Introduction to Linguistic Science

Mark Hale & Madelyn Kissock

Concordia University
2017-18
 Contents

Contents i

1 Introduction 1
1.1 What is ‘linguistics’? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 ‘Language’ vs. languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Where is ‘language’? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Mental vs. Physical Reality . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Exploring the Mind through the Physical World . . . . . . . . . . . . . . . 10
1.6 Some further implications of Figure 1.1 . . . . . . . . . . . . . . . . . . . 12
1.7 What is a linguistic system? . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 What linguistics is not about . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Phonetics 19
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Speech as a Continuum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Articulatory Phonetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Summary of relevant IPA symbols . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3 Phonology 41
3.1 Further Phonetic Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 The Phoneme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Minimal Pairs & Allophones . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 More English allophony . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Rule ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.6 Some Allophony in other languages . . . . . . . . . . . . . . . . . . . . . 51
3.7 Universal Properties and Language-Specific Properties . . . . . . . . . . . 52

i
ii Contents

3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.9 Solving phonology problems . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

4 Morphology 61
4.1 Morphemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Morphemes and Allomorphs . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Productivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4 Hierarchical Structure in Morphology . . . . . . . . . . . . . . . . . . . . 75
4.5 The ‘cranberry’ problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5 Syntax 85
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2 Grammaticality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3 The simple sentence in English . . . . . . . . . . . . . . . . . . . . . . . . 88
5.4 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.5 Constituency and Structure Building . . . . . . . . . . . . . . . . . . . . . 92
5.6 Syntactic Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.7 More Head to Head Movement . . . . . . . . . . . . . . . . . . . . . . . . 104
5.8 Structures and Their Interpretations . . . . . . . . . . . . . . . . . . . . . 107
5.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

6 Semantics 115
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.2 Lexical Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.3 Sentential Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

7 Sociolinguistics 127
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.2 Traditional Dialectology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.3 Problems with Traditional Dialectology . . . . . . . . . . . . . . . . . . . 131
7.4 The Sociolinguistic Revolution . . . . . . . . . . . . . . . . . . . . . . . . 133
7.5 Standard and Non-Standard . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.6 Gender and Linguistic Variation . . . . . . . . . . . . . . . . . . . . . . . 137
7.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

8 Historical Linguistics 143

Contents iii

8.1 Language Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

8.2 Comparative Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . 149
8.3 Relative Chronology and Subgrouping . . . . . . . . . . . . . . . . . . . . 162
8.4 Cultural Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

9 Language Acquisition 173

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.2 The Earliest Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.3 Later Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.4 Acquiring a Lexicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.5 Theories of Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.6 The Critical Period Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . 182
9.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
 Chapter 1
Introduction

Contents
1.1 What is ‘linguistics’? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 ‘Language’ vs. languages . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Where is ‘language’? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Mental vs. Physical Reality . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Exploring the Mind through the Physical World . . . . . . . . . . . . . 10
1.6 Some further implications of Figure 1.1 . . . . . . . . . . . . . . . . . . 12
1.7 What is a linguistic system? . . . . . . . . . . . . . . . . . . . . . . . . 13
1.8 What linguistics is not about . . . . . . . . . . . . . . . . . . . . . . . . 16
1.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.1 What is ‘linguistics’?

Linguistics is the study of one aspect of human knowledge, specifically, knowledge of
language. Because linguistic science concerns human knowledge, it falls under the scope
of the broader field of Cognitive Science. Other disciplines which fall under Cognitive
Science include those that study auditory perception, visual perception and motor co-
ordination. As is usual in scientific enterprises, ‘language’ has a very specific sense in
linguistics – one which does not match that found in everyday usage. Since we will be
using ‘language’ in its technical, linguistic sense, it is important at the outset to be clear
about what precisely the term means. In this context – an introduction to a scientific
pursuit – this precision has the added benefit of giving you some idea what the subject
matter of the discipline is. This chapter is intended to be helpful in this regard.
Physics is the study of the physical universe. Nevertheless, we do not normally see
physicists wandering around measuring dogs, beards, or mailboxes – although all these

1
2 Chapter 1. Introduction

things are part of the physical universe. Physicists attempt to develop general rules for
the structure and processes of the physical universe from which the properties of specific
entities – a dog, a beard, or a mailbox – follow. Similarly, the object of study in linguistics
is not your language or Fred’s language or any specific instantiation of ‘language’, but
rather the general principles which govern all languages. Specific individual languages,
like specific entities in the physical universe for a physicist, are intended to follow from
this general model of language.
Contrary to widespread everyday usage, then, a linguist is not someone who knows a
lot of languages.1 It is true, however, that a primary source of data for the general theory
of the nature of linguistic knowledge must necessarily be actually observed linguistic sys-
tems. Therefore, linguists generally know a certain amount about the structures and pro-
cesses found in a variety of languages; however, the knowledge that linguists have about
languages (other than their native language) is what we might call explicit or conscious
knowledge. By contrast, the knowledge that a native speaker has is tacit or unconscious
knowledge. (If it were not, the job of the linguist would be quite trivial.) The linguist’s ex-
plicit knowledge about the structures and processes found in the linguistic systems they
study does not give them the ability to speak these languages. If you have ever studied a
second language in some relatively formal setting, you know how inadequate conscious
knowledge of the grammar of a language is as a basis for speaking. It is only when your
knowledge becomes unconscious that you actually feel you can speak a language.

1.2 ‘Language’ vs. languages

One of the most frequent uses of the term ‘language’ in everyday speech is in such collo-
cations as ‘the English language,’ ‘the French language,’ ‘the Swahili language,’ etc. These
notions of ‘language,’ which we will call ‘socio-political,’ turn out to have no clear referent
and thus provide no meaningful basis for scientific inquiry.
We will give two examples of the difficulties which confront one attempting to de-
velop a coherent notion of ‘the English language.’ The first is literary (in spite of the fact
that written language does not fall under the purview of linguistic science in the view
of most linguists). The Modern Library Board recently released a list of the top 100 of
“what it considers to be the best novels published in the English language since 1900.”
James Joyce’s Finnegans Wake is listed as number 77, meaning, among other things, that
this board of specialists believe it is written in ‘the English language’ (and that they liked
it). A typical passage from Finnegans Wake is this nice little scene, early in the novel (pp.
15-16):

In the name of Anem this carl on the kopje in pelted thongs a parth a lone
1
Saying ‘John is a great linguist, he knows 10 languages’ sounds to a linguist like saying ‘Mary is a great
physicist, she hits home runs all the time.’ (Understanding the physics of velocity and trajectory does not,
unfortunately, have any bearing on whether one can hit a baseball accurately and far.)
1.2. ‘Language’ vs. languages 3

who the joebiggar be he? Forshapen his pigmaid hoagshead, shroonk his
plodsfoot. He hath locktoes, this shortshins, and, Obeold that’s pectoral, his
mammamuscles most mousterious. It is slaking nuncheon out of some thing’s
brain pan. Me seemeth a dragon man. He is almonthst on the kiep fief by here,
is Comestipple Sacksoun, be it junipery or febrewery, marracks or alebrill or
the ramping riots of poutiose and froriose. What a quhare soort of a mahan.
It is evident the michindaddy. Lets we overstep his fire defences and these
kraals of slitsucked marrogbones. (Cave!) He can prapsposterus the pillory
way to Hirculos pillar. Come on, fool porterfull, hosiered women blown monk
sewer? Scuse us, chorley guy! You tollerday donsk? N. You tolkatiff scowe-
gian? Nn. You spigotty anglease? Nnn. You phonio saxo? Nnnn. Clear all so.
’Tis a Jute. Let us swop hats and excheck a few strong verbs weak oach eather
yapyazzard abast the blooty creeks.

If a linguist attempted an analysis of our instantiation of knowledge of language, which

most people would call ‘English,’ and had also to account for the above data as ‘the En-
glish language,’ he or she would certainly fail to develop a coherent account of either our
knowledge or the knowledge of Joyce (as represented by Finnegans Wake).
To take a more mundane example, here is a little story from Yorkshire (taken from
Arnold Kellett, Basic Broad Yorkshire, Smith Settle, 1992). We have added a translation.

‘Thoo’s getten poison i’ thi’ sistren, that’s why thoo’s bellywark,’ says t’
docther. ‘Thoo mun a thi teeath oot.’
‘What?’ Ah says, ‘All on ’em?’
‘Aye,’ he says. ‘ivvery yan’… So Ah took ’em oot, an laad em on t’ table.
‘You’re getting poison in your well, that’s why you have a belly-ache,’ says
the doctor. ‘You have to have your teeth extracted.’
‘What?’ I say, ‘All of them?’
‘Yes,’ he says, ‘every one.’… So I took them out and laid them on the table.

While the attempt to represent this way of speaking using a modified version of the tra-
ditional English writing system leaves much unclear (we will see how linguists deal with
this problem soon), what is clear is that our language and this one are quite different.
Again, any attempt to develop a scientific analysis of our speech which must also deal
with this evidence will fail to provide insightful analysis of either linguistic system.2
What these examples show is that the notion ‘the English Language’ does not pro-
vide the linguist with a sufficiently coherent body of data to subject to meaningful sci-
entific analysis. To understand the differences between our language and that of Joyce’s
2
By the way, once we recognize that the Yorkshire linguistic system and ours are different in important
ways, if we want to keep calling one of them ‘English,’ it is probably we who are in trouble: Yorkshire is in
England, after all…
4 Chapter 1. Introduction

Finnegans Wake, that of Yorkshire, that of an Alabama farmer, or that of a life-long resi-
dent of Melbourne, Australia we cannot lump these quite diverse linguistic systems into
one big ‘English language’ bucket. Indeed, it turns out that upon sufficiently close ex-
amination, the language which each of us uses is different from the language of virtually
everyone else. Even the notion ‘the language which I use’ turns out to be rather under-
differentiated, for scientific purposes.
A cautionary note must be added at this point. Linguists continue to use such terms
and phrases as ‘English’, ‘in Tzotzil’, and ‘some languages have…’ While this is both mis-
leading and inaccurate, it is convenient, and it is done with the understanding that these
entities do not actually exist in the relevant linguistic sense.

1.3 Where is ‘language’?

The foregoing discussion reveals a serious shortcoming in traditional notions of what
‘language’ is. ‘Language’ is generally treated as something ‘out there,’ in the world. The
meaning of words is thought to be found by consulting a dictionary, the ‘correct’ rules
of grammar by looking in a grammar book. There is some notion that ‘correct English’
exists and that we are all attempting to attain this difficult target, some of us with more
success than others.
Linguists, by contrast, recognize the important role that all linguistic systems play
in the scientific pursuit of attempting to understand in general what a possible linguistic
system is. Through an examination of the diversity of human linguistic systems, linguists
have learned that there are no ‘incorrect’ linguistic systems, no ‘primitive’ languages,
and, most importantly, that language, in the relevant scientific sense, is not something
‘out there’ but rather something ‘inside’ a human being – a piece of mental machinery.
Each individual has a particular knowledge state with respect to language. As a result,
there are at least as many ‘languages’ as there are humans.3 A reasonable question at this
point might be whether this reduces the possibility of scientific inquiry into knowledge
of language to zero. If everyone’s knowledge state were completely different, it would
mean that no interesting generalizations about language (and consequently about human
knowledge with respect to it) could be made. What linguists have observed, though, is
that no matter how superficially different the languages are (in the sounds or words they
use, for example), there are striking similarities among them whether those languages
are types of Chinese, Telugu, Zulu or Pohnpeian.
Our goal, then, is to characterize what humans actually know with respect to language,
not what society/books/educational systems state that they are supposed to know.4 An
important step in creating a model of human knowledge of language is to develop de-
scriptive grammars. These contrast sharply with the prescriptive grammars that most
3
The inclusion of ‘at least’ is meant to account for individuals who are multilingual and/or multidialec-
tal.
4
We are not making a judgement about the validity of the latter. We are simply saying that what people
are supposed to know, for whatever reason, is irrelevant to linguistic inquiry.
1.3. Where is ‘language’? 5

people are familiar with from school. An example might make this clearer. If you were
to observe our speech for some period of time, you would find that we say things like the
following:

• I have to teach this morning and so don’t you.

• I didn’t do nothing all day yesterday.

• Since you never come to class, that’s all the higher a grade I can give you.

Instead of (in keeping with ‘grammar’ books):

• I have to teach this morning and so do you.

• I didn’t do anything all day yesterday.

• Since you never come to class, that’s as high a grade as I can give you.

There are several possible theories one could develop as to why we say one set of
sentences, rather than the other. These include (among others):

1. We’re of well below average intelligence.

2. We’re lazy.

3. We’re of well below average intelligence and we’re lazy.

4. We’re undereducated.

5. We’re undereducated and of well below average intelligence.

6. We’re undereducated and we’re lazy.

7. All of the above.

8. We have different linguistic systems than that of the people who wrote the gram-
mar books.

We are not really in a sufficiently impartial position to evaluate our degree of intelligence,
sloth, and education (although we both hold Ph.D.’s), but we think there are pretty com-
pelling reasons to believe that (8) provides the most productive hypothesis. Consider, for
example, the following. In English grammar books, multiple negation (‘I didn’t do noth-
ing all day yesterday.’) is considered non-standard (i.e., ‘incorrect’). By contrast, in many
linguistic systems, including those generally referred to as ‘Slavic,’ multiple negation is
6 Chapter 1. Introduction

standard. Failure to use both negatives in such sentences would be taken as a clear sign
of lack of intelligence, laziness, or lack of education. But how can it be laziness for one
of us to use multiple negation and for speaker of a ‘Slavic’-type language to fail to do so?
Either one or the other must be the lazier option, and lazy people everywhere should opt
for the same form. Instead, what counts as lazy, stupid, or lacking in intelligence is al-
ways related to some societal norm, the workings of which are quite independent of any
linguistic system.
If our linguistic systems do not reside in or derive from grammar books, where do we
have them? How did we get them? What are they? These are basic questions posed by
linguistic science. The generally accepted answer to the first of these questions is not
surprising: an individual’s linguistic system is located in their mind. It is that component
of the cognitive system which enables a person to parse (analyze) incoming speech and
to produce linguistic output.5 The property of having human-type linguistic systems ap-
pears to be restricted to human-type minds. Although efforts have been made to teach
various symbolic manipulation systems to higher apes, the fact remains that if we take
any normal human infant and expose them to human speech, they will acquire a human-
type linguistic system. This is not true, as far as can be determined, of apes, dogs, rats,
mosquitos, or potatoes. It is uniquely true of humans, to our present knowledge, and sug-
gests that the linguistic system is species-specific. Crucially, this claim has nothing to
do with whether other species have systems they use for intra-species communication
– presumably all species have methods of communicating. The claim is limited to the
linguistic system that we will be exploring, which appears to be a system found only in
humans.
The answer to the question of how we each ended up with the linguistic systems that
we have is contained in many ways in the preceding paragraph: when we were infants,
people spoke to us and around us. As in humans generally, this led to the development of
a linguistic system in each of our minds/brains (commonly known as our ‘native languages’).
The process of how this happens will be the subject of considerable further discussion.

1.4 Mental vs. Physical Reality

The conclusion of the previous discussion – that the linguistic system is a type of knowl-
edge found in every human mind/brain – is one of the primary reasons why Linguistics
as a field falls under the larger domain of Cognitive Science. It is probably not immedi-
ately clear, however, what Linguistics has in common with visual perception, for example.
Since the connections between the two, as well as the general theoretical framework of
Cognitive Science, are both very important to understanding the linguistic system, we
5
Note that we say specifically ‘component of the cognitive system’. We will discuss the important dis-
tinction between cognitive systems and physical interface systems, such as the visual and auditory systems,
shortly.
1.4. Mental vs. Physical Reality 7

will take a quick tour into the realm of physical vs. mental reality here. To start, have a
look at the illustration provided below.

This will certainly not be the first time you have seen what people generally call an
‘optical illusion.’ These sorts of pictures get passed along from person to person peri-
odically. Most people stare at the picture for a few seconds, say something like ‘Cool!
How/Why does it do that?’ and move on to some other item or activity of interest. We’re
going to make you think about it for a little bit longer than that, though, because the
question ‘How/Why does it do that?’ is actually a fairly important one.
First, let’s see how you would describe what you are seeing. Most people agree that
the picture has the following properties:

(a) There are black squares with rather thick greyish lines between them. (The picture
is a rectangle, assuming each square is of equal size, with 8 squares along the bottom
and 6 along the side.)

(b) At the intersection of the lines, there are white dots

(c) Black/grey dots appear fleetingly on the white dots as you run your eyes over the
picture

(d) Black dots never appear on a white dot if you have focussed your eyes specifically
on that dot.

We will not go further into (a) and (b) although they have their own interesting features.
What strikes people particularly about the picture (and causes them to label it an ‘optical
illusion’) are (c) and (d). Often accompanying this picture, by the way, is the text: ”Count
8 Chapter 1. Introduction

the black dots!” This, of course, turns out to be impossible to do. That’s only one of the
‘problems’ you have with this picture, though, as you can see from a more complete list
of problems below.

• You see black dots but cannot count them

• You only see black dots when your eyes are scanning the picture

• When scanning the picture, the black dots appear to move from one white dot to
another

• The black dots appear only very fleetingly on the white dots and move very fast
between them

• You know that this is an inert piece of paper and therefore it does not have any
moving parts

The final problem is the most worrisome. But the good news is that it’s not just you
having some sort of visual/mental glitch. Everyone who looks at this picture reacts the
same way. This makes it a very interesting problem for science. (If it was just you, it
would be, frankly, a lot less interesting...) There is a measurable physical object (the pic-
ture) to which humans consistently attribute the same ‘impossible’ (unattested) physical
properties (i.e., black dots and movement).
Where should we search for an explanation of these strange phenomena? It’s easy
to rule out the piece of paper. Both casual and more sophisticated measurements tell us
what the actual physical properties of the paper are and that they don’t include moving
dots. Indeed, if we want to get scientific about it, we can measure with a specialized piece
of equipment the brightness or luminosity of the image at each dot. Such a device will
uncover no black dots and no motion at all. We can also rule out the ‘air’ between you and
the paper. If it was something in the air, then there is no explanation for why you don’t
see moving black dots on this page. Let’s also rule out alien intervention – they’d have to
be some pretty tricky aliens to get all of us to think the same thing at exactly the same
time, every time. There are, of course, other logical possibilities but they are quite remote
and so should not be considered first. Essentially, we are left with you, specifically, your
eyes and your mind.
Scientists understand the mechanics of the eye quite well at this point. It is an organ
that reacts only to certain types of stimuli (light that falls within a well-defined frequency
range, but not x-rays or ultrviolet light, for example). When (processible) stimuli are in-
troduced to it, the eye passes information along to the brain. What the eye cannot do
is manipulate the stimuli themselves. But some manipulation is taking place in our pic-
ture. The physical properties of the picture are unchanging but something is taking those
properties and creating a slightly different picture for us. It is definitely not the eye itself
which is responsible for this.
1.4. Mental vs. Physical Reality 9

Not surprisingly, we conclude that it is your mind that is causing you to perceive this
picture as having moving black dots. Somehow, your mind has taken the information it
receives from the eye and constructed its own version of the picture – and that is the pic-
ture that you ‘see,’ or, more appropriately, that you ‘perceive.’ Many of the properties of
the physical picture are present in this mental picture version but the mind has also ob-
viously added some properties that the original picture does not and cannot have. Rather
crucially, the ‘mental picture’ is produced completely consistently within and across hu-
mans. Every time you see this particular optical illusion, your mind (and everyone else’s)
creates for you the same perception. This, and the countless other examples of its type,
suggests that the mind has a system for creating mental perceptions out of the stimuli that
the eye provides to it. Exploring and modelling this system is the task of researchers in
Cognitive Science who work on vision.6
It turns out that audition and language, among other areas, share this dichotomy be-
tween the physical representations of stimuli and their mental representations (what we
‘perceive’). Moreover, it seems that, like vision, the mind is strictly systematic about how
it constructs mental representations of auditory and linguistic stimuli. These are some
of the primary reasons that cause us to unite all of these areas under the domain of Cog-
nitive Science. Beyond this, all of the sub-fields within Cognitive Science are also united
by the fundamental assumption that the mind can best be modelled as a computational
system. The computational approach to human knowledge has proved fruitful and still
provides the general framework within which each of the sub-fields is explored.
Examples parallel to the optical illusion are actually fairly easy to find for language.
Some of them are so common, in fact, that we pass right over them without noticing. One
very prominent and accessible ‘linguistic illusion’ is that physical entities such as ‘words’
exist. This illusion is especially strong because we happen to have a writing system as-
sociated with English and our particular system requires white space between some sets
of letters but not between other sets.7 Although we can formally support this claim with
acoustic measurements of language, it is also observable in an everyday situation. Con-
sider those occasions when you happen to overhear people speaking a language that you
do not speak or understand. It is impossible to determine how many ‘words’ they are
saying, all you hear is a stream of unintelligible sounds. If there were actually ‘words’ in
their speech, you should be able to count them, at least, even if you don’t know what the
words mean. The fact is that there are no physical entities that correspond to ‘words,’
instead, ‘word’ is a mental construct (like moving black dots) that your linguistic system
builds for you based upon experience with specific language data. (The term (word) is
still difficult to define and may be better replaced by different units, as we will see later.)
6
If this example does not seem sufficiently compelling to persuade you that the mind controls your
perception of physical stimuli, compare the fact that the human eye can only process two dimensions with
the three-dimensional percept your mind presents to you every time you open your eyes. For a fascinating
look into human visual cognition, see Donald D. Hoffman, Visual Intelligence, 1998.
7
Upon examination, it is clear that there is no way ‘word’ could be defined based on this writing con-
vention either. And, just for information purposes, not all writing systems require white space of this type.
10 Chapter 1. Introduction

Having briefly grounded Linguistics in Cognitive Science, we will now turn to a more
detailed discussion of how we can explore the mental system responsible for our knowl-
edge of language and how the physical and mental systems used for language are related
to one another.

1.5 Exploring the Mind through the Physical World

Given that what we linguists are interested in is a property of the human mind, we have
a problem. The mind itself is not directly observable.8 Linguists have a variety of tools
which make it possible to transcend the difficulties created by this problem (which con-
fronts all scientific inquiry, since frequently the phenomena of interest are not directly
observable). The figure below may help clarify the distinction between the mind and the
physical world, and illustrate the systems that provide an interface between the two.

mental
acoustic representation
processor of acoustic
signal
acoustic ‘inter-
computational system
signal pretation’
output
articulatory of grammar
system (mental
representation)
the mind
the physical/external world

Figure 1.1: The mind vs. the physical world.

In Figure 1.1 (see following page), the linguistic computational system, also known
as the grammar is shown schematically within a double-lined circle. The computational
system/grammar has both input and output functions, on the one hand interacting with
systems that are auditory and articulatory and on the other with a component that is
8
There are new and exciting techniques for studying electro-chemical activity in the brain, such as PET
scans and MRIs, however, we know virtually nothing about the relationship between the cognitive systems
posited by scientists for the mind and their physical instantiation in brains. While recent developments
may have moved this question from the domain of mystical speculation into that of potential empirical
investigation, we need to have well-grounded theories in three domains: (1) the cognitive systems them-
selves, (2) the physical structures and processes of the brain, and (3) the relationship between (1) and (2).
Linguistics is deeply involved in (1). At present (2) is rather poorly understood. (3) remains mysterious.
1.5. Exploring the Mind through the Physical World 11

responsible for the actual meaning content of the ‘message’ which we have just labelled
‘interpretation.’ The arrows leading into the grammar indicate the processing of input
data, the arrows leading away from the grammar indicate the process of generation of
spoken output. The dotted box is intended to reveal which aspects of these processes are
mental as opposed to physical.
Examining the input path first, we can see that we start with an acoustic signal.9 If
this signal is within earshot of our schematic listener, it will enter his or her ear, be trans-
formed through fairly complex mechanisms into neural impulses, and thus ‘be heard’ in
some meaningful sense. Roughly, hearing a sound involves converting a physical set of
‘sound waves’ (variations in air pressure) into a mental representation. If the signal is hu-
man speech, this mental representation will be passed on to the grammar for linguistic
processing. Assuming the signal is parsable by the grammar, the grammar will generate
an ‘interpretation’ for the signal. Naturally, this interpretation is a mental representa-
tion, as well.
The line between physical and cognitive processes is found at that point at which the
physical effects of receipt of a sound wave by the part of the physiology responsible for
auditory processing are converted into a form which the rest of the cognitive system can
operate on. The actual sound waves do not ricochet off the inside of the skull — they
must be cognitively represented in order to be processed by higher cognitive systems.
This is indicated in the figure by the fact that the mind/physical world line cuts through
the ‘acoustic processor’ module.
If we now consider the generation of speech in this figure, we see that we start with
some kind of semantic representation (an ‘interpretation’) which the speaker wishes to
produce as speech. This semantic representation serves as the input to the grammar. The
grammar acts as a processor which converts the semantic/input representation into an
output representation. Note that the output representation is not part of the grammar
— it is the product of the grammar’s internal processing. This output representation is
then passed on to the articulatory system of the speaker, which converts it into a set of
muscle commands. The result of this processing is a set of articulatory acts performed in
response to the muscle commands. These acts will generate speech.
The line between physical and cognitive processes in speech generation is found at
that point at which the mental representations generated by the grammar are converted
into a set of physical signals sent to the articulatory organs. In Figure 1, this is indicated
by the fact that the dotted line which separates the mind from the physical world cuts
through the ‘articulatory system’ module.
Having gone through these issues in some detail (though the processes involved are
still grossly oversimplified), it should be obvious that the most readily accessible element
of this system, to those of us who are not telepathic, is the portion outside of the dotted
9
Natural (human) language may use an acoustic or a visual signal as the medium. The structure of the
argument is not affected by the medium. The technical details in the chapter are based on an acoustic signal
rather than a visual/signed one, however. The basic principles upon which the linguistic system is built
are the same for both acoustic/spoken and signed natural language.
12 Chapter 1. Introduction

box in the figure above. There are basically three aspects of this material that are more
obviously physically grounded: (1) the acoustic signal, (2) the physical structures and
processes of the auditory system, and (3) the physical structures and processes of the ar-
ticulatory system. It is the field of phonetics which investigates these matters, providing
linguists with tools which allow them to get at the most accessible aspects of the speech
system in order to test hypotheses about the underlying cognitive systems which play a
key role in this system.

1.6 Some further implications of Figure 1.1

While talking about the line between the primary interest of linguists, the mental system
called the grammar/computational system, and the systems which turn the output of the
grammar into something we can observe the effects of (the acoustic signal), there are a
few important related issues we should deal with.
It is not unusual to hear adult speakers produce strings of speech such as the following
(taken from a recorded conversation contained in a Carterette & Jones, Informal Speech,
University of California Press, 1974):10

• aj ʤəs rili ɪt ɪts ðə bɪgəst θrɪl

• I just really it it’s the biggest thrill

This speaker clearly started out saying one sentence, then shifted midstream to another
(without finishing the first). She also said the subject of the second sentence (‘it’) twice.
The result is that she has said something which is not a sentence at all — it cannot be
parsed by a grammar. It is difficult to interpret exactly what was intended (what was
she saying about herself with ‘I just really’, for example?). Part of what she said (‘it’s the
biggest thrill’) is a perfectly good English sentence, and is the only part of the string of
noise she produced that can be coherently interpreted as a sentence. Notice, for example,
that although the string starts out as if it is saying something about the speaker (‘I’), you
cannot respond to this sentence by saying ‘no, you don’t’ (or ‘no, you aren’t’).
This can be made clearer if we imagine trying to teach someone ‘English.’ One seem-
ingly sensible way to do this would be to give them sentences said by speakers of English.
Speakers say things like ‘it it’s the biggest thrill’ every once in a while. But this is not due
to the grammar which is present in their mind – it is instead an effect of the imperfect
physical system through which the grammar is forced to play its output. If we taught our
learner to say ‘it it’s the biggest thrill’ (as speakers of English do every once in a while),
when the learner’s body suffered the same electro-chemical glitch that caused our speaker
10
This is presented first in the alphabet of the International Phonetic Association, which you will learn
about later in the next chapter, then in regular English orthography. After you learn about the IPA, you
should take another look at this passage and try to read it out loud.
1.6. Some further implications of Figure 1.1 13

to say ‘it it’s the biggest thrill,’ our learner would say ‘it it it’s the biggest thrill’ (mistak-
enly saying sentence-initial ‘it’ twice). To get our learner to say ‘it it’s the biggest thrill’
with the same frequency as our speaker does, we have to teach them to say ‘it’s the biggest
thrill.’ If we do that, the learner’s body will produce — through the same processing im-
perfections that our speaker’s body has – ‘it it’s the biggest thrill’ every now and then (not
that that is necessarily a desired outcome …).
If ‘it’s the biggest thrill’ is the real output of the grammar, we would expect our speaker
to be able to tell us that. That is, even though speakers occasionally say ‘it it’s the biggest
thrill,’ they actually intend on those occasions to say ‘it’s the biggest thrill.’ And, indeed,
speakers do readily recognize that they occasionally fail to articulate precisely what they
intended.
We therefore need to carefully distinguish between the competence of the speaker,
responible for the output of their grammar (a mental representation), and their perfor-
mance on some particular occasion (responsible for the acoustic output of their body).
Bodily lapses of various sorts (lapses of memory, sneezes, etc.) and/or external physical
disturbances (high winds, dogs jumping on your chest, and so forth) modify the acoustic
signal which emerges from the body, but theses lapses/disturbances do not reflect paral-
lel lapses/disturbances in the processes of the grammar. Lewis Carroll was toying with
this distinction when he wrote, in Through the Looking Glass:

‘And you do Addition?’ the White Queen asked. ‘What’s one and one and
one and one and one and one and one and one and one and one?’
‘I don’t know,’ said Alice. ‘I lost count.’
‘She can’t do Addition,’ the Red Queen interrupted.

Of course Alice knows how to add a sequence of ones. But presented in rapid fire succes-
sion, her perceptual performance system prevents her arithmetical system from being
able to access the information necessary to answer the White Queen’s question. The per-
ceptual and articulatory processing systems thus impose limits on access to underlying
competence.
Consequently, although the output of the body provides linguists with much of the
evidence they have for the underlying grammatical system, this evidence must be used
with caution. In particular, it must be augmented by evidence which can be obtained by
checking the judgement of the speaker as to whether or not a given string they produced
involved a performance error. This evidence is known as a ‘grammaticality judgement’
— it can be used to help determine whether a string accurately reflects the output of
the grammar or not. In general, for your own grammar, you can perform this empirical
research through introspection.
14 Chapter 1. Introduction

1.7 What is a linguistic system?

This question will be the primary concern of the first two-thirds of this book. A brief
overview of the central components of the system is given below.
The linguistic system is stored in the human mind – it is some type of knowledge. As
pointed out earlier, each person’s linguistic system is somewhat different from another’s
but, at the same time, there are striking similarities in human linguistic systems. In fact,
the set of possible human linguistic systems appears to be highly restricted. These re-
strictions on human linguistic systems can be accounted for if we assume that innate (i.e.,
genetically given) knowledge restricts the types of linguistic systems humans can create.
The genetically-determined portion of the system must necessarily be supplemented by
data acquired from the environment. One way to answer the question of what ‘language’
is, in the linguistic sense, is to explore what one must know to know a language.
The traditional subfields of linguistics provide broad insight into this issue. The basic
subfields are:

1. Phonetics

2. Phonology

3. Morphology

4. Syntax

5. Semantics

Phonetics, as mentioned earlier, studies the physical properties of sounds, their articu-
lation, and their audition. Phonology is concerned with the knowledge which underlies
the ability to produce and analyze output and input of the grammar from the perspective
of structures that will relate to sound. For example, if you were asked to make up a name
for a newly created beard-trimming mechanism, you might call it a

• beard-wrangler

• zorp

• flirp

but you would almost certainly not call it a

• dnkli

• ngloopi
1.7. What is a linguistic system? 15

• [lyn] (=the usual pronunciation of French ‘lune’)

Since you easily make such judgements about words you have never encountered before,
you appear to have some knowledge of what is a ‘possible word’ in your linguistic system.
The first set of examples satisfies some criteria for ‘possible word’, whereas the latter
does not. Note that no one has given you explicit instructions not to name a new product
a ‘ngloopi.’
Morphological knowledge is reflected in your ability to analyze words generated by
linguistic systems into meaningful parts. For example, you know that ‘cats’ consists of
two parts: a part that means ‘cat’ (that’s the ‘cat’ part) and a part that means ‘more than
one’ (that’s the ‘-s’).11 You know that it is not the case that it consists of, e.g., three parts
(a part that means ‘domesticated animal,’ a part that means ‘feline’ and a part that means
‘more than one’). The internal structure of words is analyzed by the morphology of the
linguistic system, which also produces internally-complex words.
Syntactic knowledge is knowledge about the structure of sentences. You know that

• you will not pass this class if you don’t pay attention

is a possible sentence given your linguistic system. On the other hand,

• *not class this pass don’t pay if will attention you you

is not.12 Your linguistic knowledge, in particular the syntax of your linguistic system,
which governs the structuring of phrases and sentences, will not produce the asterisked
sentence and will not parse it as a ‘sentence,’ therefore it comes across instead as just a
list of words.
We would like to remind you here of the distinction that we made between social
judgements about language ‘correctness’ and the aims of linguistic science (to explore
and characterize knowledge of language). The following sentences are produced by our
linguistic systems – they are syntactically well-formed and therefore not given an aster-
isk.

• I didn’t do nothing all day.

• I have to go to school tomorrow and so don’t you.

11
Lacking the tools of phonetics at this point, we are just representing this in English orthography here.
‘Orthography,’ in its narrowest sense, means ‘correct writing’ i.e., proper spelling but is sometimes used as
a synonym for ‘writing system.’
12
Linguists mark strings of words which do not parse as sentences – so-called ‘ungrammatical’ strings –
with an asterisk, as here.
16 Chapter 1. Introduction

From a linguistic perspective, these sentences are syntactically grammatical, i.e., they
have been produced by the grammars of each of us and are judged by each of us to be
perfectly fine sentences. The fact that they may not be socially acceptable (as sometimes
indicated by prescriptive grammars) is completely irrelevant to our inquiry.13
Semantics is the study of meaning. As such, its concerns are much broader than those
of linguistics alone. However, some aspects of meaning may be linguistic in nature (the
matter is quite difficult). It is certainly the case that the grammar plays a key role in
one’s ability to assign meaning to sentences that one hears and to produce sentences that
express, at least in some vague way, the meaning one intends in some particular situation.
These ‘modules’ rather traditionally are referred to together as the grammar. Once
again, they do not mean the kind of prescriptive grammar found in traditional grammar
books, but rather a component of the cognitive system of a human – a mental organ, as
it is sometimes called, which has the responsibility for producing and parsing linguistic
information. We will study each of these modules of the grammar in turn.
There are three additional topics which we will cover in this book. The first is Soci-
olinguistics which looks at the relationship between social variables like age and gender
and linguistic variation. The second is Historical Linguistics which concerns itself with the
genetic relatedness of languages and language change over time. The third is Acquisi-
tion which, just as it sounds, is an inquiry into how knowledge of language develops in
humans. As we have briefly noted, there is considerable evidence to suggest that that
linguistic knowledge has both an innate/genetic component (commonly referred to as
Universal Grammar or UG) and an experience-based component. One of the areas that Ac-
quisition explores is which aspects of knowledge are attributable to which of these com-
ponents.

1.8 What linguistics is not about

Before proceeding, we would like to point out certain topics that we are not going to talk
about and why we are not going to talk about them. The first of these is communication.
It is true that the human linguistic system is often (but not always) used for communica-
tive purposes – so, for that matter, are Morse Code, semaphores, and picture-drawing,
among other things. Note that a particular message can be communicated using a va-
riety of different systems, such as those mentioned above. Given this fact, the message
is independent of the system (although it is true that some systems convey certain mes-
sages better than others). Linguistic inquiry of the type we have described here is an
exploration of the properties of the linguistic system, itself, not the message or messages
that are conveyed by it.
13
Although we do still casually speak of ‘English’, the grammars of particular individuals may or may
not produce the sentences here. What is grammatical for one person may be ungrammatical for another.
The point remains that these judgements have nothing to do with societal/prescriptive norms.
1.8. What linguistics is not about 17

The second topic that we will not talk about is writing systems. Human knowledge
of language existed long before writing systems developed and continues, for the vast
majority of languages, to exist independently of such systems. All humans have a fully-
developed linguistic system before they learn a writing system (if, indeed, they ever learn
one). Therefore no appeal to a writing system (nor any mention of ‘letter/spelling’) has
a place in our inquiries.
18 Chapter 1. Introduction

1.9 Exercises
Circle the correct answer.

1. A speaker’s knowledge of his/her language is:

(a) explicit knowledge.

(b) conscious knowledge.
(c) tacit knowledge.
(d) all of the above.

2. ‘I’m not doing nothing today’ is:

(a) a string of words which is ungrammatical for all humans.

(b) a string of words which is a grammatical sentence for some humans.
(c) an illogical statement.
(d) a fundamental truth.

3. ‘Syntax’ is the module of the grammar responsible for:

(a) the structure of sentences.

(b) the structure of complex words.
(c) the sounds used in human languages.
(d) that extra money you have to pay when you buy cigarettes.

4. Linguists examine various linguistic systems and develop

(a) prescriptive grammars.

(b) descriptive grammars.
(c) dictionaries.
(d) theories about which systems are best.

5. A ‘linguist’ is someone who:

(a) can speak more than five languages.

(b) can speak more than ten languages.
(c) investigates the nature of human language.
(d) speaks with proper grammar.
1.9. Exercises 19

6. Select the words below which belong in this sentence: ‘The English language is
__________ linguistic research.’

(a) a critical component of

(b) the major focus of
(c) irrelevant to
(d) one aspect of

7. ‘Phonology’ is:

(a) the study of telephonic communication.

(b) the study of telepathic communication.
(c) the module of the grammar which allows one to produce and parse speech
sounds.
(d) You phonio saxo? Nnnn.

8. An individual’s linguistic competence refers to:

(a) how clearly they enunciate.

(b) how many languages they know.
(c) the mental system containing linguistic knowledge.
(d) their ability to speak in public.
 Chapter 2
Phonetics

Contents
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Speech as a Continuum . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Articulatory Phonetics . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3.1 Consonants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Vowels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4 Summary of relevant IPA symbols . . . . . . . . . . . . . . . . . . . . . 34
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2.1 Introduction
The focus of this chapter is on one aspect of the field of phonetics. There are three major
branches of phonetics.

(a) acoustic phonetics

(b) articulatory phonetics

(c) auditory (or perceptual) phonetics

In this chapter, we will be focusing on the study of articulatory phonetics. The reasons
for this are simple: the study of acoustic and auditory phonetics generally requires access
to sophisticated equipment, laboratories, anechoic chambers, and so forth. By contrast,
the physical equipment necessary for the study of articulatory phonetics is with us all the

21
22 Chapter 2. Phonetics

time. Before we move on to articulatory matters, however, we will give a very brief illus-
tration of two types of acoustic measurements which are frequently used in phonetics.
Both will serve to underline a point made in the previous chapter.

2.2 Speech as a Continuum

Physical measurements of speech show a number of features that humans are completely
unaware of in their linguistic output. Conversely, the measurements do not show certain
features that humans believe are present in their linguistic output. We discussed a case
of the latter, the impression that humans have that there are ‘words’ when no physical
correlate of ‘word’ is present, drawing on support from casual observation of ‘foreign’
languages. Now we are in a position to illustrate the same point briefly with two types of
acoustic measurements frequently used in acoustic phonetics.
The image below shows a waveform in its top half and a spectrogram in its bottom half. A
waveform graphically represents the fluctuations in air pressure that occur when (most)
speech sounds are produced. The horizontal axis, labelled ‘zero’ may be considered the
‘at rest’ position and vertical lines above and below this position indicate fluctuations. It’s
roughly the same idea as the effect of plucking a guitar string – the string moves back and
forth to each side of its ‘rest’ position. The further away from zero/rest the line extends
(or the string moves), the louder the sound is. More critically for us, any movement away
from the zero position indicates that there is air pressure fluctuation – in our case that
translates into the presence of speechs sounds.
Spectrograms give us a rather more detailed analysis of the physical properties of
sounds. They tell us about the type of pressure fluctuation, not simply the presence or
absence of pressure fluctuation.1 A spectrographic analysis allows us to see the distin-
guishing characteristics of one vowel from another, for example, as well as one conso-
nant from another. In our image here, you can see from the spectrogram that vowels
and diphthongs have a kind of ‘barring’ pattern (two or three dark, horizontal bars) that
the consonants do not have. (Nasals also have a barred pattern, though one that is less
well-defined.)
1
Spectrograms also indicate presence or absence but are not used primarily for that.
2.3. Articulatory Phonetics 23

“These are striking images.”

We have annotated the waveform/spectrogram to include information about which sounds

go with which portions of the images. As you can see from both the waveform and the
spectrogram, there are only two actual breaks in the speech stream in spite of the fact
there are four ‘words.’2 Moreover, the annotation shows that those breaks do not occur
at any ‘word’ boundaries.3 Instead, the breaks correspond to the sounds that we represent
by ‘t’ and ‘k’ (‘g’ has some very faint activity which may be difficult to see here). So what
the physical evidence tells us is completely at odds with our mental ‘picture’ of what we
hear (and what we think we say). Even more bizarre is the fact that there is apparently no
physical evidence for ‘t’ and ‘k’ so: a) why do we think there are any sounds at all there?;
and b) how can we possibly think there are two different sounds (from no sound at all)?
Acoustic phonetics provides a wealth of information that allows us to see the physical
basis upon which our mental representations of language are built. Measurements like
these allow us at least to see where our mind’s contribution to the linguistic picture begins
and ends, although we are still rather far from understanding the details of the interface
between the organ that gets the physical (acoustic) stimulus and the cognitive (linguistic)
module responsible for providing us with a linguistic percept.

2.3 Articulatory Phonetics

In order to produce sound, air needs to be set in motion. This is a fact about the world.
Humans can make air move in a variety of ways, but in the case of speech only those
methods involving the upper respiratory tract are used. The most widely used technique
2
It’s a little harder to see in the spectrogram, but essentially, areas that are very very light indicate an
absence of signal.
3
The IPA transcription that we have used will be discussed shortly. For our purposes here, the point is
clear even for readers not familiar with the IPA.
24 Chapter 2. Phonetics

for moving air to create speech (called an ‘airstream mechanism’), used in all human lan-
guages, involves exhalation (i.e., breathing out). It is known as the pulmonic egressive
airstream mechanism. You can confirm the key role of the lungs in speech through the
following simple experiment. Make a long (5-10 seconds) [s].4 Try to pay close attention
to what you are actually doing. Do this several times to get a feel for the component parts
of the complex articulatory act. What’s going on in your mouth? Now do it again but, in
the middle of this long [s], stop the airflow coming up from your lungs without changing
the position of the articulators in your mouth. All noise ceases. Without the flow of air
from the lungs, no sound can be produced. Now make a long [s] again this time suddenly
opening your mouth as wide as possible part way through it. The [s] sound will cease.
This reveals the second critical aspect of producing speech: the modification of the flow
of air by manipulating the path through which the air is flowing — in this case in the oral
cavity (i.e., the mouth). (Airflow through this path with no modification does not make
a very interesting noise.) Figure 2.1 shows, schematically, how these two components
combine to produce acoustic output.

mouth mouth mouth

+ =

lungs lungs lungs

pulmonic
egressive oral
[s]
airstream constriction
mechanism

Figure 2.1: The components of articulation.

In Figure 2.1, one can see that the airstream mechanism alone (the leftmost schematic
upper respiratory tract) will not produce an [s], nor will the oral constriction (in this
case at the alveolar ridge, where [s]’s are made) do so alone. Only the combination of
the two acts leads to the articulation of an [s]. In English-type grammars, all speech
sounds are produced using a pulmonic egressive airstream mechanism. All languages
have sounds produced using this airstream mechanism. In addition, some languages have
sounds produced using different airstream mechanisms (‘ingressive pulmonic’, ‘glottalic
ingressive/egressive’ and so on).
4
It is traditional in linguistics to place phonetic representations of sounds in square brackets such as
these.
2.3. Articulatory Phonetics 25

The degree (none, partial, complete) and type of constriction of the airflow is the
other critical factor in producing speech sounds. The traditional division of sounds into
‘consonants’ and ‘vowels’ is also used in phonetics. Consonants are produced with some
type of airflow constriction or (partial or full) obstruction whereas vowels are produced
with a relatively unobstructed airflow. These differences between consonants and vow-
els prompt us to use different parameters to describe their articulations. As consonant
articulations are a bit more straightforward to illustrate, we will begin with consonants.

2.3.1 Consonants
Produce the following: [apapapapapa].5 Ignoring for the time being the vowels (to which
we will return later), what are you doing to the flow of air when you pronounce the [p]’s?
Produce the following: [ppppp] (that’s a really long [p], no vowel). What’s happening to
the air in your lungs? Nothing. It is totally blocked. Sounds produced with a complete
closure at some point in the vocal tract are called stops. Now produce the following:
[atatatatata] and [ttttt]. Again, during the pronunciation of the consonants, the flow of
air is totally blocked — stopped, as it were. [t] is therefore a stop as well. The various
ways of constricting the airflow (e.g., stopping it completely) are referred to as manners
of articulation.
If both [p] and [t] involve the total blockage of airflow during their articulation, how
can we tell the difference between [apa] and [ata]? Why don’t they sound the same? They
sound different because the stop closure is being made at a different point in the vocal tract
— when the air is released from the stop closure, the resulting disturbance (sound wave)
is different for [p] than for [t].6 Figure 2.2 is a schematic of the human head, with the
major places of articulation in the vocal tract labeled.
If the primary point of constriction for a consonant is at the lips (as for [p]), it is called a
labial. If the primary point of constriction is against the back of the teeth, the consonant
is called dental.7 Directly behind your teeth, on the roof of your mouth, you’ll find a
hard little ridge, the ‘alveolar ridge’, which is the beginning of the so-called ‘hard palate’.
It is at this point that most speakers of English-type grammars in North American (and
Southern England) produce, with the tip (or apex) of their tongues, their [t]-type sounds.
They are called alveolars. Segments produced behind the alveolar ridge, on the hard
palate, are known as palatals. At the end of the hard palate, the tissue becomes soft. The
soft area is known as the velum. Segments produced in this area — generally by bringing
up the back part of the tongue (the dorsum) — are called velars. Finally, hanging down at
the end of the velum is the uvula. This is the dangling piece of soft tissue that is featured
prominently when they show babies crying in cartoons. Sounds produced at this point
5
The [a] indicates the vowel of the word ‘father.’
6
This is a simplified picture of the actual acoustic description.
7
Most Western European-type grammars (but not English-type ones) pronounce sounds like [t] in the
dental point of articulation.
26 Chapter 2. Phonetics

palatal

alveolar

dental velar
labial uvular

glottal

Figure 2.2: Places of articulation.

in the oral cavity are called uvulars. The glottal point of articulation will be explained in
some detail below.
Consonants are normally described using the names of their place and manner of ar-
ticulation. [p] is thus a labial stop, while [t] is, in most English-type grammars of North
America, an alveolar stop. If you say [akakakakaka] and [kkkkk] you will realize that [k] is
produced further back in the vocal tract than either [p] or [t]. [k] is a velar stop. (English-
type grammars do not have palatal stops, which are rare, cross-linguistically, although
they do exist.)
The International Phonetic Association developed a set of symbols to represent the
sounds of natural language. This was motivated by the need to describe and represent
sounds consistently, independent of the writing system (or lack of writing system) of any
particular language. The International Phonetic Alphabet (IPA) is supplemented by a set
of diacritic marks (mainly superscripted and subscripted notations) to increase its utility.
While it is true that there is a certain amount of intra- and inter-speaker variation in
the utterance of any one particular sound (for example, every time you produce a [p],
the actual acoustic result is slightly different), the IPA captures to a sufficient extent all
the critical differences in natural language sounds. You can easily see the necessity for
such a system if you compare the following words in English orthography (recall that,
traditionally, this terms refers to the spelling system but is often now used for a writing
system) — ‘rough’ and ‘though’. The ‘gh’ in ‘rough’ represents a final consonant sound
2.3. Articulatory Phonetics 27

of [f] in ‘rough’ and represents no sound in ‘though’ (which ends in the vowel sound [o]).
While English is a rather extreme case of inconsistent symbol-sound representation, all
writing systems have this problem to some extent. The IPA allows us to overcome these
difficulties.
For the time being, we will introduce only those IPA symbols which are used to rep-
resent sounds in English. Further symbols will be added later as needed. Many of the
symbols are common to the English writing system. For example, the IPA symbol for the
velar stop is [k], the IPA symbol for the alveolar stop is [t], and the IPA symbol for the
labial stop is [p]. The task of remembering these is trivial, however, using them correctly
takes a certain amount of attention since the symbols represent sounds. Thus, the initial
sound of <cat> is represented by [k], the initial sound of <pterodactyl> is represented
by [t], and the initial sound of <pneumonia> is not represented by [p] but rather by [n].8
English has a stop in yet another place of articulation. This one really does have a spe-
cial symbol which must be learned. In addition, we do not typically write this stop at all
in standard English orthography so it may be surprising that such a thing exists. For most
North Americans, <mountain> is pronounced as [mawʔn̩]. ‘He hit me’ is pronounced as
[hi hɪʔ mi]. The sound indicated by [ʔ] in the IPA is a glottal stop.
To understand how glottal stops are produced, as well as a number of other sounds,
it is necessary to describe the glottis. Midway down your neck, noticable (especially on
males) as a bony protrusion (called an ‘Adam’s apple’), is your larynx. It is sometimes
called a ‘voice-box.’ Inside your larynx are folds of tissue known generally as the ‘vocal
chords.’ The muscles in the larynx allow manipulation of these vocal folds, controlling
the amount of tension in the folds themselves. Normally, when you’re just breathing,
these folds are apart and air flows relatively freely between them. However, you can pull
the vocal folds closed with varying degrees of strength. The space between the folds is
called the glottis. When pulled tightly closed, so tightly that the air coming up from the
lungs can’t get through the vocal folds, the glottis is closed and you are making a glottal
stop. Practice doing this by saying [aʔaʔaʔaʔaʔa]. (If the indications above didn’t make
it clear to you exactly what a glottal stop is, perhaps this will help: it is the stop in the
middle of the widely-used negator [əʔə], sometimes written ‘uh-uh.’)
This does not exhaust the stops of most varieties of English. If you observe someone
(perhaps yourself) saying [abababababa], you will notice that there is a blockage of the
airflow in that sequence as well. Where is the stop closure being made? It is labial, obvi-
ously, but we already have a labial stop, [p]. If you compare [apa] to [aba] you will notice
that, in both cases, your lips are temporarily pressed together. Why, then, do these seg-
ments sound different from one another? The answer lies in the different states of the
glottis.
In addition to being tightly closed (as for a glottal stop) or wide open (for breathing),
the glottis can adopt another posture. If you pull your vocal cords loosely closed and
then push air through them (by normal exhalation), they will vibrate. You can mimic
8
Carat brackets (< >) are used to indicate orthographic representations when it is necessary to distin-
guish these from phonetic representations.
28 Chapter 2. Phonetics

this behavior with your lips. If you hold them very tightly closed, they will stop the flow
of air. If you hold them apart, the air will flow between them unimpeded. But if you
hold your lips loosely together and then push air through them, you will get a mildly
embarrassing kind of noise from the flapping of your lips in the passing breeze. When
you make this kind of noise in your glottis, the resonating cavity of your head shapes
it into that mellifluous sound which is your voice, much as the lip-buzzing sound that a
trumpet-player blows into the business end of a trumpet is converted by the resonating
cavities in the trumpet into pleasant trumpety sounds.
The technical name for the sound produced by this rapid flapping of the vocal folds is
voicing. Segments which are produced while the vocal folds are vibrating are therefore
called ‘voiced’ sounds. Since this voicing takes place some distance away in the glottis, it
can be superimposed upon oral constrictions of various types, e.g., stops. The difference
between [p] and [b], both of which are labial stops, is that during the production of [p]
the vocal folds are pulled away from one another, allowing the air to flow through them
unobstructed (although the air is ultimately blocked by the lips). During the production
of a [b], by contrast, the vocal folds are pulled together, vibrating in the flow of air coming
up from the lungs. [b] is therefore a voiced labial stop, while [p] is a voiceless labial stop.
You can establish for yourself that this is the case with a simple experiment. Note
that you can make a very long [p] — in fact, you can make a [p] that lasts just as long
as you can hold your breath. On the other hand, if you try to make an equally long [b]
you will discover that you have a little aerodynamic problem. To keep the vocal folds
vibrating (which you must do so that the [b] does not turn into a [p]), you have to keep
pushing air up from your lungs through the loosely-closed glottis. But your lips are closed,
so there’s nowhere for the air coming up from lungs to go. Once the capacity of your
oral cavity is exhausted, you can’t push any more air through the glottis and voicing will
cease. (Attempting to exceed the capacity of your oral cavity will cause the look and feel
of imminent explosion.)
There are voiced versions of the alveolar and velar stops as well. [d] as in <duck> is
a voiced alveolar stop and [g] as in <get> is a voiced velar stop. There is, of course, no voiced
glottal stop. Since to make a glottal stop you must hold your vocal folds tightly closed,
allowing no air to escape, and to produce voicing you have to push air through the glottis
causing the vocal folds to vibrate, it is physically impossible to do both simultaneously.
There is yet another sound made with the lips blocking the flow of pulmonic egressive
air. This sound is [m] as in <mouth>. Can you tell whether [m] is voiceless or voiced? As-
suming that you replied ‘voiced’, you are correct. You can feel your vocal chords vibrating
during the articulation of [m] if you press your fingers to your throat at the point of your
larynx and say [mmmmm ðæts gʊd] (‘mmmmm that’s good’). We pointed out above that
you get an ‘explosive’ feeling if you try to make a very long [b] because of a build-up of
pressure from the need to push air through the glottis constantly to keep up the voicing.
But if [m] is voiced, why doesn’t your head explode when you make a really long [m]?
(You may safely test this.) Obviously, the pressure is being released somehow — the air
is getting out. Just as obviously, the air is not being released through your mouth, which
2.3. Articulatory Phonetics 29

is closed at the lips. This means it must have another escape route. Given the anatomical
structure of the human head, the possible escape routes are limited. A good candidate
would be one of those orifices connected to the vocal tract, since that’s where the pres-
sure is building up in the case of a long [b]. Now, as you know if you have ever (1) burst
out laughing while taking a drink, or (2) unwisely left your mouth open when diving into
water, your nasal sinus passage is directly connected to the oral cavity.9 The relationship
between the nasal and oral cavities can be seen in Figure 2.2 where the velum is where
air will divert between the two passages.
From Figure 2.2, it would appear that air should always be able to pass through the
nasal cavity, but this is not the case. The velum is a ‘movable’ part although it is difficult
to sense this movement consciously. In particular, it can (and does) move back and block
access to the nasal passage from the oral cavity. Fewer nerve endings in that general
area reduces your ability to feel the velaric movement, but you may try the following.
The vowel written [ɛ] in the IPA is the vowel of <bet> [bɛt]. The vowel written [ɛ̃] is
the vowel of French <fin> ‘end’ [fɛ̃].10 Pronounce the following: [ɛɛ̃ɛɛ̃ɛɛ̃ɛɛ̃ɛɛ̃]. It’s pretty
hard to do that, actually, except very slowly. If you do it, you will notice that nothing in
your mouth moves at all, nor does the airflow through the glottis vary (both [ɛ] and [ɛ̃]
are voiced). Why do the two vowels sound different? Because the velum is moving back
& forth to open (for [ɛ̃]) and close (for [ɛ]) access to the nasal sinuses.
Consonants produced with the velum open (i.e., lowered) are called nasals. As we
have already established, [m] is a labial nasal. Parallel to the alveolars [t] and [d] we have
the alveolar nasal [n], as in <nasal> [nezl ̩]. Parallel to the velar stops [k] and [g], we have
the velar nasal [ŋ] as in <sing> [sɪŋ]. Nasals are classified as stops, since there is an oral
occlusion at the normal stop positions.
To summarize, the set of stops (including the nasals) is given in the table in Figure 2.3.

labial alveolar velar glottal

voiceless p t k ʔ
voiced b d g
nasal (voiced) m n ŋ

Figure 2.3: Stops.

The official way to refer to stops, and consonants generally, is to state their voicing, place,
and manner, in that order. [b] is thus a ‘voiced labial stop,’ [n] is a ‘voiced alveolar nasal’11
and [k] is a ‘voiceless velar stop. Every speech sound must, in addition, be either oral or
9
It is true that the ears, and even the eye sockets (in some mysterious way), are also connected to
the naso-oral cavity, as demonstrated several times by invitees to a certain late-night TV show. These
connections are a lot less direct, however, and air is not released through these passages during speech.
10
If this is not a promising clue, it is also close to the sound that Bugs Bunny made at the beginning of
his standard question ‘What’s up doc?’ usually written as ‘eh’ and articulated as though he had a cold.
11
Since nasals are almost always voiced, the ‘voiced’ designation is sometimes left off for nasals.
30 Chapter 2. Phonetics

nasal. (The velum will either be raised or lowered.) Because the default value for speech
sounds appears to be ‘oral’ (more sounds are oral sounds than are nasal sounds), we typi-
cally specify this parameter only for those sounds which are nasal (i.e., we do not usually
call [k] a voiceless velar oral stop, for example, although it would be perfectly correct to
do so).
The segments we have examined so far all involved total obstruction of the airflow
through the oral cavity; however, it is possible to partially obstruct, rather than com-
pletely block, the flow of air. One technique for doing this is to force the air flow to go
through a very narrow passage. Since the air is coming up from the lungs at a more or less
constant rate, when it is forced through a small opening it must speed up significantly to
get out of the way of the rest of the air rushing up from the lungs. The narrow channel
causes this rapidly rushing air to become quite turbulent and these sounds are charac-
terized by a great deal of ‘noise’ (in the technical sense). Sounds that are made in this
manner are called fricatives. A typical example of a fricative, quite widely found in the
languages of the world, is the voiceless alveolar fricative [s] (as in ‘Simpsons’ [sɪmpsn̩z]).
An [s] is articulated by using the apex (tip) of the tongue to block all air flow through the
oral cavity except for a narrow channel between the tongue and the alveolar ridge. [s] has
a voiced counterpart in [z] (also in ‘Simpsons,’ but at the end). Indeed, a very good way to
get a feel for exactly what is going on with your vocal chords is to pronounce [szszszszsz]
— when you do this, you can feel the vocal chords engaging as you go from [s] to [z].
Grammars of the English type, like many grammars around the world, lack labial frica-
tives, though they do exist in some languages. Acoustically very similar to the labial frica-
tive is the labiodental fricative, which is produced by forcing the air through a very small
set of channels between the upper teeth and the lower lip. The official IPA symbol for the
voiceless labiodental fricative is [f], as in ‘failure’ [fæljr̩]. The voiced labiodental fricative
is [v] as in ‘veggie’ [vɛʤi].
Most English-type grammars also have a pair of interdental fricatives — quite rare,
cross-linguistically. These sounds are produced by placing the apex of the tongue be-
tween the teeth, forcing the air through small channels between the teeth and the tongue.
The voiceless interdental fricative is found in words like ‘thrilling’ [θrɪlɪŋ]. The voiced in-
terdental fricate is found in words like ‘bathe’ [beð].
The alveopalatal fricatives are produced with the blade of the tongue making a nar-
row closure in the region just behind the alveolar ridge at the very frontmost edge of
the (hard) palate. There is a voiceless one, [ʃ], as in ‘shake’ [ʃek], and a voiced one, [ʒ], as
in ‘measure’ [mɛʒr̩]. Frequently, a non-standard (i.e., non-IPA) symbol is used for these
sounds in North America. The voiceless alveopalatal fricative is often written [š] and the
voiced one [ž]. We will continue to use the standard, IPA symbols for these sounds but
they may appear in their non-standard forms elsewhere.
English type grammars typically lack both palatal and velar fricatives, though such
segments are not rare cross-linguistically. Many linguists treat [h], as in ‘hop’ [hɔp], as
a fricative. In this case, the frication/turbulence is found at the glottis (caused by air
rushing through vocal folds that are drawn together slightly). We will treat [h] as a glottal
2.3. Articulatory Phonetics 31

fricative here.
It is useful to deal with a rather peculiar type of segment at this point — one called
an affricate. The affricates start out with a stop closure, but instead of simply releasing
the closure and allowing the air to flow freely after the stop release, only a small portion
of the closure is released, leading to the type of narrow channel generally characteris-
tic of fricatives. These segments are thus combinations of stops and fricatives, neither
having the full duration of corresponding full segments of those types. English has two
of these affricates, a voiced-voiceless pair. They are produced in the alveolar and alveo-
palatal area (the stop portion being essentially alveolar, though slightly further back, and
the fricative portion being essentially alveo-palatal). They are called the alveo-palatal
affricates, the voiceless [ʧ], as in ‘chunk’ [ʧəŋk], and the voiced alveo-palatal affricate,
[ʤ], as in ‘junk’ [ʤəŋk]. As with the alveo-palatal fricatives, there are special symbols
in widespread use in North America for these two segments. The voiceless alveo-palatal
affricate is frequently written [č] and the voiced one [ǰ].
To summmarize what we have covered about fricatives and affricates, we have in-
cluded the chart in Figure 2.4.

labio-dental interdental alveolar alveo-palatal glottal

Fricatives
voiceless f θ s ʃ (š) h
voiced v ð z ʒ (ž)
Affricates
voiceless ʧ (č)
voiced ʤ (ǰ)

Figure 2.4: Fricatives and affricates.

We see the same general pairing of voiceless and voiced segments at each place of artic-
ulation (except for glottal, where it would make no sense) as we saw in the stop table in
Figure 2.3.
The remaining segments are harder to describe the articulation of and they do not,
unfortunately, fit into nice neat tables. English-type grammars have two liquids: [r] and
[l] as in ‘really’ [rili]. The articulation of [r] is quite complex. It is formally character-
ized as a ‘retroflex approximant’ — ‘retroflex’ because the tongue tip is curled backward
(such that the underside of the tip is parallel to the alveolar ridge) and ‘approximant’
because the tongue approaches but does not touch the alveolar ridge. The symbol [r] is
not the standard IPA symbol for this sound — the IPA symbol is [ɻ]. Due to the difficulty
of accurately reproducing this symbol when writing by hand as well as the difficulty in
remembering the exact orientation of it, we will not use it here but will substitute ortho-
graphic [r]. (The latter is also an IPA symbol — it stands for the sound of a ‘trill’.) Of the
many r-type sounds found in natural language, the English type is most unusual. The [l]
32 Chapter 2. Phonetics

is also, technically, an approximant, but a lateral approximant. It is made by pushing the

apex of the tongue up against the alveolar ridge and allowing the air to flow around the
sides of the tongue. The ‘lateral’ designation is because of this pattern of airflow. While
it may not, at first glance, seem to fit into the approximant category, as described above,
the critical portion of the articulation of [l] is its ‘laterality’ — the areas where the tongue
approximates but does not actually touch, the sides of the alveolar ridge. For convenience,
we will continue to use the term ‘liquids’ for [r] and [l]. To distinguish between these two
liquids, the terms rhotic (after the Greek rho) and lateral (as above) are used for r-like and
l-like sounds, respectively.
Nasals and liquids, in English, share something in common with vowels. They may
form what is called the nucleus of a syllable all on their own. The nucleus might be de-
scribed as the ‘heart’ of the syllable — if there is no nucleus, there is no syllable. Inter-
estingly, everyone has good intuitions about syllables and this really means that they
have good intuitions about syllabic nuclei. For example, if we were to say aloud a long
and unfamiliar word, you would be immediately able to tell us how many syllables were
contained in it. People appear to do this in the same way that they can follow the beat of
music by tapping their fingers or feet. For example, the word ‘antidisestablishmentarian-
ism’ has twelve nuclei and, thus, twelve syllables. Usually, it is vowel sounds which carry
the ‘beat’ (i.e., form the nuclei) however liquids and nasals may do so as well under cer-
tain conditions. When a consonant, such as a liquid or nasal, acts as a syllabic nuclei, we
call it a syllabic or vocalic consonant. Syllabic status for liquids and nasals is indicated by a
little bar under the standard IPA symbol for that segment as in the following examples:12
‘master’ [mæstr̩], ‘bird’ [br̩d]; ‘snap, crackle and pop’ [snæp krækl ̩ n̩ pap], ‘castle’ [kæsl ̩];
‘button’ [bəʔn̩], ‘fasten’ [fæsn̩]; ‘something’ [səʔm̩ ], ‘chasm’ [kæzm̩ ].13 Certain aspects of
syllable structure are complex and still not very well understood. The brief notes above
will be sufficient for our purposes.
As far as the major consonants needed to discuss most English-type grammars in a
somewhat superficial way, there are only two segments left. These are usually called
glides or semivowels.14 The labiovelar glide is transliterated [w] and found in words such
as ‘wacky’ [wæki]. It is made by rounding the lips and retracting the back of the tongue
toward the velum. The palatal glide has the official IPA symbol [j] and is found in words
such as ‘yucky’ [jəki].15 In this case, the blade of the tongue is placed very close to the
palate but not close enough to cause frication. The alternate ‘semivowel’ designation is
the result of the fact that these sounds are virtually identical to two of the vowel sounds
[u] and [i] (discussed below) except in their position within the syllable (critically not the
12
This subscripted bar is one of the set of diacritics used to supplement the IPA that we mentioned
earlier.
13
Some people, transliterate the syllabic [r] with a vowel symbol specially designed for that purpose,
[ɚ].
14
Strictly speaking, phonetically, these two sounds also fall under the ‘approximant’ category but it is
more common and, we think, less confusing, to use the term ‘glide’.
15
Many people use the symbol [y] to designate this sound, but as the IPA symbol [y] is officially used for
a relatively critical vowel sound that exists in a number of languages, it is a good idea to keep these distinct.
2.3. Articulatory Phonetics 33

nucleus position). Their position within the syllable is like that of consonants.
Although not always covered in introductory texts, we consider the following sounds,
universally present in English-type grammars, to be worth describing. First, there are
dental stops in addition to labial, alveolar, velar and glottal. If you contrast the pronoun-
cation of ‘ten’ [tɛn] and that of ‘tenth,’ you will find that the latter should be transliterated
[tɛn̪θ], where [n̪] is the symbol for the dental nasal. Similarly, for those speakers who have
a stop before the final fricative in a word such as ‘eighth’, it is a voiceless dental stop: [et̪θ].
More frequent is the voiced alveolar flap or tap (use either term). This sound is found in the
middle of words like ‘middle’ and ‘butter,’ and its official IPA symbol is [ɾ]. ‘Middle’ and
‘butter’ should therefore be transliterated [mɪɾl ̩] and [bəɾr̩], respectively.

2.3.2 Vowels
English-type grammars are extremely rich in vowels. Whereas with consonants there
were only a few ‘funny’ symbols of the IPA that needed to be learned, in the case of vowels
virtually all of the symbols have unexpected values for speakers of English. This may seem
burdensome but since vowel symbols will be in extensive use, it is unavoidable.
Consonants are generally classified according to the point of greatest constriction in
the vocal tract — the number of such points used in human language is quite restricted.
Vowels, by contrast, are distinguished from one another by the shape of the (usually oral)
resonating cavity. The shape of this cavity can be modified in three major ways. The
lower jaw can be raised or lowered (making the resonating cavity smaller or larger, re-
spectively). The body of the tongue can be bunched up in the back of the oral cavity or the
blade of the tongue can be pushed up toward the palate. Finally, the lips may be rounded
or spread during the articulation of the vowel. Thus, three important parameters for
vowel articulation are:

• height of the lower jaw (high, mid, low)

• position of tongue (front, central, back)

• lip rounding (rounded, unrounded)

If you raise your lower jaw as high as possible (without making such a tight constriction
in the oral cavity that you make a consonant) and push your tongue forward with your
lips spread (i.e., not rounded) you can produce the sound [i] of ‘beat’ [bit]. [i] is therefore
a high, front, unrounded vowel. When the doctor wants to look down your throat, he
or she says ‘say [a]’ (if he or she says ‘say [i]’, it may be time to look for a new doctor)
because [a] is a low vowel (low, central and unrounded). [a] is the vowel of ‘father’ [faðr̩].
The vowel [u], like [i] is made with the lower jaw up (i.e., it’s a high vowel). However,
unlike [i], to make an [u] the back of your tongue is bunched up towards the velum (but
not close enough to cause frication) — it is therefore a back vowel. Like all back vowels
in English-type grammars, [u] is round.
34 Chapter 2. Phonetics

These three vowels more or less define the edges of the vowel space. You should
practice making the following sequence of sounds: [iuaiuaiuaiuaiua]. This will help you
get a feel for the space within which most vowels are produced. This space is typically
represented as in the figure below.

HI
i
u
MID

a
LO
FRONT CENTRAL BACK

Figure 2.5: The vowel space.

English-type grammars have many more vowels than just these three, of course. For
example, if you say [iaiaia] very slowly and stop the jaw-lowering involved about halfway
between the end of the [i] and the start of the [a] you’ll be producing a mid, front, un-
rounded vowel: [e], as in [bet] ‘bait.’ Similarly, in the back area, if you make a vowel
about half-way between a high [u] and a low [a], you’ll be producing an [o], as in [bot]
‘boat.’ These five vowels are very widely found in the languages of the world.
At the high and mid heights, however, English-type grammars have additional vowels.
These contrast with the vowels cited above at the high and mid points by virtue of the
degree of vocal tract constriction required for their articulation. Vowels such as [i] and [u]
are generally called ‘tense’ vowels because the muscles of the vocal tract are constricted
during their articulation. By contrast, the vowel of a word like ‘bit,’ which is, like [i], high,
front and unround, is called ‘lax’. The IPA symbol for the vowel of ‘bit’ is [ɪ], thus [bɪt].
Corresponding to tense [u] in the high, back, round space we have the lax vowel [ʊ] of
‘put’ [pʊt]. Articulated in the mid, front, unround area (near [e]) is the lax vowel [ɛ] of
‘bet’ [bɛt]. Finally, corresponding to the tense version of the mid, back, round vowel ([o]),
we have a lax version [ɔ] in words such as ‘law’ [lɔ].
One of the big differences between various dialects of English concerns the distri-
bution of vowel sounds in specific words, often the sounds [ɔ] and [a]. In some dialects,
‘caught’ is pronounced as [kɔt] and ‘cot’ as [kat]. In others, the two words are homophonous
— both are pronounced as [kɔt]. Similarly, some speakers say [pat] and [sari] for what oth-
ers pronounce as [pɔt] ‘pot’ and [sɔri] ‘sorry.’ You can see how convenient the IPA is for
allowing us to describe the differences between various linguistic systems!
English-type grammars have two other simple vowels. The first is the low, front, un-
round vowel [æ] of [bæt] ‘bat.’ The other is the so-called schwa (its name is pronounced
2.3. Articulatory Phonetics 35

[ʃwa]), whose symbol is [ə]. This vowel is found in words such as [bət] ‘but’ or [sofə] ‘sofa.’
Both [æ] and [ə] are lax, as is the other low vowel of English [a].
With the introduction of the tense/lax distinction, we now have four parameters used
to describe vowels in English, as below.

1. height of the lower jaw (high, mid, low)

2. position of tongue (front, central, back)

3. lip rounding (rounded, unrounded)

4. tense vs. lax

We are not quite done with vowels yet, however, In addition to the ‘simple’ vowels
just described, English has a number of diphthongs — i.e., vowels which involve move-
ment from one vowel position to another during their articulation. By tradition, they are
transliterated as sequences of vowel+glide. These diphthongs include [oj] as in [boj] ‘boy,’
[aj] as [bajd] ‘bide,’ and [aw] as in [bawt] ‘bout.’ The trajectory of tongue/jaw movement
during these diphthongs is indicated in Figure 2.6.16

i
HI
u
oj
MID
aw
aj
o
LO a
FRONT CENTRAL BACK

Figure 2.6: The diphthongs of English.

For completeness’ sake, we will add two more vowel symbols, both of which stand
for central vowels. The first is called ‘barred i’ [ɨ] and represents the sound in the final
syllable of <roses> [rozɨz]. It is similar in articulation to schwa but somewhat higher. The
second symbol is a carat [ʌ]. This symbol is used by some people in place of schwa when
the syllable is stressed. For various reasons, we do not follow this usage and instead use
schwa in both stressed and unstressed syllables.
16
To be technically precise, the tense mid and high vowels of English are also diphthongs, front vowels
having a [j] off-glide and back ones a [w] off-glide. This causes some people to transliterate them [ij], [ej],
[uw], and [ow]. We do not do this, for reasons which are too complicated to go into.
36 Chapter 2. Phonetics

2.4 Summary of relevant IPA symbols

The tables below present listings of the symbols we have studied so far as well as some
examples of words which generally, in English-type grammars, exemplify the use of these
symbols. Again, the critical descriptive information for consonants consists of place of
articulation, manner of articulation, and voicing. For the vowels, what is important are
the features height, backness, lip-rounding, and tense or lax.
A number of the ‘unusual’ IPA symbols have their own names. Since the names fre-
quently have something to do with the sounds represented by the symbols, they may be
helpful to learn. Greek symbols retain their Greek letter names: θ is ‘theta’; ŋ is ang(ma);
ɛ is ‘epsilon’; and ʊ is (sort of) ‘upsilon.’ Two symbols are borrowed from Old English:
ð is ‘eth’ (pronounced [ɛð]; and æ is ‘ash.’ For ease of reference alone, we include the
following: ɔ is called ‘open o’ and the ə is called ‘schwa.’

labial dental
& &
labio- inter- alveolar alveo- palatal velar glottal
dental dental palatal
stops
voiceless p t̪ t k ʔ
voiced b d̪ d g
nasal m n̪ n ŋ
fricatives
voiceless f θ s ʃ h
voiced v ð z ʒ
affricates
voiceless ʧ
voiced ʤ
liquids
rhotic rɾ
lateral l
glides w j

Figure 2.7: Consonants

2.4. Summary of relevant IPA symbols 37

i
HI
ɪ ɨ u
ʊ
e ə
MID o
ɛ
ɔ
LO æ a
FRONT CENTRAL BACK

Figure 2.8: Vowels (without diphthongs)

ʔ [bəʔn̩] ‘button’; [hi hɪʔ mi] ‘he hit me’

θ [θrot] ‘throat’; [bæθ] ‘bath’; [tiθ] ‘teeth’
ð [ðiz] ‘these’; [beð] ‘bathe’; [tið] ‘teethe’
ʃ [liʃ] ‘leash’; [mɛʃ] ‘mesh’; [ʃɪvr̩] ‘shiver’
ʒ [liʒr̩] ‘leisure’; [mɛʒr̩] ‘measure’
ʧ [ʧr̩ʧ] ‘church’; [bæʧ] ‘batch’
ʤ [ʤəʤ] ‘judge’; [bæʤ] ‘badge’
ŋ [sɪŋ] ‘sing’; [fɪŋgr̩] ‘finger’
ɾ [bəɾr̩] ‘butter’; [mæɾr̩] ‘matter’ & ‘madder’
j [jɛlo] ‘yellow’; [kjut] ‘cute’; [traj] ‘try’
i [bit] ‘beat’ and [fit] ‘feet’
ɪ [bɪt] ‘bit’ and [fɪt] ‘fit’
ɨ [rozɨz] ‘roses’
e [bet] ‘bait’ and [fet] ‘fate’
ɛ [bɛt] ‘bet’ and [fɛt] ‘fête’
æ [bæt] ‘bat’ and [fæt] ‘fat’
u [but] ‘boot’ and [fud] ‘food’
ʊ [pʊt] ‘put’ and [fʊt] ‘foot’
o [bot] ‘boat’ and [flot] ‘float’
ɔ [bɔt] ‘bought’ and [fɔt] ‘fought’
a [faðr̩] ‘father’
ə [bət] ‘but’ and [(ɛlmr̩) fəd] ‘(Elmer) Fudd’
aj [bajd] ‘bide’ and [faj] ‘fie’
oj [bojd] ‘Boyd’ and [fojl] ‘foil’
aw [bawt] ‘bout’ and [lawd] ‘loud’

Figure 2.9: Examples for ‘unusual’ IPA symbols.

38 Chapter 2. Phonetics

2.5 Exercises
1. [θ] is a

(a) voiced labiodental fricative.

(b) voiceless alveolar fricative.

(c) voiceless interdental fricative.

(d) voiceled palatal fricative.

2. A voiceless alveolar fricative is transcribed in the IPA with the symbol

(a) [ʃ]

(b) [s]

(c) [θ]

(d) [ɚ]

3. [u] is a

(a) a high, back, round vowel.

(b) a high, front, unround vowel.

(c) a high, back, unround vowel.

(d) a high, front, round vowel.

4. Examine the following dialogue between children and answer the question. [Note: This
and the following excerpts were transcribed from an actual conversation between young
children. They are mildly incoherent and have many performance errors thus they are
quite similar to adult speech …]

A: aj æ fɔr sɪstr̩z.

B: ju dən hæv fɔr sɪstr̩z. jə oni hæv θri.

A: suzi kæθi mi n̩ bani.

B: ʔɪts na kawntɪŋ ju. jr̩ naɾə sɪstr̩. jr̩ naʔ jr̩ sɪstr̩.
2.5. Exercises 39

How many sisters does Miss A have?

(a) Four, but she thinks she has three.

(b) Three, but she thinks she has four.

(c) Five, but she thinks she has 4.

(d) The details of Miss A’s family situation cannot be deduced from this dialogue.

5. Examine the following child’s monologue and answer the question.

A: aj hæv ðiz awʔ ɔrɛɾi ɪn ðɪs wən kem awt. læs najt maj mawθ əz ɔl bliɾɪŋ. no, naʔ
frəm ðæt, maj hol gəm ərawn æn maj hol tiθ. ɪt ɔmos ɪʔ drɪpt awɾə maj mawθ. maj
brəðr̩ hæʔ wɛl hi hæd e tiθ awʔ. hɪz mawθ wəz ɔl bliɾi ən i hæʔ hi kʊnnivən stæn
frəm ɪt. ɪʔ ɪɾɔlmos drɪpdawn tu ɪz to. tu ɪs fit. æn maj mam hæɾən apr̩eʃɪn. ʔapr̩eʃɪn
ɔnr̩stəmɪk.

How far did the blood from A’s brother’s mouth almost drip down to?

(a) To his bellybutton.

(b) To his knees.

(c) To his ankles.

(d) To his toes or to his feet, the text is a little unclear on the point.

6. Examine the following children’s conversation and answer the question.

A: pɛpr̩z ə ʤr̩mɪn ʃɛpr̩d.

B: jə no wət. ʔə boj sɛd ðɛt hi ðɛt pɛpr̩ hɛd ʔəm rebiz. bət mi n̩ hɪm wən ovr̩ tə pɛɾɪt.
ɛn ɪ dɪdn̩ ivn̩ æv rebiz. hi wəz rɔŋ.

A: pɛpr̩ dɪd.

B: aj pɛɾəd hɪm. hi dɪdn̩ ivn̩ hæv rebiz. f i hæd rebiz hi wʊɾə bɪʔ mi. bɪʔn̩ mi ʔn̩ ʔəm
dajd. əhə.

C: maj dɔgz hævɪŋ bebiz n̩ hi dɪdn̩ bajt.

B: hæv rebiz.

C: bebiz.
40 Chapter 2. Phonetics

B: bebiz. wɛl ðe donʔ bajʔ wən ðe hæv rebiz. bebiz. bət. wən ðe ʔæv ʔm̩ rebiz jæ ðe du
bajt.

Which of the following statements is true?

(a) According to A, ‘Pepper’ had rabies.

(b) According to B, ‘Pepper’ did not have rabies.
(c) C is a little confused about whether the discussion concerns ‘babies’ or ‘rabies.’
(d) All of the above.

7. Examine the following child dialogue and answer the question.

A: wəns wəns hi kip ʤəmp ʤəmpɪŋ ʤəmpɪŋ æn ðɛn ðɪs θɪŋ slajs awn ðɛn hi fɛl ovr̩ tu
ðəʔ əðə bɛd æn i star krajɪn ɛn aj kʊʔən gɛ tə bɛd so aj aj hæftə wek əp pəʔəm bæk
ɪm maj krɪb.
B: ʔɪn jr̩ krɪb.
A: no naʔ ɪm maj krɪb. aj don hæv ə krɪb.
B: ju sɛ pəʔɪm bæk ɪn jr̩ krɪb.
A: aj min ɪn hɪz krɪb. aj don hæv ə krɪb.

Does Miss A have a crib?

(a) She said she did, but she doesn’t.

(b) She said she didn’t, but she does.
(c) No, B has a crib.
(d) No, B said she had a crib (but she doesn’t).

8. Examine the following first-grader monologue and answer the question.

A: bɪfɔr aj juz tə hæv ə br̩d. ɪʔ dajd. aj juztə hæv ə ræbɪt. ðæʔ dajd. wi gaɾə dɔg ðæʔ
dajd. ən wən wi gaɾə kæt wi sold ɪt əwe. so wi gat dɪsajdəd tə gɛɾənəðr̩ dɔg.

Given A’s past experience with pets, what are the prospects for his new dog?

(a) Promising — it will probably lead a long and healthy life.

(b) Excellent — it can look forward to many years of happy pethood.
2.5. Exercises 41

(c) Fine — I’m sure it will be well taken care of.

(d) Not so great, it should quickly find itself a new home.

9. Examine the following first-grader conversation and answer the question.

A: æn aj hæv ə θawzɛnz ɔmos θawzənz əv ænəml ̩z ɪm maj bɛəd. aj oni ga wən tr̩dl ̩.

B: aj ləv ræbɪs.

C: aj gaʔʔəbawʔ tɛn fajv ræbiʔs. fajv ræbɪʔs.

A: aj hæv ə dɔg. hɪz nem ɪz dæjəndi.

B: aj əv ə pɛwokiʔ bəɾi dəzən tɔk jɛt.

A: æn hiz ə æn izə winr̩ dɔk. hi wɪgl ̩z ɪz teəl ɛvri tajm.

B: du ju kɔlɪm ɛnitajm ə winr̩. ɛnitajm.

A: hi ʤəmʔs ɔn ə ʧer̩z ən ɛvriθɪŋ. maj faðr̩ waps ɪm wən.

B: aj gaɾə slo ænɪml ̩. ən gɛs hu hi ɪz. ðə tɔrdɪs.

What is B’s slow animal?

(a) A rabbit.

(b) A wienerdog.

(c) A tortoise.

(d) A parakeet.

10. One of the four strings below is likely to be accurate — the others have serious errors
which make it very unlikely that you or we would ever say them that way. Circle the one
which has the good transliteration in it.

(a) ɪf ju kn̩ rid ðɪs, jə məjt pæs ðɪs klæs.

(b) ðiz kwɛSʧin̩z ar rili stʊpid.

(c) maj hɛad hurts.

(d) ði aj pi e ɪz ə ril pain ɪn ði æs.

 Chapter 3
Phonology

Contents
3.1 Further Phonetic Details . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 The Phoneme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Minimal Pairs & Allophones . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4 More English allophony . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Rule ordering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.6 Some Allophony in other languages . . . . . . . . . . . . . . . . . . . . 51
3.7 Universal Properties and Language-Specific Properties . . . . . . . . . 52
3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.9 Solving phonology problems . . . . . . . . . . . . . . . . . . . . . . . . 53
3.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.1 Further Phonetic Details

The symbols of the IPA we have learned allow us to represent in much greater detail the
speech of most people with English-type grammars than would ever be possible using the
English writing system alone, which is unsystematic, arbitrary, and confusing. The IPA
also includes symbols to transcribe the basic segments of all other known human lan-
guages. However, as linguists studied in greater detail the sounds of even well-known
grammars, such as those of the English type, they discovered that there were small de-
tails of articulation which the IPA alone failed to capture. For example, depending upon
the dialect, speakers pronounce [æ] differently, sometimes further front and sometimes
further back. The IPA has been supplemented with many additional symbols (which, for-
tunately, you don’t have to learn, even though we did). So [æ̟ ] indicates, through the use

43
44 Chapter 3. Phonology

of a diacritic (in this case, a subscripted plus-sign), a further front low front unround
vowel such as is found in some pronunciations of words such as ‘cat.’ The contrast be-
tween a ‘more rounded’ and a ‘less rounded’ type of [o] can be indicated as well ([o̜] for
the less rounded and [o̹] for the more rounded). A voiceless velar stop produced with
lip-rounding, as in the normal pronunciation of a word like ‘cool,’ is indicated as follows:
[kw ul]. The set of possible symbols increased dramatically as phoneticians made more
and more precise observations about the details of articulation.
With the advent of the use of computers to analyze the acoustic properties of human
speech, however, it quickly became apparent that the goal of having a different symbol
for each ‘speech sound’ could never be attained. The computer revealed that, in fact,
each of the [æ]’s that a single speaker produces (if one were to record them saying, e.g.,
[kæt] ten times) is slightly different. We would need an infinite number of symbols with
diacritics to represent all the details of speech, and that’s a lot of symbols…
It became clear that a better goal was not to pay attention to all of the differences be-
tween the [æ]’s or [k]’s that speakers produced, but to focus on differences that might be
used in some human language to indicate a change in meaning. An example might help
with this. Speakers with English-type grammars normally produce round consonants be-
fore round vowels (remember, all back vowels are also round in English-type grammars).
So we say [kw ul] ‘cool,’ [tw u] ‘two,’ and [mw un] ‘moon.’ However, if you were to say [kul]
or [mun] (with no rounding of the consonant), everyone would know that you meant
‘cool’ and ‘moon,’ respectively. In the case of speakers with Marshallese-type grammars,
however, the difference between round and unround consonants is very important. For
example, [am nikw nikw ] means ‘your cloth’ but [amw nikw nikw ] means ‘our (i.e., belonging
to me and some other people, but not you) cloth.’ Marshallese speakers thus have to pay
attention to the rounding of [m] (and of other consonants as well), whereas speakers of
English do not. Since Marshallese [am] ‘your’ and [amw ] ‘our (excluding you)’ differ only
in the lip-rounding that we hear on the labial nasal, any speaker of Marshallese who did
not attend to this contrast would be very confused (and very confusing).
On the other hand, the difference between [ʃ] and [ʧ] is of no consequence in Mar-
shallese. For example, the verb which means (according to the dictionary) ‘to go bare-
foot’1 can be pronounced [ʃintəbw ] or [ʧintəbw ] — no Marshallese speaker will pay any
more attention to this difference than you would to the difference between [kw ul] and
[kul].2
The contrast between round and unround [m] or [k] is thus very different from the
distinction between a slightly fronter [æ] (i.e., an [æ̟ ]) and a slightly backer one (indicated
in the IPA by [æ̠ ]). No human language uses the difference between these [æ]’s to signal a
difference in meaning, in the way Marshallese uses the [m]:[mw ] contrast, or English the
[ʃ]:[ʧ] one.
1
This verb also means ‘to eat only one food.’
2
Correspondingly, we use two different phonetic transcription systems depending upon whether we
want to illustrate all distinctions present (narrow transcription) or simply those distinctions which are
contrastive in a language (broad transcription).
3.2. The Phoneme 45

The second important implication of the difference between the status of rounding or
affricates in English and Marshallese is that it appears that each grammar uses some set
of features of sounds to pay attention to and ignores others (though the ones they ignore
may be just as present as those being attended to, as is the case with consonant rounding
in English or the [ʃ]:[ʧ] variation in Marshallese). It appears, then, that part of what you
know, as a speaker with a particular grammar, is which aspects of the speech stream mat-
ter for conveying differences in meaning and which ones do not.3 Since the knowledge of
what to ignore and what to pay attention to appears to vary from grammar to grammar
(so that, e.g., Marshallese-type grammars differ from English-type grammars in this re-
gard), you must learn to ignore certain aspects of the speech signal when you learn your
grammar. The next section will explore the nature of the knowledge you acquire with
respect to this.

3.2 The Phoneme

Imagine that you are about to get married (or, if that makes you too nervous, or seems
too unlikely, imagine that you are about to attend someone else’s wedding — if that too
is beyond the scope of your imagination, you will have to develop your own analogy).
The wedding ceremony — to be a real wedding — must have certain features. For exam-
ple, someone other than the two people getting married must be present (to perform the
ritual ceremony). Depending on your culture, various other things will have to happen
if the two people are to be truly married. One could say that the things that determine
whether or not what happens is actually a wedding (like is there someone there for you
to marry, is there someone there to perform the ceremony, is that person lawfully em-
powered to marry a given couple, etc.) are criterial as to whether or not a wedding has
taken place. In addition to these criterial features, every wedding will have many many
features which are incidental — i.e., irrelevant to the question of whether or not what took
place was actually a wedding. For example, the weight of the groom does not determine
whether or not a wedding has taken place, nor does the time of day or year, the size of
the bride’s feet, the number of people named ‘Bob’ who attend, or an infinite number of
other aspects. Note that every wedding has these incidental features – at every wedding
the groom weighs some amount, it takes place at a given time of day and year, the bride’s
feet have a certain dimension, there are some number (≥ 0) of people named ‘Bob,’ etc.
Similarly, we could say that in English-type grammars the difference between [ʃ] and
[ʧ] is criterial for the determination of meaning, whereas the difference between [m] and
[mw ] is incidental — i.e., present but not decisive for whether or not the word ‘moon’ is
interpreted by a speaker of English to be the round white object which sometimes ap-
pears in the sky at night. By contrast, in Marshallese the difference between [ʃ] and [ʧ] is
incidental to the determination of the meaning of a given string, whereas the difference
between [m] and [mw ] is criterial. Speakers must produce and pay attention to those seg-
3
Like all linguistic knowledge, this knowledge is, unfortunately, tacit, or hidden from consciousness.
46 Chapter 3. Phonology

ments which determine meaning differences (otherwise they won’t know what anyone
is saying) but can safely ignore those differences which do not influence meaning (who
cares if you say the [m] of ‘moon’ with lip-rounding or not? it’s going to mean ‘moon’
either way). The sound differences that matter for a given grammar are called phone-
mic differences. Differences that do not matter for the purpose of determining meaning
(but are present, like the dimensions of the bride’s feet at the wedding) are called pho-
netic differences. Phonemic sounds are contrastive to the speaker and The set of sounds
of a given grammar, defined in such a way as to list only those aspects of their produc-
tion that are criterial for the determination of meaning, are known as the phonemes of
that language. Thus an /m/4 which has no indication as to whether or not it involves
lip-rounding is a phoneme of English, whereas in Marshallese /m/ (an unrounded labial
nasal) and /mw / (a rounded labial nasal) are separate phonemes. Phonemes are what one
must learn to pay attention to in order to ultimately derive accurate semantic represen-
tations. Again, learning what sounds have phoneme status is part of acquiring particular
languages, since different languages have different phonemes.

3.3 Minimal Pairs & Allophones

How can we determine what the set of phonemes for a given grammar are? One technique
which works quite well is that of substitution. The idea is quite simple. Since, in English
[bæt] means ‘bat’ and [pæt] means ‘pat,’ the difference between [b] and [p] is a phonemic
one. This is shown by the fact that if we substitute the [b] of [bæt] with [p], we get a
difference in meaning. Words which differ in meaning and have that difference signalled
by just a single segment are called minimal pairs (meaning ‘minimally-distinct pairs’ of
words). Since the only difference between [bæt] and [pæt] resides in the [b]:[p] contrast,
we can infer that these two segments are used to signal meaning differences. From the
point of view of people who have English type grammars, these sounds are contrastive or
phonemic. On the other hand, if we substitute an unrounded [m] for the [mw ] of [mw un]
‘moon,’ giving [mun], the newly created word may sound a little funny, but it still means
‘moon.’ The difference between [m] and [mw ] is therefore not a phonemic one – they are
non-contrastive.
Distinctions which are merely phonetic (rather than phonemic), although they do not
play any role in determining meaning, are generally found to have a well-defined distri-
bution. For example, if you isolate the vowel of ‘men’ (by starting out to say ‘men’ but
then, instead of producing the final nasal, simply prolonging the vowel) you will discover
that this vowel is, for speakers with English-type grammars, nasalized (i.e., the velum is
lowered as it is for nasal consonants). It contrasts, therefore, with the vowel of a word
such as [mɛt] ‘met’ which (as you will discover if you isolate it using a similar technique)
is not nasalized. That is, if we want to be phonetically precise, ‘men’ should be transliter-
4
Phonemes are placed between slashes, to keep them clearly distinct from phonetic segments, which
are in square brackets.
3.3. Minimal Pairs & Allophones 47

ated [mɛ̃n]. (The ‘tilde’ diacritic indicates nasalization of vowels.) In the output of English
speakers, we find both [ɛ] (in ‘met,’ ‘bet,’ etc.) and [ɛ̃] (in ‘men,’ ‘Ben,’ etc.). Similarly,
in the speech of speakers with French-type grammars we find both these vowels, [ɛ] in
words like [fɛ] ‘fait’ (fact), [ɛ̃] in words like [fɛ̃] ‘fin’ (end). Notice that the same segments
are contained in the phonetic output of both types of speakers. However, as the exam-
ples given above show, in French if you substitute [ɛ̃] for the [ɛ] of ‘fait’, you get [fɛ̃] ‘fin,’
which means something completely different. However, if you substitute the [ɛ̃] of ‘men’
into ‘met’ in English, the resulting [mɛ̃t] may sound a little odd but the listener will still
understand you to have said the past tense of the verb ‘meet’. They may infer that you
have a cold or that you are performing that phonetic experiment which involves placing
an olive up your nose but no one would think you said anything but ‘met’ (i.e., the substi-
tution does not lead to a change in meaning, unlike in French). While French [fɛ] and [fɛ̃]
form a minimal pair, we cannot construct two words with different meaning in English
using the [ɛ]:[ɛ̃] contrast.
In fact, all English vowels are nasalized before nasals and only before nasals (barring
some physical — and therefore non-linguistic — problem, such as a cold). In French, as
[fɛ̃] ‘fin’ shows, this same statement is not true (we find nasalized vowels which are not
before nasals). Since the difference between [ɛ] and [ɛ̃] is a phonemic one in French-type
grammars, we refer to [ɛ] and [ɛ̃] as phonemes of French.
But what is the relationship between nasalized and non-nasalized vowels in English-
type grammars? As we’ve seen, since the distinction cannot express a difference in mean-
ing, there is no question of the nasalized and non-nasalized vowels being separate phonemes.
The distribution of nasalized vowels is, in English but not in French, completely pre-
dictable. We can state the prediction in the form of the following ‘rule’ of grammars of
the English-type:

(3.1) Rule 1: Nasalize vowels before nasals.

This rule5 accounts for the fact that in English nasalized vowels occur always and only
before nasals. The vowel of [mɛt] will not get nasalized by Rule 1, for example, since it
is not before a nasal. On the other hand, imagine that in your brain you store the word
‘men’ as /mɛn/ (without nasalization). When you run /mɛn/ through the grammar, a
process such as that in (3.2) will take place.

(3.2) /mɛn/ =⇒ (via Rule 1) [mɛ̃n]

The stored form /mɛn/ is converted by the application of (3.1) into a form with a nasalized
vowel, which you then send on to the articulatory system and ultimately pronounce with
nasalization.
5
Note that we are using the term ‘rule’ in the standard way it is used in linguistics. It is quite different
from a rule like ‘No pushing at the drinking fountain’ which one can follow or not follow as one chooses. In-
stead it represents the knowledge that a speaker has which gives rise to the distribution of phonetic features
such as vowel nasalization.
48 Chapter 3. Phonology

As we pointed out above, English-type grammars also generally round consonants

which occur immediately before round vowels. We can state this ‘consonant rounding’
rule as in (3.3).

(3.3) Rule 2: Make consonants round before round vowels.

Now imagine that we have ‘moon’ stored as /mun/. To derive the actual pronounced form
of this word, the grammar applies its phonological rules to the phonemic form /mun/
which is the input to the grammar. The result is a phonological derivation, such as that
in (3.4) for ‘moon.’

(3.4) /mun/ =⇒ (via Rule 1) [mũn] =⇒ (via Rule 2) [mw ũn]

The final form has both nasalization of its vowel (by Rule 1) and lip-rounding of its conso-
nant (by Rule 2). Note that since Rules 1 and 2 apply to all input forms to the grammar, the
output of these rules, taken together, will create a situation in which [ɛ̃] will occur only
where [ɛ] does not (before nasals), and vice-versa. The two phonetic segments [ɛ̃] and [ɛ]
are in complementary distribution — one never occurs in the same environment as the
other.
This type of analysis carefully distinguishes between what kind of information you
store in your head about the sounds in a word and what kind of instructions you send to
your articulators in order to get them to produce the word. At the level of mental stor-
age, you need only store enough information about ‘men’ to distinguish it from ‘mean,’
‘moan,’ ‘moon,’ ‘man,’ ‘met,’ ‘mess,’ etc. Since nasalization of vowels and rounding of con-
sonants is never relevant to keeping meanings distinct in English-type grammars, there is
no reason to store nasalization and consonant rounding information in our brains about
English words. Nasalization and consonant rounding are predictable aspects of English-
type grammar output. This contrasts sharply with the cases of French (with respect to
nasalization) and Marshallese (with respect to consonant rounding). If we did not store
nasalization information in French, [fɛ̃] and [fɛ] would both be stored /fɛ/. Similarly, if
we did not store consonant roundedness information in Marshallese, [amw ] ‘our’ and [am]
‘your’ would both be stored as /am/. Clearly we cannot write straightforward rules for
the stored forms of these French and Marshallese words which would insert nasalization
and rounding in just the right words.6
The segments used in our mental dictionary are the phonemes of the language. As
a result of the application of phonological ruels such as those in (3.1) and (3.3), these
phonemes can be realized in a variety of ways. So, for example, the phoneme /m/ in En-
glish can be realized as [mw ] (before round vowels) or as [m] (elsewhere), given Rule 2.
The phoneme [ɛ] can be realized as [ɛ̃] (before nasals) or as [ɛ] (elsewhere), given that
English grammars have Rule 1. The various phonetic realizations of a phoneme (and
6
Obviously, we can posit a rule which, for French, says ‘nasalize the [ɛ] in the word for ‘end,’ but notice
that since the rule explicitly mentions the word that it applies to, we might just as well posit the nasalization
as an inherent property of the word and get rid of the ‘rule’ altogether.
3.4. More English allophony 49

we have not given anything like a complete list here) are called the allophones of that
phoneme. So, in English-type grammars, the phoneme /ɛ/ has the allophones [ɛ̃] and
[ɛ] (at least — as we noted, the list is not complete), while the phoneme /m/ has at least
the allophones [mw ] and [m]. It will be useful to think of the phonological component as
consisting of two levels of abstract mental representations with a computational system
that ‘converts’ one type of representation into another. The most abstract level is the
level of the phoneme, or the underlying representation (UR) and the less abstract is the
phonetic level, or surface representation. In the production of speech, the phonolog-
ical computation system converts underlying (phonemic) representations into surface
(phonetic) representations. These surface representations are then further converted
into articulatory gestures (a process external to the grammar). In the comprehension of
language, relevant auditory input (speech sounds) is converted into surface (phonetic)
representations (again, a process external to the grammar). The surface representation is
then converted by the phonological component into an underlying (phonemic) represen-
tation. Again, unpredictable, and therefore important to remember, information is stored
only at the underlying or phonemic level.

3.4 More English allophony

When a native speaker with an English-type grammar says the word ‘pick’ one can ob-
serve a burst of air following the release of the stop closure of the initial segment. You
can test this out by putting a candle, match, or lighter flame in front of your lips when
you say ‘pick’ (not too close). The flame will flicker dramatically after the release of the
stop. This burst of air is known as ‘aspiration.’ All English voiceless stops are aspirated
when they are the initial segment of a stressed syllable. Thus ‘pick’, ‘tick’, and ‘kick’ are
pronounced, by native English-type speakers, as [ph ɪk], [th ɪk], and [kh ɪk].7
The aspiration of voiceless stops is not found when the voiceless stop is not syllable-
initial. So, e.g., in words like ‘spat’, ‘stat’, and ‘scat’ there is no aspiration ([spæt], [stæt],
and [skæt]). The aspiration of voiceless stops is thus predictable — a voiceless stop will be
aspirated if it is the first segment of a stressed syllable, otherwise it will not. English-type
grammars thus have a rule such as in (3.5) below.

(3.5) Rule 3: Aspirate voiceless stops at the beginning of stressed syllables.

We can thus say that the phonemes /p/, /t/, and /k/ have allophones as in (3.6-3.8).

(3.6) /p/ has the allophone [ph ] at the beginning of stressed syllable, the allophone [p]
elsewhere

(3.7) /t/ has the allophone [th ] at the beginning of stressed syllable, the allophone [t]
elsewhere
7
A superscripted ‘h’ is the diacritic for aspiration.
50 Chapter 3. Phonology

(3.8) /k/ has the allophone [kh ] at the beginning of stressed syllable, the allophone [k]
elsewhere

It should be clear at this point that allophones are generated by phonological rules.
One more case of allophonic alternation in English-type grammars should suffice to
make the principle of the phoneme clear to you. If you say the words ‘bat’ and ‘bad’ to
yourself, you will notice that there is in fact a significant difference in the length of the
two vowels. The vowel of ‘bad’ is considerably longer than the vowel of ‘bat.’ It turns out
that in English-type grammars, vowels are much longer before voiced consonants than
they are before voiceless consonants. Vowel length in the IPA is indicated by a colon-
like symbol [ː]. Thus, ‘bad’ would be transliterated [bæːd], while ‘bat’ is [bæt] (no length-
mark).8 English-type grammars thus have a rule such as that in (3.9).

(3.9) Rule 4: Lengthen vowels before voiced consonants.

The application of this rule to a stored lexical item such as /bæd/ (for which no length is
stored, since length is predictable) produces the output representation [bæːd].
We can see the combined effects of the phonological rules we have posited for English-
type grammars if we examine the phonological derivation of a word such as /konhɛd/
‘conehead.’ The stored lexical item — with all of the predictable information stripped
out — is /konhɛd/, as the phonemic bracketing indicates. Rule 1 says that vowels before
nasals must be nasalized in English-type grammars. It therefore converts /konhɛd/ to
[kõnhɛd]. Rule 2 says that consonants are rounded before round vowels, of which [õ]
is one. Rule 2 therefore turns [kõnhɛd] into [kw õnhɛd]. Rule 3 says that voiceless stops
at the beginning of stressed syllables are aspirated. It therefore converts this form into
[kwh õnhɛd]. Finally, according to Rule 4, vowels are lengthened in English-type grammars
before voiced consonants. Rule 4 thus changes our form into the actual output form:
[kwh õːnhɛːd].

3.5 Rule ordering

So far, the rules we have examined have not required any explicit ordering. For example,
it matters little to the final outcome whether we apply Rule 1, which nasalizes vowels,
before or after we apply Rule 4, which lengthens vowels before voiced consonants. If we
apply Rule 1 first, we’ll get a nasalized vowel, which, being a vowel, will lengthen by Rule
4. If we apply Rule 4 first, on the other hand, we’ll get a lengthened vowel. Rule 1 will
nasalize this lengthened vowel (since it nasalizes all vowels), and the result will be exactly
the same.
In some cases, however, explicit ordering is required in order to derive the correct
result. Examine the alternations in (3.10ab).
8
This notion of ‘length’ should not be confused with the somewhat ideosyncratic use of this term in
traditional English grammar, where the diphthong [aj] is called ‘long i’ and other such peculiarities. A long
vowel is, not surprisingly, a vowel that lasts longer than a short vowel but has no difference in quality.
3.5. Rule ordering 51

(3.10) (a) ‘fat’ [fæt] : ‘fattest’ [fæɾəst], ‘hit’ [hɪt] : ‘hitting’ [hɪɾɪŋ]
(b) ‘wide’ [wajd] : ‘widest’ [wajɾəst], ‘kid’ [kɪd] : ‘kidding’ [kɪɾɪŋ]

Since in their simple form ‘fat’ ends in a /t/ and ‘wide’ in a /d/, it seems clear that we
need a rule which converts both /t/ and /d/ to [ɾ] under certain conditions. The details
need not detain us too seriously at this point, so we will state the rule as in (3.11).

(3.11) Turn an alveolar stop into a voiced, alveolar flap ([ɾ]) when it is found between
two vowels and at the beginning of an unstressed syllable.9

In /fæt/ the /t/ is not at the start of a syllable at all (it is at the end of a stressed syllable),
as is the /d/ of /wajd/. For this reason, these segments surface as [t] and [d], respectively.
However, in ‘fattest’ and ‘widest’ the /t/ and /d/ have become the initial segment of the
second syllable, which is unstressed.10 They therefore undergo the ‘flap’ rule, taking the
form indicated in (3.10ab). This rule is pervasive in most varieties of North American
English (though it is rare in grammars of the English type found in England).
Now examine the date in (3.12ab).

(3.12) (a) [ajz] ‘eyes’, [lajz] ‘lies’, [trajd] ‘tried’, [trajb] ‘tribe’
(b) [əjs] ‘ice’, [ləjs] ‘lice’, [trəjt] ‘trite’, [trəjp] ‘tripe’

The diphthong found in the (b) examples in (3.12) is clearly different from that found in
the (a) examples. In fact, many North American varieties of English, particularly those
found in and near to Canada, have the rule in (3.13), which triggers the different realiza-
tions of /aj/.

(3.13) /aj/ becomes [əj] before voiceless consonants

This rule is responsible for the data in (3.12) as well as for alternations such as [flaj] ‘fly’
vs. [fləjt] ‘flight.’
Given the ‘aj-raising’ rule of (3.13), ‘write’ will have the underlying form /rajt/ and
‘ride’ will have the underlying form /rajd/. The fact that /rajt/ will be output as [rəjt] is
predictable (since /t/ is voiceless) and therefore will not be stored in the mental lexicon.
What happens when we add the agent suffix /r̩/ to these verbs? We get phonemic repre-
sentations that have the form /rajtr̩/ ‘writer’ and /rajdr̩/ ‘rider’ (notice that these forms
have two syllables — with stress on the first syllable — whereas the verbs were monosyl-
labic). The conditions for both the Flapping and the Raising rules are met by the form
/rajtr̩/ — the /t/ is at the start of the second, unstressed syllable, and the /aj/ is before
a voiceless consonant. What order do we apply the rules in? (3.14a) shows the result of
9
Both of these conditions must be true and the term ‘vowels’ here includes all segments that ‘act like’
vowels for purposes of syllable structure, thus syllabic liquids and nasals count as vowels.
10
Rather than go through a lengthy explanation of syllable structure here, we ask that you trust us on
the relative positions of segments within syllables.
52 Chapter 3. Phonology

applying the Flapping rule before the Raising rule, (3.14b) shows the results of applying
the rules in the opposite order.

(3.14) (a) /rajtr̩/ > (Rule 11) [rajɾr̩] > (Rule 13) [rajɾr̩] (no change, because [ɾ] is voiced
and ‘aj-raising’ applies only before voiceless consonants)
(b) /rajtr̩/ > (Rule 13) [rəjtr̩] > (Rule 11) [rəjɾr̩]

Note that the output of (3.14a) is identical to the output we get for ‘rider’: /rajdr̩/ >
(Flapping rule) [rajɾr̩]. However, ‘rider’ and ‘writer’ are not homophonous (i.e., they don’t
sound alike). ‘Writer’ has the [əj] diphthong in it, even though the flap which follows the
diphthong is voiced! Therefore, (3.14b) is the correct result. We must specify that the
Raising rule applies before the Flapping rule. As this example shows, certain phonological
rules which generate allophones must be ordered with respect to one another (Flapping
and Raising), whereas the ordering of certain other rules with respect to one another is
not significant (as we saw with nasalization and lengthening of vowels). When you are
solving a phonology problem, you must always ask yourself whether or not the rules you
have posited require ordering.
You may have noticed that the phonological rules which we have posited apply to
classes of segments. Rule 1 applies to vowels before nasals, Rule 2 involves consonants and
round vowels, Rule 3 applies to all voiceless stops, Rule 4 before voiced consonants, etc. We
do not generally find rules applying before random sets of sounds — sounds which share
nothing in common with one another. So, for example, a rule like ‘lengthen vowels be-
fore /b/, /s/, and /l/’ would be quite odd. The classes to which rules apply are generally
known as natural classes of segments — you should always try to state any observed
phonological processes in terms of the properties of the set of segments involved rather
than by simply listing the segments. Thus ‘vowels are nasalized before nasals’ is a better
scientific hypothesis than ‘vowels are nasalized before /m/, /n/, and /ŋ/’, even though
in English-type grammars the two statements produce the same results (since /m/, /n/,
and /ŋ/ are the only nasals in English). ‘Better scientific hypotheses’ have the following
properties: 1) they are the most general statements possible which make no false predic-
tions; 2) they are stated in the most economical fashion; 3) they are based on empirical
evidence; and 4) they are testable (and thus falsifiable).
Many phonological rules have the result of making one segment more like an adjacent
segment. This is true of both the nasalization and rounding cases we have just discussed.
This process is called assimilation. There are, of course other types of rules – rules which
delete segments, rules which insert segments, rules which make a segment less like an ad-
jacent segment (dissimilation), and so on. Assimilation rules are by far the most common,
however.
3.6. Some Allophony in other languages 53

3.6 Some Allophony in other languages

The principles developed on the basis of English data above are, of course, general lin-
guistic principles which can be used in the analysis of any language. All languages have
phonological rules. The specifics of these rules differ from language to language but the
general structure of the rules is the same. Because the basic structure of phonological
rules is the same in all natural language, the same principles can be used to determine
the specifics of phonological rules in particular languages. For example, examine the
data in (3.15) from Malay (spoken in Malaysia and Singapore).11

(3.15) (a) [kərəta] ‘car’

(b) [mãkan] ‘eat’
(c) [mãrah] ‘scold’
(d) [mə̃laraŋ] ‘forbid’
(e) [rumãh] ‘house’

Is nasalization predictable in Malay? [Take a moment to think about it — you might want
to look at the data to help you decide…] Yes! Malay appears to have a rule like that in
(3.16).

(3.16) Nasalize all vowels which immediately follow a nasal.

The phonemic representation of these words in the minds of Malay speakers are thus those
in (3.17). (Recall that phonemic representations have redundant, predictable information
stripped out of them.)

(3.17) (a) /kərəta/ ‘car’

(b) /makan/ ‘eat’
(c) /marah/ ‘scold’
(d) /məlaraŋ/ ‘forbid’
(e) /rumah/ ‘house’

If we apply the rule in (3.16) to the forms in (3.17) we will get the data in (3.15), as desired.
Now examine the Khmer phonetic data in (3.18).

(3.18) (a) [pɔːŋ] ‘to wish’

(b) [tɔp] ‘to support’
(c) [kat] ‘to cut’
11
As is normal in such examples, the data is presented only to reveal a principle of analysis. The actual
facts of nasalization in Malay are slightly more complicated.
54 Chapter 3. Phonology

(d) [ph ɔːŋ] ‘also’

(e) [th ɔp] ‘to be suffocated’
(f) [kh at] ‘to polish’

Is aspiration predictable in Khmer, as it is in English? [We’ll put this here to give you
time to think about it before you answer…] No! /p/ and /ph / are separate phonemes in
Khmer (even though in English they are allophones of /p/). If they were allophones of
/p/ as in English, the words ‘to wish’ and ‘also’ would both be stored in the lexicon as
/pɔːŋ/. But notice there is no way to write a rule (without mentioning the specific forms
involved, which makes the rule useless) which would cause this form to surface with as-
piration in ‘also’ but without aspiration in ‘to wish.’ The phonemic forms of these words
in Khmer must therefore contain specifications as to whether or not a given voiceless
stop is aspirated, while in English no such specification is required (because aspiration is
predictable).

3.7 Universal Properties and Language-Specific

Properties
If much of the human linguistic system is innate, i.e., attributable to Universal Gram-
mar, we would expect many properties of the system to be universally present across
languages. Indeed, all human languages have a phonological component where abstract
representations of sound are processed – this must be part of UG. On the other hand,
it is fairly clear that some part of human linguistic knowledge is acquired through ex-
posure to a particular language environment and is thus language-specific. One of the
central questions in linguistics, then, is ‘What is a possible human linguistic system?’ Note
that the answer to this question will necessarily include both universal and language-
specific properties. So while all languages have phonological rules, not all languages have
a phonological rule which causes voiceless stops to be aspirated in certain positions. The
specifics of rules can only be determined through examining the relevant data.12 There
is another aspect of phonology which falls into the language-specific realm. In Chapter
1, we noted that there were certain combinations of sounds, certain ‘words,’ that native
speakers would immediately identify as ‘not possible words.’ For English, we gave exam-
ples such as ‘ngloopi’ and ‘dnkli.’ All native speakers have intuitions of this type — what
sounds can be put together, what combinations of sounds can form a syllable, etc. — about
their native language. These language-specific properties are called the phonotactics of
the language. So ‘ngloopi’ violates a phonotactic constraint of English — English does
not allow the sequence ‘ngl’ at the beginning of the syllable. Recall, however, that simply
because it does not occur in English, does not mean that it cannot occur in some other
12
This means that you should not carry over what you know about the specifics of English phonological
rules to other languages.
3.8. Summary 55

language. It may be the case that this is a universal restriction on language (i.e., not a pos-
sible syllable-initial sequence in human language) but the only way to determine whether
it is actually the case is through examination of the relevant empirical evidence. (It turns
out that other languages do have initial clusters of that type.)

3.8 Summary
We have seen in this section that the lexicon (the mental dictionary) stores only unpre-
dictable information about the sounds in particular lexical items in a language. This re-
quires the notion of the phoneme or underlying representation — which is the name for the
segments used in our mental dictionary — and the notion allophones or surface representa-
tions — which is the name for the various phonetic realizations of those phonemes. The
various realizations themselves are generated by the application of phonological rules —
sometimes ordered, sometimes not – to the representations stored in the mental lexi-
con. Grammars differ both with respect to their phonemes and their phonological rules.
Because their phonological rules differ, different grammars may produce different allo-
phones for a given underlying segment.

3.9 Solving phonology problems

The basic question is always the same: What sorts of things can a human know with re-
spect to the sound system of natural language? We approach this question by looking at
phonetic data from a ‘language’ and then we try to figure out the unconscious system that
produces such forms.
The various kinds of detailed questions that we ask are the following. Are two or more
segments in complementary distribution (i.e. never occur in the same environment) and
are they thus allophones of a single phoneme? Or, are the segments in question separate
phonemes (listed distinctly in the phonological inventory - i.e. the underlying/’mental’
representation of sounds in a particular language)? Note that if you are told to consider
more than two segments, it may be the case that only two out of three are allophones (or,
of course, that none are).
A somewhat mechanical approach to answering this question involves the following
steps. Step (1): List all the environments. (The environments are just the ‘surroundings.’)
That is, mechanically list all segments (including word boundaries, as well as tone or syl-
lable stress if they look like they might be important) which occur to the immediate left
and to the immediate right of the segments in question. Step (2): Consider the left side.
If there is no overlap, e.g. the environment on the left is never the same for Segment A
as it is for Segment B, then they are in complementary distribution and you have solved
the problem. If there is overlap, then follow the same step for the right side. (You should
keep in mind during this process that when two items are allophones, they will share a
large number of phonetic features in common although ’trick’ problem sets, like whether
56 Chapter 3. Phonology

/h/ and /ŋ/ are allophones in English rarely occur.) If you have found nothing interest-
ing in the way of complementary distribution on either the left or right sides, proceed
to Step (3). Step (3) is to consider that the environment could include both sides — that
is, consider the fact that a combination of left and right adjacent segments is the envi-
ronment. A hypothetical example would be that Segment A is only found between two
/i’s/. Alternately, you may need to view things from a slightly broader perspective such
as whether Segment A occurs only between two vowels. Or, you may need to consider
things such as tone or syllable stress. After this examination, if you find that there is no
overlap in environment, you may conclude that the segments are allophones. Otherwise,
(if you consistently find an overlap) they are phonemes.
Writing up the solution involves formalizing your findings, to some extent. If you con-
clude that two or more segments are allophones of a single phoneme, you need to decide
which of the allophones you will choose to represent the phoneme, or underlying repre-
sentation (UR). Since we assume that language is represented economically in the minds
of speakers, we choose the segment which is found in the greatest variety of environments
as the UR/phoneme. This is the ‘elsewhere’ segment. As an illustration, we will use nasal-
ization of vowels in English which we discussed earlier. In this case, the non-nasal (oral)
vowels occur in the greatest variety of environments — everywhere except before nasals.13
Non-nasal vowels are thus the ‘elsewhere’ case and should be chosen as the phonemes. If
we were to choose nasal vowels as the phonemes, we would have to list as the environ-
ment for the rule every segment in the language apart from the three nasal consonants.
This would be both uneconomical and also miss a generalization about phonological pro-
cesses since nasalization of vowels before nasals is clearly an assimilation process.
Linguists have developed a ‘shorthand’ for writing rules which is in widespread use.
A basic understanding of the form of these statements will be a useful thing for you to
acquire. Basically, phonological rules have the form in (3.19).

(3.19) A → B / X

This formalism is to be read ‘A becomes (→) B in the environment (/) in which A follows
X. The ‘ ’ tells you where the affected segment is in the string if the rule is to apply to
it. Thus, the nasalization rule would be rewritten as in (3.20).

(3.20) vowel → nasal vowel / nasal

This means vowels are nasalized before nasals. The Malay vowel nasalization rule in
(3.16), by contrast, would be written as in (3.21).

(3.21) vowel → nasal vowel / nasal

13
This is true because oral vowels occur before stops, affricates, fricatives, word boundaries, voiced
sounds, voiceless sounds, and so on. Nasal vowels, on the other hand, only occur before nasal consonants,
a much more restricted set than the oral vowel set and one which, in addition, forms a natural class.
3.9. Solving phonology problems 57

This means that vowels are nasalized after nasals. If you had a language which nasalized
vowels only when the vowel was between two nasals, the rule would have the form in
(3.22).

(3.22) vowel → nasal vowel / nasal nasal

This rule specifies both a preceding and a following nasal being required to trigger nasal-
ization of vowels.
Note that for both the English and Malay cases, we do not list every vowel and its nasal
counterpart individually in the statement of the rule. Instead, we have made the most
general statement possible which still captures the data. If a rule applies to all vowels,
then there is no reason to be more specific than just stating ‘vowels.’ The vast majority
of rules apply to natural classes of segments and natural classes may be easily described
using only a few features, e.g., high vowels; labials; velars; stops; lax vowels; voiced con-
sonants; and so forth.
We often present derivations after we have reached a solution to some data set.
Derivations allow us to see the stepwise process that takes a phonemic or underlying rep-
resentation and converts it into a phonetic or surface representation. By going through
the steps to derive some or all of the phonetic forms in the data set, we are able to double
check to see that our solution properly accounts for all the data. A derivation consists of:

• the underlying form (this should always include the segment that is in the speaker’s
phonological inventory, i.e. the phoneme or UR and should appear in slashes);

• the rule/rules (Note: if there are more than one rule, they may have to be ordered
with respect to one another — recall the English Flapping and Raising rules.)

• the surface form (in square, phonetic brackets).

If the word [bĩn] is given in the data for nasalized vowels in English, a derivation of [bĩn]
would look like this:

/bin/

nasalization rule

[bĩn]

Many phonology processes are sensitive to syllable structure (cf., the aspiration rule
of English). There are some general rules of thumb about figuring out where one syllable
begins and where another ends (and they have nothing to do with the spelling and dictio-
nary conventions for English). Briefly, the number of syllables is equal to the number of
58 Chapter 3. Phonology

vowels (including, of course, diphthongs and syllabic/vocalic consonants). If the struc-

ture contains alternating consonants and vowels, then the syllable breaks are after each
vowel, as indicated by the dots (CV.CV.CV.CV). If the vowels are separated by two conso-
nants, the syllable break should be placed between the two consonants (CVC.CVC.CVC). If
there are three or more consonants, you’re on your own... (but no data that we give you
will require you to determine the syllable structure of such strings).
Finally, you should keep in mind that the presence or absence of particular phonolog-
ical rules is very language-specific (since it is based on experience with particular lan-
guage data during acquisition). Therefore, it is important to look at each new data set
as: 1) completely independent of any other data set; and 2) entirely representative of all
the data you need to determine an answer to the question. So the fact the English has
an aspiration rule has no bearing on, for example, the solutions to any of the phonology
problems at the end of this chapter.
3.10. Exercises 59

3.10 Exercises
The correct choice of phonological is a rule stated in the most general terms possible
that still produce the correct data.

1. Mokilese (an Oceanic language spoken on Mokil). Examine the following data, not-
ing the distribution of voiceless (with a ring under the vowel) and voiced vowels.

a) [pi ̥san] ‘full of leaves’

b) [uduk] ‘flesh’
c) [kaskas] ‘to throw’
d) [poki] ‘to strike something’
e) [tu̥pu̥kta] ‘bought’
f) [pu̥ko] ‘basket’
g) [ki ̥sa] ‘we two’
h) [pil] ‘water’
i) [apid] ‘outrigger support’
j) [luʤuk] ‘to tackle’
k) [su̥pw o] ‘firewood’
l) [kamw ɔki ̥ti] ‘to move’
m) [isi ̥k] ‘to shake’
n) [tun] ‘to drip’

a. Which rule accounts for the voiceless vowel allophones in the data above?

(a) Vowels become voiceless before voiceless stops.

(b) High vowels become voiceless between voiceless consonants.
(c) High vowels become voiceless after voiceless consonants.
(d) Vowels become voiceless between voiceless consonants.

b. What should the phonetic realization of Mokilese /pisa/ ‘fish body part (from
the start of the tail to just below the gills)’ be?

(a) [pisḁ]
(b) [pisa]
(c) [pi ̥sa]
60 Chapter 3. Phonology

(d) [pi ̥sḁ]

c. The phonemic representation of Mokilese ‘bought’ is:

(a) /tu̥pu̥ktḁ/
(b) /tupuktḁ/
(c) /tu̥pu̥kta/
(d) /tupukta/

2. Rennellese. Examine the data below from the Rennellese language, paying partic-
ular attention to the distribution of [t] and [ʧ].

a) [haʧi] ‘break’
b) [hakaʧeːʧeː] ‘stroll’
c) [tasi] ‘one’
d) [ʧege] ‘run’
e) [tuː] ‘ground pigeon’
f) [ʔata] ‘picture’
g) [huʧi] ‘banana’
h) [taŋata] ‘man’
i) [hakatoːtonuŋaohie] ‘hard to explain’
j) [tou taʔe nei] ‘your excrement here [1sg. humilitive pronoun]’

a. What rule would account for the distribution of [t] and [ʧ] in Rennellese?

(a) /t/ becomes [ʧ] before front vowels.

(b) /ʧ/ becomes [t] before back and low vowels.
(c) /t/ becomes [ʧ] before back and low vowels.
(d) /ʧ/ becomes [t] before front vowels.

b. You chose to state the rule that way based on

(a) counting the number of occurrences of each sound and picking the one that
occurred most often.
(b) determining that one of the sounds occurred in a wider variety of environ-
ments than the other sound and should therefore be considered the phoneme/UR
3.10. Exercises 61

(c) counting the sounds that preceded [t] and [ʧ] and picking the one that had the
fewest different sounds preceding it
(d) the feeling that one of the two sounds was more likely to be the phoneme than
the other.

c. The phonemic representation of Rennelese ‘run’ is:

(a) /ʧege/
(b) /tege/
(c) either /ʧege/ or /tege/
(d) /run/

3. West Tarangan (Maluku, Indonesia). Examine the data below from the West Tarangan
language. The voiceless, bilabial fricative [ɸ] is like an [f], but without the involve-
ment of the teeth. Consider [ɸ] and [p].

a) [ɸit] ‘night’
b) [ɸɔp] ‘pig’
c) [ɸaɸə] ‘ground’
d) [ɛtaɸakə] ‘it capsizes’
e) [ɔlɸɔt] ‘egret’
f) [gurep] ‘nail’
g) [nakɸɔ] ‘I will carry’
h) [tɔp] ‘short’

a. Which of the following statements accounts for the distribution of these allo-
phones?

(a) You can’t tell when [ɸ] will appear and when [p] will appear.
(b) /ɸ/ → [p] at the ends of words, [ɸ] elsewhere.
(c) /ɸ/ → [p] after vowels, [ɸ] elsewhere.
(d) /ɸ/ → [ɸ] between vowels, [p] elsewhere.

b. The phonemic representation of West Tarangan ‘pig’ is:

(a) /ɸɔp/
(b) /pɔp/
62 Chapter 3. Phonology

(c) /pɔɸ/
(d) /ɸɔɸ/

4. Sawai (Maluku, Indonesia). Consider Sawai [d] and [ɾ].

a) [ndɛɾɛɾɛ] ‘he follows’

b) [mɛɾmɛɾ] ‘a plant’
c) [mdiɾkɛ] ‘you (sg.) move’
d) [deko] ‘friend’
e) [mdi] ‘tree trunk’
f) [dɛɾe] ‘wall’
g) [wɔɾɛ] ‘already’
h) [dɛl] ‘girlfriend’
i) [ndɛ] ‘he goes’
j) [nɛɾoɾ] ‘he asks’

a. Which of the following statements accounts for the distribution of these allo-
phones.

(a) /d/ → [ɾ] between vowels, [d] elsewhere.

(b) /d/ → [ɾ] when not word-initial, [d] elsewhere.
(c) /d/ → [ɾ] after vowels, [d] elsewhere.
(d) /d/ → [ɾ] before vowels, [d] elsewhere.

b. What is the phonemic representation of Sawai ‘he follows’?

(a) /ndɛdɛɾɛ/
(b) /nɾɛɾɛɾɛ/
(c) /ndɛɾɛdɛ/
(d) /ndɛdɛdɛ/
 Chapter 4
Morphology

Contents
4.1 Morphemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Morphemes and Allomorphs . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Productivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4 Hierarchical Structure in Morphology . . . . . . . . . . . . . . . . . . . 75
4.5 The ‘cranberry’ problem . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

4.1 Morphemes
Until now, we have been concerned only with the sound system of language. We have
paid little attention to things that seem like they might be somewhat important such as
‘meaning.’ While we have learned that English /kæts/ will have the form [kh æts] when
the grammar is finished processing it, we have done nothing to account for the fact that
speakers with English-type grammars will instantly recognize that /kæts/ is a ‘complex’
word, in the sense that it contains more than one meaningful element.1 The study of the
internal structure of complex words of this type is known, in linguistics, as morphology.
The word /kæts/, to be specific, appears to consist of two components: /kæt/, which
means something like ‘cat’, and /s/, which means something like ‘more than one’. Put
together, as here, they therefore mean ‘more than one cat.’ Notice that although /kæt/
consists itself of three phonological segments, it cannot be further divided into mean-
ingful component elements (so, e.g., it is not the case that /k/ means ‘four-pawed,’ /æ/
1
We are using slanted, ‘phonemic’ brackets here for the moment but the representation within the
brackets is not, as you will see, the underlying form of the morpheme ultimately.

63
64 Chapter 4. Morphology

‘pet’ and /t/ ‘coughs up furballs’). /kaets/ has exactly two meaningful elements within
it.2 Note, moreover, that it is not the case that any old combination of /kæt/ and /s/ gives
rise to the meaning ‘more than one cat’: neither /skaet/, /ksæt/, nor /kæst/, all of which
consist of a combination of /s/ and /kæt/, mean ‘more than one cat.’ Only /kæts/ means
‘more than one cat.’ Why should this be so?
It is also worthwhile to observe that if you and a friend were walking along and saw
what appeared to be a very large and oddly colored hairpiece spread out on the hood of
a car, and your friend said ‘What’s that?’ you, with your masterful detective skills, might
be able to answer ‘/kæt/!’ However, if your friend asked, ‘How many cats are there?’
you could not answer ‘/s/!’ to mean ‘more than one!’ That is, /kæt/ appears to mean
‘cat’ all on its own, while /s/ means ‘more than one’ only when attached to some noun.
In particular, as we saw in the paragraph above, it must be attached to the end of some
noun. Thus /kæts/ means ‘more than one cat,’ /hæts/ means ‘more than one hat,’ and
/læps/ means ‘more than one lap.’
A string of one or more phonemes which conveys a particular meaning is known as
a morpheme. Thus /kæt/ ‘cat’ is a morpheme, as is /s/ ‘more than one.’ [Note that
this does not mean that every sequence /kæt/ in English contains the morpheme /kæt/
– /skæt/, e.g., does not – any more than every /s/ in English means ‘more than one.’]
Moreover, /kæt/ is what is generally referred to as a free morpheme, meaning it can be
used on its own to convey the meaning ‘cat’ (as in the example in the last paragraph),
while the plural marker /s/ by contrast is known as a bound morpheme, since it must
be ‘bound’ to another morpheme for it to convey its meaning. A bound morpheme is
also frequently called an affix. Morphemes are the smallest units which carry meaning.
Phonemes, on the other hand, are the smallest units which signal a difference in meaning
but are not meaningful in and of themselves (they are just sounds).
Bound morphemes come in various types, depending upon where they attach to their
host morpheme. Affixes which attach at the end of their host, like the English plural
marker /s/ in /kæts/, are known as suffixes. English also has prefixes, which attach
to the beginning of their host word. For example, /ri-/ ‘again’ as in /ritaj/ ‘retie’ (i.e.,
‘tie again’), or /ən-/ ‘not’ as in /ənbəlivəbl ̩/ ‘unbelievable’ (i.e., ‘not able to be believed’).
Note that /ənbəlivəbl ̩/ has both a prefix (/ən/) and a suffix (/-əbl ̩/), added to the ‘free
morpheme’ /bəliv/ ‘believe’.3
Extremely rare in English-type grammars, indeed for some speakers non-existent, is
the third logical possibility for the placement of an affix relative to its host morpheme:
infixation. Infixes are placed somewhere inside their host morpheme. Since infixes are
rare to non-existent in English, we will cite a few examples from other languages. Exam-
ine the singular : plural pairs in Oaxaca Chontal (Mexico) in the data below.
2
To be clear, the ‘meaningful element’ /æt/ ‘at’ is phonologically contained within /kæts/—but there is
no /æt/ element contributing to the overall semantics of /kæts/. Therefore, to be accurate, it is not ‘at’
contained within /kæts/, but rather a sequence of segments which sound like ‘at’.
3
We use a hyphen to indicate where the affix attaches to its host. In the case of an infix—a type of affix
we are about to introduce in the main text—there would be hyphens on both sides.
4.1. Morphemes 65

(4.1)
Oaxaca Chontal Plurals
singular plural
/tuwa/ ‘foreigner’ /tulwa/ ‘foreigners’
/lipo/ ‘possum’ /lilpo/ ‘possums’
/sewiʔ/ ‘magpie’ /selwiʔ/ ‘magpies’
/mekoʔ/ ‘spoon’ /melkoʔ/ ‘spoons’

How does one make plurals in Oaxaca Chontal? Unlike other affixes, whose placement is
clear by virtue of their names, one must always specify precisely where to infix an infix.
For example, if all you stated was that Oaxaca Chontal made its plurals by infixing /l/
‘more than one’ and that /sewiʔ/ was the word for ‘magpie,’ we would not know whether
the plural was /slewiʔ/, /selwiʔ/, /sewliʔ/, or /sewilʔ/! An infix is not randomly inserted
anywhere within a morpheme, its placement is rule-governed. You must determine, after
examining the data, exactly what the rule for placement of the infix is. (For example, if
you were to say ‘after the first vowel’, this would work for Oaxaca Chontal).
Here’s some more data on infixation, just to make it clear to you. This comes from
Agta (a Phillipine language).

(4.2)
Agta Tense Formation
present past
/gafutan/ ‘grabs’ /ginafut/ ‘grabbed’
/daŋagan/ ‘hears’ /dinaŋag/ ‘heard’
/hulutan/ ‘follows’ /hinulut/ ‘followed’
/paligatan/ ‘hits’ /pinaligat/ ‘hit (past)’

Notice first that Agta has a suffix which indicates ‘present tense’: /an/.4 So /gafutan/
‘grabs’ is a complex word, having two parts: /gafut/ ‘grab’ (the root) and the suffix /an/
‘present tense.’ The past tense form of this verb is created by taking the root which means
‘grab’ (/gafut/) and infixing into it the past tense morpheme /in/. Where is the past tense
morpheme infixed in Agta? That’s right, after the first consonant of the host morpheme
(the verb).
It is critical that you realize that morphological surface forms are not derived from
one another. That is, if you want to make the past tense of ‘hit’ in Agta, you do not first
make the present /paligatan/, then subtract the present tense suffix /an/, then insert
the past tense infix /in/. Why bother making the present tense through suffixation in
this way when what you want to say is the past tense? This would be comparable in
English to saying that, in order to make the past tense of the verb /glajd/, you take the
4
As in phonology, the abstract mental representation which serves as input to the grammar is placed
between slashes; thus, both phonemic and morphemic representations are cited with slashes.
66 Chapter 4. Morphology

3rd singular present form /glajdz/, subtract the 3rd singular marker /-z/ and add the
past tense marker /-əd/. This seems pretty silly when stated in terms of English to you,
as an English speaker, and it would seem just as silly if proposed to an Agta speaker for
Agta. To say the past tense of ‘hit’, you take the morpheme that means ‘hit’ (/paligat/)
and infix into it the past tense marker (/in/), giving /pinaligat/. There is no reason to
mess around with the present tense form.
We will offer one more example. This one comes from Katu, a Katuic language of
Vietnam.

(4.3)
Katu Infixation
simple form infixed form
/gap/ ‘cut’ /ganap/ ‘scissors’
/juut/ ‘rub’ /januut/ ‘cloth’
/panh/ ‘shoot’ /pananh/ ‘crossbow’
/piih/ ‘sweep’ /paniih/ ‘broom’

The infix here is clearly /an/, but what does it mean? Well, ‘scissors’ are what you use
to ‘cut’ with, ‘cloth’ is what you use to ‘rub’ with, a ‘crossbow’ is what you ‘shoot’ with
(if you are a Katu, — we will refrain from the obvious corollary), and a ‘broom’ is what
you ‘sweep’ with. The nouns identify the instrument used to perform the action of the
verb. They are therefore called ‘instrument nouns’ and the infix /an/ is used to derive
instrument nouns from verbs. As in the Agta case above, this infix is placed after the
initial consonant of the verb which hosts the infix.
There is one more kind of morpheme. This is usually referred to as a ‘process mor-
pheme’ since, rather than consisting of some string of phonemes, it requires, instead,
modification of the morpheme to which it attaches. We will consider only one type of
process morpheme here: reduplication. Examine the data from Northern Tepehuán be-
low.

(4.4)
Northern Tepehuán Reduplication
singular plural
/toʃi/ ‘rabbit’ /totoʃi/ ‘rabbits’
/kəli/ ‘man’ /kəkəli/ ‘men’
/mara/ ‘son’ /mamara/ ‘sons’
/tova/ ‘turkey’ /totova/ ‘turkeys’
/ʃiəgi/ ‘older brother’ /ʃiʃiəgi/ ‘older brothers’
/sukuli/ ‘younger brother’ /susukuli/ ‘younger brothers’
/dəgi/ ‘rat’ /dədəgi/ ‘rats’
4.1. Morphemes 67

Can you figure out how Northern Tepehuán forms the plural of these words? Of course
you can. One takes the initial consonant and vowel of the base morpheme and prefixes
them (in their original order) to that base. This process, doubling some of the material al-
ready present in the morpheme to which the process applies, is known as ‘reduplication.’
The meaning of reduplication in these Northern Tepehuán words is ‘more than one.’
Below is another case of reduplication. (It’s a popular process …) This one comes from
our old friend, Agta.

(4.5)
Agta Reduplication
simple form reduplicated form
/adanuk/ ‘long’ /adadanuk/ ‘very long’
/addu/ ‘many’ /adaddu/ ‘very many’
/apisi/ ‘small’ /apapisi/ ‘very small’
/abikan/ ‘near’ /ababikan/ ‘very near’

Here, the initial vowel plus consonant sequence of these words is being reduplicated,
the doubled segments being prefixed in their original order to the base morpheme. The
meaning is intensification (i.e., an extreme version of whatever the base morpheme means).
It is necessary, at this point, to add a couple of cautionary statements, the first about
how one should regard data generally and the second about bound morphemes (in par-
ticular, infixes). We will use the Southern Barasano below to illustrate our points.5

5
A vowel with a tilde over it is nasalized, as was stated earlier.
68 Chapter 4. Morphology

(4.6)
Southern Barasano
complex word form gloss
1. /wabõ/ ‘she goes’
2. /bɪ ̃dĩbĩ/ ‘he goes upstream’
3. /tɪdibɪ/ ‘I returned’
or ‘you return’
4. /bɪ ̃dĩkoabĩ/ ‘he really goes upstream’
5. /tɪdiboabɪ/ ‘I unexpectedly return’
or ‘you unexpectedly return’
6. /wakoabɪ/ ‘I really go’
or ‘you really go’
7. /tɪdirũtũbĩ/ ‘he continues to return’
8. /bɪ ̃dĩboabõ/ ‘she unexpectedly goes upstream’
9. /warũtũbĩ/ ‘he continues going’
10. /tɪdikoabõ/ ‘she really returns’
11. /bɪ ̃dĩrũtũbɪ/ ‘I continue to go upstream’
or ‘you continue to go upstream’
12. /waboabõ/ ‘she unexpectedly goes’

This appears rather complicated. The first thing to realize regarding data in general (in
case it is not already obvious) is that the morphological entities found in one language
do not necessarily, or even usually, correspond in a nice one-to-one way to those found
in another language.6 Thus, while intensification is done through reduplication in Agta,
in English-type grammars we usually use a free morpheme (‘very’) to express the same
meaning. So, although we find all kinds of things in the English translation of the South-
ern Barasano words above, it is clear that all of the meanings contained within the En-
glish – including those many meanings which are expressed by free morphemes (words)
in the English – are expressed by various types of affixes in Southern Barasano. To solve
such a problem, it is necessary to identify which parts of the Southern Barasano express
which meanings. The meanings will not necessary match each individual word of the En-
glish – sometimes English uses multiple words to express what is expressed in Southern
Barasano by a single affix.
Let us first identify the Southern Barasano rendering of the pronouns of the English.
These include ‘he’, ‘she’ and, rather weirdly (given the languages of the world), some-
thing that apparently means both ‘I’ and ‘you’ (since every sentence involving ‘I’ in the
data above can also be translated ‘you’, and vice-versa). If you look at all of the sentences
involving ‘he’ in the English translation, you will see that they each end in /bĩ/. Appar-
ently the suffix /bĩ/ means ‘he’. If you look at those sentences which involve ‘I’ and ‘you’,
you will see similarly that the suffix /bɪ/ occurs in all Southern Barasano words which
6
This is just as true in phonology and syntax as it is in morphology.
4.1. Morphemes 69

are translated with ‘I’ and ‘you’ in English. How about ‘she’? Well, it is a little hard to
tell without further analysis whether ‘she’ is /bõ/ or /abõ/ – since all ‘she’ sentences end
with the same last three segments in Southern Barasano. We will keep the options open
for ‘she’ for the moment and pass on to identify what in the Southern Barasano corre-
sponds to English ‘really’. If you look at the relevant examples, you will quickly see that it
is /koa/. So ‘I/you really go’ in (6) appears to consist of /wa/ ‘go’ plus /koa/ ‘really’ plus
/bɪ/ ‘I/you’ (thus meaning ‘I/you really go’).7 That means that ‘she goes’ in (1) is probably
/wa/ ‘go’ plus /bõ/ ‘she.’ Now we have gotten ‘she’. How does Southern Barasano express
the fact that the action indicated by the verb happened unexpectedly? Clearly, /boa/ is
responsible for this part of the meaning of the Southern Barasano words above. And how
does Southern Barasano express that the action indicated by the verb is continuing to go
on? – with /rũtũ/. This leaves the verbs (which appear to be morphemes from the data)
/wa/ ‘go’, /bɪ ̃dĩ/ ‘go upstream’,8 and /tɪdi/ ‘return’.
Now comes the second caution. Given that the analysis of (9) is /wa/ ‘go’ plus /rũtũ/
‘continue’ plus /bĩ/ ‘he’ (therefore meaning ‘he continues going’), is /rũtũ/ an infix? The
answer is a resounding NO. An infix is inserted into a morpheme. /rũtũ/ is not inside
some other morpheme – it’s merely been suffixed to the verb /wa/ before /bĩ/ was suf-
fixed. That is, we must distinguish between a series of suffixes (or prefixes) and a true
infix. To be an infix, you must be inside another morpheme, not merely between two other
morphemes. Do not confuse infixes with suffixes or prefixes. In addition, when we find
multiple suffixes, as here, we must state, when we give a solution to the problem, what
order the suffixes attach in. (This obviously applies to prefixes, as well.)
The simplest type of morphological problem consists merely of segmenting complex
words into their component parts. Just for practice, we will give three instances of this
type of problem here. The answer will be provided after the presentation of the data for
all three, to encourage you to try to solve these problems before looking at the answers.
The task, in case it is not clear, is to identify the morphemes, state whether they are
bound or free, and, if bound, whether they are prefixes, suffixes, infixes, or reduplication
(a process morpheme). If there are multiple prefixes or suffixes, state the order.

7
There appears to be nothing here indicating ‘present tense’ – this is not unusual in the languages of
the world.
8
Note that in Southern Barasano, unlike in English, ‘go upstream’ appears to be a unitary morpheme –
i.e., it does not contain ‘go’. Again, this kind of mismatch between languages is common.
70 Chapter 4. Morphology

(4.7)
Michoacán Nahuatl
/nokali/ ‘my house’
/nokalimes/ ‘my houses’
/mokali/ ‘your house’
/ikali/ ‘his house’
/kali/ ‘house’
/nopelo/ ‘my dog’
/mopelo/ ‘your dog’
/mopelomes/ ‘your dogs’
/pelo/ ‘dog’
/nokwahmili/ ‘my cornfield’
/ikwahmilimes/ ‘his cornfields’
/kwahmili/ ‘cornfield’

(4.8)
Isthmus Zapotec
/ʒigi/ ‘chin’
/kaʒigi/ ‘chins’
/ʒigibe/ ‘his chin’
/ʒigiluʔ/ ‘your (sg.) chin’
/kaʒigitu/ ‘your (pl.) chins’
/diaga/ ‘ear’
/kadiagatu/ ‘your (pl.) ears’
/kadiagadu/ ‘our ears’
/biʃozedu/ ‘our father’
/biʃozetu/ ‘your (pl.) father’
/kabiʃozetu/ ‘your (pl.) fathers’

(4.9)
Pocomchí
simple form derived form
/suk/ ‘good’ /suksuk/ ‘delicious’
/raʃ/ ‘green’ /raʃraʃ/ ‘very green’
/nim/ ‘black’ /nimnim/ ‘jet black’
/kaq/ ‘red’ /kaqkaq/ ‘very red’
/saq/ ‘white’ /saqsaq/ ‘very white’
Pocomchí /q/ is a voiceless, uvular stop. Can you find your uvula?

• The Answer to the Michoacán Nahuatl Problem

4.2. Morphemes and Allomorphs 71

– There are three free morphemes in the data: /kali/ ‘house’; /pelo/ ‘dog’; and
/kwahmili/ ‘cornfield’.

– There is one suffix in the data: /mes/ ‘plural’.

– There are three prefixes in the data: /no/ ‘my’; /mo/ ‘your’; and /i/ ‘his’.

• The Answer to the Isthmus Zapotec Problem

– There are three free morphemes in the data: /ʒigi/ ‘chin’; /diaga/ ‘ear’; /biʃoze/
‘father’9
– There are four suffixes in the data: /be/ ‘his’; /luʔ/ ‘your (sg.)’;10 /tu/ ‘your
(pl.)’; and /du/ ‘our’.
– There is one prefix in the data: /ka/ ‘plural’.

• The Answer to the Pocomchí Problem

– The free morphemes are all those listed as ‘simple forms’. The other mor-
pheme is complete (or ‘full’) reduplication, which indicates intensification. As
we have now seen, reduplication may be partial (cf. Agta) or full as in Pocom-
chí.

4.2 Morphemes and Allomorphs

It would be nice if morphological life were as simple as what we have seen so far – unfor-
tunately, it is not. The English plural forms below illustrate some further complexities.

9
Note that although ‘father’ is not found standing on its own in the data given, the most general solution
to the problem — and thus the most valuable — is that which can see beyond the limited data given. Since
all of the other nouns which show up with the various prefixes and suffixes present in the data can also
stand on their own as ‘free morphemes,’ it is logical to conclude that the same is true of ‘father’.
10
Isthmus Zapotec, like many human languages, distinguishes between a singular ‘you’ and a plural one.
English used to do this with ‘thou’ vs. ‘ye’.
72 Chapter 4. Morphology

(4.10) English Plurals

singular plural
/kæt/ ‘cat’ /kæts/ ‘cats’
/rɪp/ ‘rip’ /rɪps/ ‘rips’
/bæk/ ‘back’ /bæks/ ‘backs’
/klɪf/ ‘cliff’ /klɪfs/ ‘cliffs’
/kæd/ ‘cad’ /kædz/ ‘cads’
/rɪb/ ‘rib’ /rɪbz/ ‘ribs’
/bæg/ ‘bag’ /bægz/ ‘bags’
/hajv/ ‘hive’ /hajvz/ ‘hives’
/bənænə/ ‘banana’ /bənænəz/ ‘bananas’

From this list, you can see quite clearly that whereas some nouns make their plural by
suffixing /s/, others do so by suffixing /z/. While it is possible that we all just memorize
which nouns take /s/ and which take /z/, there is considerable evidence that this is not
correct. In order to see why, examine the figure below.

Suppose we tell you that this is a picture of the dreaded /birdɨd slorg/. We can safely
assume that you have never heard of a /birdɨd slorg/ before since we just created it afresh.
If we were to ask you, now, what the following is a picture of:

you would almost certainly give the answer that this is a picture of two /birdɨd slorgz/.
Psycholinguistic experiments have shown that this is precisely what speakers with English-
type grammars answer in such a testing situation. The interesting question is this: why
do all people with English-type grammars answer this question in the same way? You
cannot have known what the plural of /slorg/ was before you were asked, since you had
never heard the word before, and we did not tell you the plural.
Notice that if we now told you that we had made an error in judgement and that in fact
the first picture above was not a /birdɨd slorg/ at all, but rather the benign /birdɨd zorp/,
and extracted from you what the second picture contained, you would say it showed
4.2. Morphemes and Allomorphs 73

‘two /birdɨd zorps/.’ Again, this has been conclusively demonstrated in experiments per-
formed on willing subjects.
If you examine the list of English plurals given in the table in (4.10) above, the reason
for this otherwise inexplicable behavior should be clear. English regular plurals (i.e., ig-
noring things like ‘sheep’ : ‘sheep’, ‘ox’ : ‘oxen’, ‘goose’ : ‘geese’ – to which we will turn our
attention momentarily) appear to add /s/ after voiceless segments, but /z/ after voiced
ones. Now consider the data below.

(4.11)
More English Plurals
singular plural
/mɛs/ ‘mess’ /mɛsɨz/ ‘messes’
/mez/ ‘maze’ /mezɨz/ ‘mazes’
/ræʃ/ ‘rash’ /ræʃɨz/ ‘rashes’
/gəraʒ/ ‘garage’ /gəraʒɨz/ ‘garages’
/mæʧ/ ‘match’ /mæʧɨz/ ‘matches’
/bæʤ/ ‘badge’ /bæʤɨz/ ‘badges’

These examples show that if a noun ends, in English, in an alveolar or alveo-palatal frica-
tive or an affricate, it takes the plural marker /ɨz/ (regardless of the voiced or voiceless
status of its final segment).
For regular plurals in English, the precise form of the plural marker (/s/, /z/, or /ɨz/)
is fully predictable. In keeping with our general practice of not admitting predictable in-
formation into the lexicon, we therefore want to store only one of these forms. The others
can be derived by a rule which produces the variant realizations of this one underlying
morpheme.11 The variant realizations of a morpheme are called allomorphs, much as the
variant realization of a phoneme are known as allophones. The regular plural marker
in English therefore has three allomorphs: /s/, /z/, and /ɨz/. The rules governing these
forms are stated informally below.

• Rules for English Regular Plural Formation (informal statement, to be revised shortly)

1. If the noun ends in an alveolar fricative or an affricate, add /ɨz/.

2. If the noun ends in a voiceless consonant which does not fall under Rule 1, add
/s/.
3. If the noun ends in a voiced consonant which does not fall under Rule 1, add
/z/.
11
In addition, we should note that speakers, if asked, appear to be unaware that they produce three
different forms when they produce plurals. This lends further support for the fact that only one of the
forms needs to be stored (explicitly memorized) and the others are produced automatically by rule.
74 Chapter 4. Morphology

When we were dealing with allophones above, it became clear that when we had, for
example, two allophones of a given phoneme ([p] and [ph ] as allophones of the phoneme
/p/, for instance) we needed one rule to produce them. This is because if we posit a rule
which aspirates voiceless stops (thus producing, under the proper conditions, [ph ] from
the phoneme /p/) the other allophone ([p]) would surface by default if we simply applied
no rule at all to it. The same is true for allomorphs. We need to pick one of the regular
plural allomorphs (/s/, /z/, or /ɨz/) as the underlying morpheme, then by positing two
rules, we can produce two allomorphs. The third allomorph will be the result of doing
nothing at all to the underlying morpheme, and thus does not require a rule at all.
Now imagine that we were to select /ɨz/ as the underlying form of the regular plural
morpheme in English. We would need (1) a rule which deleted the schwa (/ɨ/) except
after an alveolar or alveo-palatal fricative or affricate, (2) a rule which forced the /z/ that
resulted from the first rule to assimilate (i.e., adopt the same value for some feature, in
this case voicing) to the voicelessness of a final voiceless consonant, and (3) a statement
that the first rule must be applied before the second.
If we assumed /s/ was the underlying form for the regular plural morpheme, we would
need (1) a rule which inserted schwa before this /s/ if the noun to which the plural marker
was attached ended in an alveolar or alveo-palatal fricative or affricate, (2) a rule that as-
similated /s/ to the voicing of the segment immediately preceding it, and (3) a statement
that the first rule applies before the second (since the forms affected by the first rule
need to come out /ɨz/, not /ɨs/). Note that the second rule required under this assump-
tion cannot be a general phonological rule of English, since the final ‘s’ of /fɔls/ ‘false’ does
not undergo it (i.e., /fɔls/ is not pronounced [fɔlz]).12
Finally, if we assumed /z/ was the underlying form for the regular plural morpheme,
we would need (1) a rule which inserted schwa before this /z/ if the noun to which the
plural marker was attached ended in an alveolar fricative or an affricate, (2) a rule that
assimilated /z/ to the voicelessness of the segment immediately preceding it, and (3) a
statement that the first rule must apply before the second.
The second of these options (in which the regular plural morpheme is /s/) is the least
satisfactory for the following reason. It is more economical to have a system with fewer
exceptions than one with more exceptions. Because phonological rules operate without
exception, a solution which relies on phonological rules is preferred. It is not always
possible to have a purely phonological solution to a morphology problem, but when it
is possible, that option should be chosen. One of the rules which choosing /s/ as the
morpheme requires does not appear to be generally true of English, given the existence
of words such as /fɔls/, thus, you cannot have a phonological solution, if you select this
particular underlying form. This problem does not plague either other solution. So, how
do we pick between the two remaining options? Is it /z/ or /ɨz/? The key here is in the
way in which the first rule is stated for each of these options. In the case of the /ɨz/
choice, the rule says ‘delete schwa except after an alveolar or alveo-palatal fricative or
12
One characteristic of phonological rules, as opposed to morphological rules, is that they operate
‘across the board’ – there are no exceptions to phonological rules.
4.2. Morphemes and Allomorphs 75

affricate’. In the case of the /z/ choice, the rule says ‘insert a schwa after an alveolar or
alveo-palatal fricative or affricate’. Casually termed, for the first choice, /ɨz/, we need
a rule which says ‘don’t delete schwa here’ (where ‘here’ is after an alveolar fricative or
affricate). For the second choice /z/, we need a rule that says ‘do insert schwa here’ (same
‘here’ as above). It is a general principle of grammar, for reasons of economy such as those
stated earlier, that rules are stated in positive terms with well-defined environments.
Therefore, ‘do insert schwa here’ (where ‘here’ is the well-defined environment of ‘after
alveolar fricatives and affricates’) is the preferred choice. The final choice, then, for the
regular plural morpheme in English-type grammars is /z/.

• Regular Plural Allomorph Creation Rule (informal, final version)

1. Insert a schwa if the plural morpheme follows an alveolar or alveo-palatal

fricative or affricate.
2. z → s after voiceless consonants

The rules apply in the order stated. A similar rule accounts for the regular third person
singular present tense verb forms in /z/ (with allomorphs /z/, /s/, and /ɨz/ as in /goz/
‘goes’, /kɪks/ ‘kicks’ and /wɪʃɨz/ ‘wishes’) and the possessive.
The English plural is a fairly complex case so let us look at something simpler. Examine
the Sierra Popoluca data below.13

(4.12) Sierra Popluca Possessed Forms

singular plural
1. tək ‘house’ antək ‘my house’
2. petkuj ‘broom’ ampetkuj ‘my broom’
3. meːme ‘butterfly’ ammeːme ‘my butterfly’
4. heːpe ‘cup’ anheːpe ‘my cup’
5. kawah ‘horse’ aŋkawah ‘my horse’
6. nəːyi ‘name’ annəːyi ‘my name’
7. suuŋ ‘cooking pot’ ansuuŋ ‘my cooking pot’

What kind of affix expresses possession by the speaker in Sierra Popoluca? A prefix. What
are its allomorphs? Well, the affix shows up in the following forms in the data above: /an/
(1,4,6,7), /am/ (2,3), and /aŋ/ (5). ‘Is the distribution predictable?’ is obviously the next
question. (If you simply keep on reading to find out the answer, you will never learn to
solve problems of this type on your own. Take a minute to look at the actual data.) Yes,
the distribution is predictable. /am/ occurs before /p/ and /m/, /aŋ/ occurs before /k/,
13
We will omit brackets of any type from now on unless they are relevant. It should be reasonably clear
what level, phonetic or phonemic, we are dealing with at any particular time.
76 Chapter 4. Morphology

and /an/ occurs before /t/, /h/, /n/, and /s/. Is there a reason /am/ might occur before
/p/ and /m/? Yes, indeed. /p/ and /m/ are both labial and the nasal at the end of the
allomorph of ‘my’ that occurs before words that start with these consonants is labial, too.
Does it make sense that we get the allomorph /aŋ/ before /k/? Yes it does. /k/ is a velar,
and /aŋ/ is the variant which has a velar nasal. Does is make sense that we get /an/ before
/t/, /h/, /n/, and /s/. Well, partly. /t/, /n/, and /s/ are all alveolars, as is the /n/ at the
end of this allomorph for ‘my’, but what about /h/. Is /h/ alveolar? Of course not. So
when we add ‘my’ to a noun in Sierra Popoluca the final nasal of that morpheme always
matches the initial consonant of the noun in place of articulation, except in the case of
/h/. What that tells us that the form before /h/ must be the default form of the morpheme
– i.e., the form that surfaces when nothing special happens to it. So we should assume
that /an/ is the underlying form of ‘my’ and that it assimilates in place of articulation
to following labial and velar consonants. Thus underlying /an/ will become /am/ before
labials and /aŋ/ before velars. It won’t do anything before /s/ or /h/ — and thus will
come out unchanged (i.e., as its underlying form, /an/). This is clearly the correct result.

4.3 Productivity
The case of the English plural, which we discussed above, illustrates something else quite
important about morphology. Depending upon one’s theory of morphology, one might
say that there are two kinds of morphemes: productive and unproductive. A productive
morpheme can be freely used with ‘new’ lexical items, such as the word ‘slorg’ which we
invented in the previous section. Any native speaker of English can generate the plural
of a noun such as ‘slorg’ (e.g., if we had made up ‘blorg,’ ‘norg’, ‘lorg’, or ‘horg’, everyone
who has an English-type grammar would agree that their plurals are /blorgz/, /norgz/,
/lorgz/, and /horgz/). Another example is the English nominal prefix ‘ex-’ (meaning ‘for-
mer’). If we shaved off the beard of the ‘slorg’, it could be an ‘ex-slorg.’ Similarly, given the
other nouns we just made up, we can make ‘exblorg,’ ‘exnorg,’ ‘exlorg,’ and ‘exhorg.’ The
sense of ‘he’s an exblorg’ would immediately be clear – he used to be a blorg, but isn’t one
anymore. Note that this all works even if you do not know exactly what a ‘blorg’ is (which
is presumably the case). The regular plural and ‘ex-’ are both productive morphemes.
We can contrast this with an unproductive morpheme such as the suffix in ‘warmth’,
which clearly consists of ‘warm,’ meaning something like ‘warm,’ and /-θ/, meaning ‘the
property designated by the adjective to which I am suffixed.’ Thus ‘warmth’ means ‘the
property designated by the adjective ‘warm’.’ We find the same suffix in ‘strength’ (from
‘strong’ – note the wacky vowel change), ‘length’ (from ‘long’), ‘depth’ (from ‘deep’),
‘width’ (from ‘wide’), ‘filth’ (from ‘foul’!!), etc. The first thing you notice is that, un-
like, for example, ‘-ness,’ which does a similar job, you cannot make words like *‘uglith’
(the abstract property shared by all ugly things), *‘coolth’ (or *‘cilth’ or *‘celth’ or what-
ever, meaning the abstract property shared by all ‘cool’ things), or *‘hetth’ (the abstract
property shared by all ‘hot’ things – with the vowel change seen in ‘long’:‘length’ and
‘strong’:‘strength’). The abstract nouns made by /-θ/ are listable and finite – no new ones
4.4. Hierarchical Structure in Morphology 77

can be created (unlike plural and ‘ex-’ forms). ‘Ugliness,’ ‘coolness,’ and ‘hotness’ are all
fine, on the other hand. In fact, ‘-ness’ is quite productive. If we make up a new adjective,
like ‘slarm’ meaning ‘both chunky and bearded’, we can easily say that someone is richly
endowed with ‘slarmness’, but *‘slarmth’ is not going to be understood by anyone, even
by someone who you just told what ‘slarm’ means.
How are we to explain the difference between productive and unproductive morphol-
ogy? They key resides in the fact that while it is possible, even trivial, to list all abstracts
nouns in /-θ, it is not possible to list all plurals in /-z/. Any newly formed word will make
its plural in /-z/, but no newly-created words will make abstracts in /-θ/. How are we to
capture this fact in the model of linguistic knowledge we have been working on? Listable,
non-predictable information, such as the fact that ‘strong’ has an abstract ‘strength,’14
must be stored in the lexicon, just like unpredictable phonological information is stored
there (e.g., the fact that the word for ‘bean’ starts with a voiced bilabial stop). On the
other hand, if a noun does not have an irregular (and thus unpredictable) plural stored
with it (like ‘sheep,’ plural ‘sheep’), its plural is predictable: it will be in /-z/ (which may
be articulated as [s], [z], or [ɨz]). It would be silly to list all these regular plurals in the
lexicon – every noun which is not irregular in the plural has this regular plural. So the
lexical entry for ‘dog’ will not contain any information about its plural form – then the
default, predictable /-z/ plural will be used with this noun. On the other hand, ‘sheep’
will contain the information in its lexical entry that says, ‘Oh, by the way, my plural is
‘sheep.’
Now we see why newly coined words like ‘slorg’ make a plural in /-z/. When we told
you that there is a word ‘slorg’, we did not tell you that it had any kind of irregular plu-
ral. Therefore, you stored it in your lexicon with no information about how its plural is
formed. This is, of course, just like the way ‘dog’ or ‘banana’ is stored. All such nouns
show the plural morpheme in /-z/. Productive morphology is simply predictable, default
morphology. Non-productive morphology is unpredictable, irregular morphology.

4.4 Hierarchical Structure in Morphology

So far, we have looked at complex words as if the type of affixation and order of affixes
was all that was involved – this was something of a simplification. Consider the com-
plex word /ən-taj-əbl ̩/. We have used to dashes to separate this word into its component
morphemes. Note, however, that the word itself is ambiguous, i.e., has more than one
interpretation. On the one hand, it means ‘able to be untied,’ as in: ‘that knot is untieable,
I untied it in just five minutes!’ On the other hand, it can also mean ‘not able to be tied,’
as in ‘that knot is definitely untieable, I’ve been trying to tie it for twenty years and I’m a
14
Note that as a speaker with an English-type grammar, this fact is ‘predictable’ by you in some sense.
But that’s cheating – it’s predictable precisely because you already know that it is the case. The question of
predictability is one of whether or not someone who isn’t cheating – by already knowing the answer – can
predict whether ‘strong’ makes ‘strength’ or ‘strongness.’ They cannot.
78 Chapter 4. Morphology

world-renowned master at tying knots!’ How are we to account for the fact that this one
word has two very different interpretations?
Let us examine the matter of ambiguity in more detail. Ambiguity comes in two well-
defined types: lexical and structural ambiguity. Lexical ambiguity is based on the fact
that often different meanings are associated with an identical phonological form in the
lexicon. (These are homophones.) For example, the phonological form /najt/ is associ-
ated with a meaning which refers to a period of darkness that takes place every 24 hours
(on Earth), as well as with a meaning which refers to a medieval kind of guy who rides
around on a horse in armor or whatever. If I say, ‘the /najt/ was very dark’ you don’t
know whether I’ve just made a claim about the part of the day after sunset or about some
swarthy guy on a horse. Or, take a look at the following example:

• She can’t /ber/ children.

This sentence has at least the following meanings, based on the many different En-
glish words which have the same phonological form (/ber/). (1) She can’t stand children.
(2) She can’t give birth to children. (3) She can’t carry children (‘she can’t /ber/ the chil-
dren across the water — they’re too heavy’). (4) She can’t make children naked (it’s too
embarrassing).15 All of these various interpretations are based on the simple fact that the
phonological sequence /ber/ is associated with numerous, disparate meanings — in this
case, in particular, in several distinct verbs. It is thus a case of lexical ambiguity.
Structural ambiguity, by contrast, centers around the fact that the same elements
may be combined in more than one order, with the interpretation matching the order of
combination. The ‘un-tie-able’ example is a case of structural ambiguity. The two inter-
pretations are derived as follows:

(4.13) (a) combine /ən/ and /taj/ to get /əntaj/ ‘untie’, add /-əbl ̩/ giving ‘able to be [
untie ] -d’
(b) combine /taj/ and /-əbl ̩/ to get /tajəbl ̩/ ‘able to be tied’, add /ən-/ giving
‘not [ able to be tied ]’

/-əbl ̩/ in general means ‘abled to be Xed’ (so ‘kickable’ = ‘able to be kicked,’ ‘fakable’ = ‘able
to be faked,’ etc.), so when it gets added to /əntaj/ in (4.13a) it turns ‘untie’ into ‘able to
be untied.’ In (4.13b) we have added /-əbl ̩/ to ‘tie’ directly, giving ‘able to be tied.’ Then,
we negated this with /ən-/, giving ‘not able to be tied.’
Unfortunately, in the morphology we’ve been talking about thus far, there is no way
to represent the order of affixation. Put another way, the fact that ‘un-’ and ‘tie’ are closer
to one another (giving ‘untie’, to which ‘-able’ is added) in the first example, whereas ‘tie’
and ‘-able’ are closer to one another (giving ‘tieable’ to which ‘un-’ is added) in the second
15
For some speakers, there is yet another verb /ber/ — it means ‘to maul like a bear’ — so our example
sentence could also mean, for such speakers, ‘she can’t maul children like a bear (her teeth aren’t sharp
enough).’
4.4. Hierarchical Structure in Morphology 79

is not captured by the morphological analysis that says that ‘untieable’ is ‘un-tie-able.’ It
is standard practice in modern linguistics to show the order/closeness facts by means of
tree structures, as in (4.14) and (4.15).

(4.14) Tree for (13a)

Adj

V -əbl ̩

ən- V
taj

(4.15) Tree for (13b)

Adj

ən- Adj

V -əbl ̩
taj

Let us examine these trees in some detail, because you will see similar things later on in
the chapter on syntax. In the first tree (4.14), the fact that, in the interpretation given
in (4.13a), ‘ən-’ and ‘tie’ are closer together is captured by graphically designating them
as sisters in the tree (sisters are nodes of the tree which have a common mother node –
in (4.14) ‘un-’ and ‘tie’ are daughters of the node labelled ‘V’). The label ‘V’ indicates that
when ‘un-’ attaches to a verb like ‘tie’ the result is still a verb (i.e., ‘un-’ prefixation does
not change the part of speech of the morpheme to which it attaches). The little subtree
over ‘un-’ and ‘tie’ is itself a sister to ‘-able’ – indicating that ‘-able’ is further away from
‘tie’ than is ‘un-’ (it’s an ‘aunt’ of ‘tie’, if you will – rather than a sister; your sister is a closer
relative than your aunt, right?). In (4.15) by contrast, ‘-able’ is a sister of ‘tie’. When ‘-able’
gets added to a verb (like ‘tie’), the result is not a verb, but rather an adjective – this is
indicated by the fact that the mother of ‘tie’ and ‘-able’ is labelled ‘Adj’ for ‘adjective.’ In
this tree, ‘tie’ and ‘-able’ are sisters, and thus closer to each other than either is to ‘un-.’ By
contrast, ‘un-’ is a sister of the complex adjective ‘tie-able.’ Since ‘un-’ doesn’t change the
part of speech of an element that it attaches to, the result of adding ‘un-’ to an adjective
is still an adjective. This is indicated by the label ‘Adj’ at the top of the tree.
Trees encode what we usually call ‘hierarchical’ information about the structure of
complex words which is lacking in a simple representation such as ‘un-tie-able.’ Since
that information is critical to the interpretation of the word, it must be part of its repre-
sentation.
Finally, we can see that ‘structural ambiguity’ results when there are two (or more)
ways in which multiple morphemes can combine into a complex morpheme (and, con-
sequently, two or more tree structures). Since you cannot, without further information,
80 Chapter 4. Morphology

deduce which of the two possible structures is actually in the mind of the person talk-
ing to you if someone just comes up and asks you what ‘untieable’ means, the string of
morphemes is ambiguous. You can assign it both or either of the possible, hierarchical
analyses. We will return to this matter when we discuss syntax.

4.5 The ‘cranberry’ problem

In general, morphological segmentation is quite straightforward, given sufficient evi-
dence. It does not take a lot of effort to figure out from ‘untie,’ ‘unwrap,’ ‘uncover,’ and
‘unbutton’ that English has a verbal prefix ‘un-’ which means ‘to reverse the action desig-
nated by the verb’ (so ‘unwrap’ is to reverse the action of wrapping, etc.). However, con-
sider the following examples: ‘lukewarm,’ ‘huckleberry,’ ‘cranberry’, and ‘hamburger.’
It is clear enough that the ‘warm’ part of ‘lukewarm’ means ‘warm,’ but what does the
‘luke’ part mean? Notice the weirdness of *‘lukecool,’ *‘lukesnowy,’ and *‘lukesmart’ (an
asterisk is used in linguistics to indicate ill-formed structures). Similarly, the ‘berry’ part
of ‘huckleberry’ and ‘cranberry’ appears to mean ‘berry,’ but what on earth is a ‘huckle’
or a ‘cran’? And ‘hamburger’ is rather worse. A ‘fishburger’ is a sandwich made of fish,
a ‘chickenburger’ is one made of chicken but a ‘hamburger’ is not one made of ‘ham’ but
rather one made of ground beef. A ‘cheeseburger’ is not a cheese sandwich, but instead
something made of ground beef with cheese on it. Does the ‘ham’ in ‘hamburger’ mean
‘ground beef’? Does the ‘cheese’ in ‘cheeseburger’ mean ‘ground beef with cheese on it’
instead of meaning just ‘cheese’? The whole affair is assuming nightmare proportions.
Because the first widely discussed example of such a phenomenon – the partial analyz-
ability of what appear to be complex words – involved ‘cranberry,’ these sorts of examples
have come to be called ‘cranberry morphemes.’ Ocean Spray has compromised this ter-
minology by the invention of such beverages (with their corresponding neologisms) as
‘Cranapple,’ and ‘Crangrape.’ They have not, however, solved the basic linguistic prob-
lem which is – are these simple words (consist only of a single morpheme) or complex
words (consist of more than one morpheme)? If they are simple, then we have to some-
how force ourselves to ignore those sequences of sounds which are meaningful, such as
‘berry.’ If, on the other hand, they are complex, we must find some meaning to assign
to sequences like ‘cran’ since every morpheme, by definition, must have a meaning as-
signed to it. This is a difficult question which has been, at various times, hotly debated in
the literature. While we have no doubt that you find the topic totally engaging as well, we
have decided it is of less central importance than many other topics and will not pursue
it further here.

4.6 Summary
This chapter illustrates the ways in which the grammar stores and manipulates mean-
ingful units or morphemes. We see that, as in phonology, there are two levels of represen-
4.6. Summary 81

tation with a computational component interfacing between the two. For morphology,
the level of underlying representation is the morphemic level and the less abstract (sur-
face) level consists of allomorphs. Morphological computations create complex words from
simple words through systematic processes for combinining free and bound morphemes.
82 Chapter 4. Morphology

4.7 Exercises
In answering questions about morphemes, be specific about type of affix, type of process,
and so on.

1. Southern Barasano

Southern Barasano
singular plural
kahea ‘eye’ kahe ‘eyes’
bitia ‘bead’ biti ‘beads’
kĩa ‘cassava’ kĩ ‘cassavas’

a) Which of the following statements is true about Southern Barasano?

(a) Southern Barasano has a prefix /a/ which marks the singular.
(b) Southern Barasano has a suffix /a/ which marks the singular.
(c) Southern Barasano has a prefix /a/ which marks the plural.
(d) Southern Barasano has a suffix /a/ which marks the plural.

2. Bahnar – note that the symbol ɲ is used to indicate a palatal nasal (as in Spanish
<ñ>, Italian <gn>, or some English-speakers’ pronunciation of ‘onion’ [əɲjən]).

• həɲír ‘sweaty person’

• təbúl ‘person to sleep curled up with’
• məlwɛ́ŋ ‘drunk person’
• həháh ‘laughing and giggling girl’
• həɲír həɲóɲ móɲ ‘each of many sweaty people’
• məlwɛ́ŋ məlwóɲ móɲ ‘each of many drunk people’
• həháh həhóɲ móɲ ‘each of many laughing and giggling girls’

a) What morphological process do we see at work in the Bahnar data?

(a) replication
(b) reconstruction
(c) reduplication
(d) stuttering

b) How would you say ‘each of many people to sleep curled up with’ in Bahnar?
4.7. Exercises 83

(a) təbúl təbúl móɲ

(b) təbúl təbóɲ móɲ
(c) təbóɲ təbúl móɲ
(d) jəbədəbədu

3. Xavante

Xavante
N your N
du ‘stomach’ ʔaddu ‘your stomach’
ʔra ‘child’ ʔajʔra ‘your child’
hiʔrãti ‘knee’ ʔajhiʔrãti ‘your knee’
tɔ ‘eyes’ ʔattɔ ‘your eyes’
ʔwa ‘tooth’ ʔajʔwa ‘your tooth’
brɔ̃ ‘wife’ ʔajbrɔ̃ ‘your wife’
ʃɛːrɛ ‘hair’ ʔaʃʃɛːrɛ ‘your hair’
paːra ‘foot’ ʔajpaːra ‘your foot’
bãːbã ‘father’ ʔajbãːbã ‘your father’

a) What are the allomorphs of the morpheme meaning ‘your’ in Xavante?

(a) ʔaj, ʔat, ʔad, and ʔaʃ
(b) ʔaj and ʔaʃ
(c) ʔa
(d) jəbədəbədu

b. What is the underlying morpheme meaning ‘your’ in Xavante?

(a) ʔaj, ʔat, ʔad, and ʔaʃ
(b) ʔaʃ
(c) ʔad
(d) ʔaj

c. The precise rule responsible for the allomorphs of ‘your’ in Xavante is:
(a) /j/ assimilates completely to (i.e., becomes the same as) following stops
(b) /j/ assimilates completely to (i.e., becomes the same as) following alveo-
lars and alveopalatals
(c) /j/ assimilates completely to (i.e., becomes the same as) following voice-
less consonants
84 Chapter 4. Morphology

(d) /j/ assimilates completely to (i.e., becomes the same as) following stops
and fricatives

4. Waorani

Waorani
abo ‘I see’
adã ‘she sees’
amõ ‘we (inclusive) see’
ãmo ‘I say’
ãŋã ‘he says’
kæ̃ nã ‘she eats’
kæ̃ mõ ‘we (inclusive) eat’
kækã ‘he does’
kæmõ ‘we (inclusive) do’

a) The allomorph(s) of ‘I’ in Waorani is or are:

(a) /bo/ and /mo/
(b) /mo/
(c) /bo/
(d) /abo/

b) The allomorph(s) of ‘she’ in Waorani is or are:

(a) /da/
(b) /dã/ and /nã/
(c) /da/ and /na/
(d) /na/

c) The allomorph(s) of ‘he’ in Waorani is or are:

(a) /kã/ and /ŋã/
(b) /ka/ and /ŋa/
(c) /a/ and /ã/
(d) /ŋã/

d) The rule governing the distribution of all of these allomorphs in Waorani is:
(a) nasalize stops after non-nasal vowels
(b) denasalize stops after non-nasal vowels
(c) nasalize stops after nasal vowels
4.7. Exercises 85

(d) denasalize stops after nasal vowels

e) If you were about to marry a Waorani person and were asked during the mar-
riage ceremony to assent to the marriage, you would probably say:
(a) kæmo
(b) kæbo
(c) kæ̃ mo
(d) hə

5. Copainalá Zoque

Copainalá Zoque
present tense past tense
kenpa ‘he looks’ kenu ‘he looked’
sikpa ‘he laughs’ sihku ‘he laughed’
witpa ‘he walks’ wihtu ‘he walked’
kaʔpa ‘he dies’ kaʔpu ‘he died’
nahpa ‘he kicks’ nahpu ‘he kicked’
sospa ‘he cooks’ sohsu ‘he cooked’

a) The morpheme which marks the past tense in this variety of Zoque is:
(a) the suffix /u/
(b) the prefix /u/
(c) the infux /u/
(d) reduplication

b) The morpheme which marks the present tense in this variety of Zoque is:
(a) the suffix /a/
(b) the suffix /pa/
(c) the infix /p/
(d) the infix /a/

c) There is some variation in the verbal roots involving /h/. What rule triggers
this variation?
(a) delete /h/ if it is followed by a stop
(b) delete /h/ if it is followed by two consonants
(c) delete /h/ if it is not followed by two stops
(d) delete /h/ whenever you don’t feel like saying it
86 Chapter 4. Morphology

d) What is going on with Zoque ‘kick’?

(a) The root is /nah/ and the present tense has an irregular /p/.
(b) The root is /nahp/ and there is a rule which deletes one of two adjacent
/p/’s.
(c) The root is /nahp/ and there is a rule which deletes /p/ after /h/.
(d) The root is /kɪk/ and nothing’s going on with it.

e) Do the rules of /p/ deletion and /h/ deletion need to be ordered with respect
to one another?
(a) No.
(b) Yes. /p/ deletion needs to occur after /h/ deletion.
(c) Yes. /h/ deletion needs to occur after /p/ deletion.
(d) Yes, but it does not matter which order you use.
 Chapter 5
Syntax

5.1 Introduction
Having surveyed the structure of sounds and of complex words, we now turn to the study of the
structure of phrases and sentences. Actually, as far as our every day experience of language goes,
we are generally confronted by strings of words, put together in a manner made possible by the
syntactic system of the grammar in question. It is therefore the sentence which represents the
most salient aspect of language in actual use.

5.2 Grammaticality
It is apparent that some strings of words represent possible sentences for a given grammar, while
other strings of words do not. Compare the strings in (5.1a) and (5.1b) below, for example.

(5.1) a. The clown likes only fresh twinkies.

b. *Twinkies only the likes fresh clown.

Anyone with a grammar which permits them to read this book knows that while the string of
words in (5.1a) represents a sentence of English, those same words arranged as in (5.1b) do not.
In particular, (5.1a) appears to have a relatively straightforward interpretation (it asserts of some
specific clown that he has the property of liking only fresh twinkies, as opposed to stale twinkies
or fresh asparagus).1 By contrast, the string of words in (5.1b) does not appear to assert anything
about anyone (or anything). One can easily imagine contradicting (5.1a): ‘no, no, he likes stale
twinkies as well,’ for example, but what could one say to (5.1b) except ‘what are you trying to
say?’
1Note that the clown in question may or may not like fresh asparagus, the sentence says nothing about the
matter one way or another.

79
80 Chapter 5. Syntax

It is traditional in modern syntactic literature to refer to a string of words which speakers

with the relevant grammar consider to be a ‘sentence’ as a grammatical string or a grammatical
sentence, while a string of words which does not have this property is generally called ungram-
matical and is marked with an asterisk, as in (5.1b) above.
It is important to try to understand precisely what the term ‘grammatical’ means in modern
linguistics – as with other terms we have seen, the intended force of this term is technical and
highly specific.2 First, grammaticality is not a question of whether or not you have ever heard a
particular sentence or whether the string involved is one you are likely to say. For example, it is
probable that you have never said the sentence in (5.2).

(5.2) I just whacked my little toe against the P-Z volume of the Oxford English Dictionary
again!

It may be for many of you that it is extremely unlikely that you would ever say such a sentence –
e.g., you may not have the Oxford English Dictionary, or you may be really careful when you walk
around barefoot. Nevertheless, even for those of you for whom utterance of such a string is vir-
tually unthinkable, it is clearly a perfectly well-formed sentence for speakers with English-type
grammars. (Syntactic well-formedness is synonymous with grammaticality.) Well-formedness,
therefore, has nothing to do with likelihood of utterance. It is also the case that well-formedness
has nothing to do with length of utterance – a well-formed sentence may be arbitrarily long. For
example, if we had you make up the longest sentence you could think of, we could simply add
‘We know that’ to the beginning of that sentence, and the result would be well-formed. You, in
turn, could add ‘You know that’ to the beginning of our sentence (so that we get ‘You know that
we know that…’) and we could all go on making the sentence longer and longer in this manner
for days, weeks, or years on end. The resulting sentence would be boring to listen to and kind
of useless, but it would be well-formed, and therefore grammatical. The likelihood that anyone
is going to say a sentence that it would take us years of dedicated effort to construct is, we are
happy to say, vanishingly small.
Secondly – and this is sometimes a difficult point for non-linguists to fully appreciate – gram-
maticality is about syntactic well-formedness only. It therefore does not matter if the string of
words makes sense at all when one is attempting to determine whether that string is grammati-
cal. Look at the examples in (5.3).

(5.3) a. The refrigerator hid under the mouse.

b. Excitement said that the television had a new hairdo.
c. My dog always wins at poker.

Contrast these with the parallel sentences in (5.4).

2Its more literal reading – producible by the grammar – is actually very applicable here.
5.3. The simple sentence in English 81

(5.4) a. *Under mouse the hid refrigerator the.

b. *Said the had new a that excitement television hairdo.
c. *Poker my always at wins dog.

Clearly, the examples in (5.3), while semantically odd, are well-formed in a way in which those
in (5.4) – which involve exactly the same words – are not. One can easily imagine somewhat
fantastic stories in which each of the sentences in (5.3) might occur (like ‘the tale of the tiny
animate refrigerator,’), but the strings of words in (5.4) cannot be part of any story, no matter
how fantastic.3 Notice that the oddness of the sentences in (5.3) is essentially oddness about the
world – refrigerators don’t normally ‘hide,’ excitement can’t ‘speak,’ TVs don’t have hairdos, and
dogs don’t play poker. The world, and human conceptions of it, change all the time. It would be
quite a bad thing, therefore, if your grammar did not permit you to say things just because they
were not currently sensible. Language is flexible enough to serve your needs whatever bizarre
situation you happen to find yourself in.
Basically, for at least certain types of words (nouns, for example), syntactic well-formedness
has nothing to do with the ‘meaning’ of the word in question. Your grammar does not care
whether you are talking about ‘dogs’ or ‘refrigerators’ – you can construct exactly the same sen-
tences using either noun.4 The same may be said for adjectives, e.g. – every sentence about ‘a
brown refrigerator’ is equally well-formed if instead we say ‘a large refrigerator’ or ‘a skeptical
refrigerator’ (though those formed with the latter phrase may be semantically odd). This is the
contrast between syntax and semantics: syntax is about syntactic well-formedness, semantics is
about meaning.

5.3 The simple sentence in English

The basic unit of syntactic analysis is the sentence. A (grammatical) sentence is a string of words
which the syntactic component of the grammar can assign a coherent structural analysis to. The
strings in (5.1a), (5.2), and (5.3a-c) all represent sentences, while those in (5.1b) and (5.4a-c) do
not. A sentence is made up of words, which, as we saw in our discussion of morphology, may be
either (morphologically) simple or complex. As the contrast between (5.3) and (5.4) suggests, in
English-type grammars at least, the arrangement of words is critical as to whether or not a given
string represents a (grammatical) sentence or an (ungrammatical) string.5
As we stated, the semantics of the individual words does not play a key role in determining
syntactic well-formedness. What does turn out to be extremely relevant to this determination
3Unless, of course, the fantasy is that the language of the story in question is no longer a variety of English. But
you would be pushing it …
4A noun (just in case) is a word which designates an entity, perhaps abstract. Given a sufficiently liberal notion
of ‘thing,’ the traditional definition (‘a noun is a word which designates a person, place, or thing’) is adequate for
our purposes.
5The latter is sometimes called an ungrammatical ‘sentence’ which is something of a contradiction in terms,
given what we have just said.
82 Chapter 5. Syntax

is the so-called syntactic category of the word in question. While we can substitute virtually any
noun for ‘dog’ in the sentence ‘the dog is ill-behaved’ – giving ‘the cat is ill-behaved,’ ‘the tree
is ill-behaved,’ ‘the nation is ill-behaved,’ ‘the excitement is ill-behaved,’ etc. – if we attempt to
place a verb (like ‘drowns’ or ‘believes’) in the position of the noun ‘dog’, we get ungrammatical
strings: ‘*the drowns is ill-behaved’ and ‘*the believes is ill-behaved’. (Recall that ungrammatical
strings are marked with an asterisk).
Also clearly relevant to grammaticality are certain morphological properties of the words in
question. Examine the sentences and ungrammatical strings in (5.5).

(5.5) a. The dog sits on the carpet.

b. *The dog sit on the carpet.
c. The dogs sit on the carpet.
d. *The dogs sits on the carpet.

In general, verbs must agree with the number of their subject in English-type grammars (though
there is some variation in the dialects).6 That is, if the subject is singular (‘the dog’) the verb must
be singular (‘sits’), while if the subject is plural (‘the dogs’) the verb must be plural (‘sit’). Such
agreement processes are common in human languages. Agreement will not be a major concern
of ours in our brief survey of syntactic processes – but it is quite important in advanced syntactic
theorizing.
What does one need in order to have a sentence? Examine the strings in (5.6).

(5.6) a. *The dog on the carpet.

b. *Sits.
c. *The sits.
d. The dog sits.
e. The dog sits on the carpet.
f. Dogs sit.
g. Dogs sit on carpets.

Why are the strings in (5.6a-c) ungrammatical? Superficially, the contrast between (5.6a) and
(5.6e) would seem to indicate that the problem with (5.6a) is that it lacks a verb. Similarly, the
contrast between (5.6b) and (5.6f) would seem to indicate that the problem with (5.6b) is that
it lacks a subject. (5.6c) tells us that ‘the’ is not a possible subject (and thus (5.6c) doesn’t fix the
problem with (5.6b)) – whereas (5.6d) tells us ‘the dog’ is a possible subject. It appears, then, that
6The ‘subject’ is what the sentence makes an assertion about – in the sentences in (5.5) it is one or more dogs.
We will return to this issue shortly.
5.3. The simple sentence in English 83

the minimal sentence will have an appropriate subject and a verb.7 This makes sense, of course.
Sentences basically predicate, or attribute, some property (expressed by a verb and its satellites)
of some entity (expressed by a noun or pronoun and its satellites). Thus (5.6f) predicates of the
class of entities designated as ‘dogs’ that they have the property of engaging in sitting. (5.6g)
asserts that entities known as ‘dogs’ have the property of sitting on entities known as carpets.
(5.6d), by contrast, says only that some specific dog – known to the speaker and hearer (this is
what ‘the’ does) – has the property of sitting. Note that it does not say that dogs in general don’t
sit, it is silent on the general properties of dogs. Every declarative8 sentence predicates some
property of some entity.
(5.6c) indicates that there are restrictions on what can be a subject. Typical types of subject
can be seen in (5.7).

(5.7) a. The dog sits on the carpet.

b. The big dog sits on the carpet.
c. The big dog in the corner sits on the carpet.
d. She sits on the carpet.

What is interesting about these examples is that in (5.7a-c) there is always a noun present in the
string that represents the subject (‘the dog’; ‘the big dog’; ‘the big dog in the corner’) and that if
we leave that noun out, the strings become ungrammatical (*the sits on the carpet; *the big sits
on the carpet; *the big in the corner sits on the carpet). In (5.7d) there is a pronoun (a substitute
for a noun phrase) which cannot be omitted in English-type grammars (*sits on the carpet).
Similarly, acceptable predicates appear to require a verb.

(5.8) a. The dog sits.

b. The dog is fat.
c. The dog drinks the water.

Again, without their verbs, these strings are not sentences (*the dog; *the dog fat; *the dog the
water).9
It appears, then, that a sentence consists of two basic parts: a subject, which designates some
entity in the universe about which something is going to be said, and a predicate, which desig-
nates some property which the sentence asserts is true of the entity designated by the subject.
The subject must contain a noun or a noun phrase substitute (a pronoun); the predicate must
7We will consider a little later imperative sentences such as ‘Leave!’ which appear to lack a subject.
8We will consider other types of sentences – interrogatives, for example, in what follows.
9Note that it does not follow that you cannot say the strings. If someone asks ‘Who sat on the carpet?’, the
answer ‘the dog’ is perfectly acceptable. You might, however, consider what the answer ‘the dog’ means – e.g., can
it mean ‘the dog runs’ or ‘the dog sleeps on the carpet’?
84 Chapter 5. Syntax

contain a verb. It is important to note that judgements about the grammaticality of these sen-
tences do not depend upon overt knowledge of traditional grammatical terms like ‘noun’ and
‘verb.’ You will find the ill-formed sentences just as ill-formed and the well-formed ones just as
well-formed whether you can name the parts of speech contained in them or not.
These are not properties of English sentences, only. As far as we know, these are universal
properties of natural language. One might say, instead, that to qualify for ‘sentencehood’ in a
human brain, a string must contain both a subject and a predicate.

5.4 Categories
As we have seen, there are apparently restrictions on what can function as a subject of a sen-
tence and on what can function as a predicate. It makes sense, if we are to try to understand
how sentences are constructed, to explore these restrictions – what makes for an acceptable
sentence is undoubtedly in part a function of what makes for acceptable components (subject
and predicate) of a sentence.
We will first take a look at acceptable subjects. In (5.6f) the subject is simply the noun ‘dogs.’
We can conclude from this that a noun alone can be a subject. We will designate nouns in general
by the symbol N. In (5.7a) the subject is ‘the dog.’ ‘the’ is what linguists call a ‘determiner’ (like, in
English, ‘a(n),’ ‘his’, ‘my’, ‘that,’ ‘these,’ etc.). We will designate determiners by the symbol D. An
acceptable subject thus may be either N or D N. One of the ways to think about the relationship
between N and D N is that a determiner may be present (‘the dogs’) but is not required (‘dogs’)
– i.e., a determiner is optional. Notice that the noun is not optional (*the is happy, *my is here).
If we designate optional material by placing it in parentheses, we can say that a subject may
have the form ‘(D) N’ – that is, an optional determiner followed by a noun. The subject of (5.7b),
‘the big dog,’ contains, in addition to a determiner and noun, an adjective, ‘big.’ If we designate
adjectives with the symbol A, we can now say that a subject may have the form ‘(D) (A) N’ – i.e.,
an optional determiner, an optional adjective, and a noun.
It is not entirely accurate, however, to say that ‘(D) (A) N’ is a generalization about subjects.
Examine the sentences below.

(5.9) a. The cat sits on dogs.

b. The cat sits on the dogs.
c. The cat sits on the big dogs.

These examples indicate that the same N, (D)N, or (D)(A)N pattern occurs in non-subject posi-
tions. In (5.9), they are the object of the preposition10 ‘on’. Given that this, too, is possible, we do
not want to connect N, (D)N, and (D)(A)N only to subjects. It turns out, in fact, that these same
patterns occur in many different positions in the sentence. Interestingly, wherever you can have
10A preposition is a word which expresses spatial, temporal or causal relationships – typical examples from
English include in, on, over, above, under, through, down, across, etc.
5.5. Constituency and Structure Building 85

an N, you can also have the other forms which include an N. We can make a generalization about
this by calling an N and its satellites a Noun Phrase or NP, for short. It thus appears that Noun
Phrases (NPs) are fundamental building blocks of the sentence. They are required in subject
position, but they also occur elsewhere (as 5.9a-c show).
We have not yet considered the subject of (5.7c), ‘the big dog in the corner.’ Notice that al-
though this subject contains two nouns, only one of them is the real or structural subject (namely,
the ‘dog’ – i.e., it is the dog in (5.7c) which is sitting on the carpet, not the corner that is doing
the sitting). ‘the corner’, by contrast, is the object of the preposition ‘in’. Let us designate prepo-
sitions with the symbol P. Notice that ‘*the dog in sits on the carpet’ is ungrammatical. This is
because ‘in’ requires an object when used as a preposition (prepositions are pre-posed to their
nominal objects). We can therefore posit another type of a phrase, call it a Prepositional Phrase
(PP), which consists of a P and an NP (i.e., a preposition and its noun phrase object). A Noun
Phrase, as (5.7c) shows, can also contain a PP – thus the structure of Noun Phrases appears to be
the following: (D) (A) N (PP). As more and more items are incorporated into our ‘possible NP’,
though, we need to ask whether all those constituents have the same relationship to one another.
(Recall the ‘untieable’ example from the preceding chapter that outlined different possible mor-
pheme relationships in words of three or more morphemes.)

5.5 Constituency and Structure Building

The internal structure of phrases turns out to be fairly complex. Once again, it is speakers’ in-
tuitions, in large part, which have led us to believe this. We have gotten very consistent results
about syntactic categories and structure from what we call constituency tests. Coordination is
one such test, illustrated below.

(5.10) The dog [sits on the sofa] and [sleeps on the bed].

(5.11) The dog sits [on the sofa] and [on the bed].

(5.12) The dog sits on [the sofa] and [the bed].

In each of the above examples, two elements belonging to the same syntactic category have
been conjoined – two VP’s, two PP’s, or two NP’s. Interestingly, however, it seems that only like
categories can be conjoined by ‘and.’

(5.13) *The dog bites with his teeth and the mailman.

This is further support for the hypothesis that the syntactic component is sensitive to cat-
egories such as N, PP, and so forth. In addition, it suggests that groupings of certain elements,
such as PP’s, form sub-constituents within the sentence.
Another type of constituency test is substitution. The results of these tests strongly support
the idea that syntactic structure is hierarchical in nature.
86 Chapter 5. Syntax

(5.14) The dog sits on the carpet and the cat sits there, too. [there = ‘on the carpet’]

(5.15) The dog sits on the carpet and the cat does, too. [does = ‘sits on the carpet’]

In the above, ‘there’ substitutes for a PP and ‘does’ substitutes for a VP.
There is another substitution, called ‘one substitution’ that we can do for elements in an NP.
We will go through ‘one substitution’ in some detail to show how the argument for constituent
structure actually works and what the internal structure of an NP looks like. Consider the sen-
tences below.

(5.16) a. I saw a big yellow house and you saw one, too. [one=‘a big yellow house’]
b. *I saw a big yellow house and you one, too. [one=‘saw a big yellow house’]
c. I saw a big yellow house and you saw a small one [one=‘yellow house’]
d. I saw a big yellow house and you saw a small white one [one=‘house’]
e. *I saw a big yellow house and you saw a one house [one= ‘big yellow’]
f. *I saw a big yellow house and you saw big yellow one, too. [one=‘a house’]

In each case, the part of the phrase that we understand ‘one’ to be substituting for is indicated
in square brackets. These sentences suggest the following:

• ‘one’ can only substitute for NP-related constituents (including NP itself) but not for verbs,
for example.

• ‘one’ cannot substitute for just any random constituents within the NP

• the NP seems to have an internal structure which is more complex than the linear order
of the words in the NP suggests.11

One fairly clear pattern stands out here – the noun is always substituted for. From this, and
from the interpretation of constituents in the NP, all of which seem to be giving us more infor-
mation about the N, it seems like the syntactic structure of this phrase is built up around the
noun. Given that and also the fact that subjects require a noun (but not an adjective, for exam-
ple), we’ll consider the noun the head of the NP. Based on the one-substitution results above,
which suggest that ‘house’, ‘yellow house’, ‘big yellow house’ and ‘a big yellow house’ are all pos-
sible constituents of the NP, we can illustrate the syntactic structure graphically using syntactic
trees, as below.12
11If ‘one’ can substitute for the determiner/indefinite article (‘a’) and for the noun, as we see from (5.16a-e), why
is (5.16f) ungrammatical?.
12Bracketing is also used to illustrate syntactic structure but the hierarchical nature of the structure is not quite
as clear in that system, among other things.
5.5. Constituency and Structure Building 87

(5.17) N

N
house

(5.18) X?

Adj N
yellow house

(5.19) X?

Adj X?
big
Adj N
yellow house

(5.20) NP

Det X?
a
Adj X?
big
Adj N
yellow house

There is a crucial property shared by each of these trees that we believe to be a general property
of all syntactic structure. This is that syntactic structure-building operations create only binary
(or unary) relationships – that is, at most two constituents can be connected to one another in
a single operation. We refer to this type of structure as binary branching.13
We still have to capture structurally the relationships that speaker intuitions suggest hold
true within the phrase, for example, that the adjectives each modify the noun instead of mod-
ifying each other. One way to do this is to formally build in levels of representation between
the head of the phrase, N in this case, and its highest node NP. Traditionally, this has been done
through X’ Theory (read ‘x bar theory’). X’ Theory mandates that for every head X, a correspond-
ing phrasal category, XP, is projected and any constituents that modify X are joined in at X’ (x-
bar) levels. Not all constituents that modify the head are exactly equal, however. There are
13This is considered a significant boost for the child acquirer, since the child’s hypothesis space is then more
limited.
88 Chapter 5. Syntax

two types of constituents that, when present, have a special status with respect to the head of a
phrase. One of these is the specifier and the other is the complement. These two constituents are
structurally related to the head of the phrase in the way that is shown below.

(5.21) a) XP b) XP

ZP X’ X’ ZP
Spec Spec
X WP WP X
Head Complement Complement Head

c) XP d) XP

ZP X’ X’ ZP
Spec Spec
YP X’ X’ YP

X WP WP X
Head Complement Complement Head

These schematic trees illustrate several important aspects of syntactic relationships, using let-
ter variables (so X, W, Y and Z could be any syntactic categories). Perhaps the most important
of these is the sisterhood relationship. As with genealogical trees, sisters share a single ‘mother’
node. In the trees above, the X heads and their WP complements are sisters. There are a number
of additional sisterhood relationships in these structures which we leave to you to identify. Cru-
cially, these sisterhood relationships are independent of linear order. As pair (a) and (b) and pair
(c) and (d) illustrate, X and WP occur in different linear orders – X precedes WP linearly in (a)
and (c) and X follows WP linearly in (b) and (d). However, the syntactic relationships remain the
same across all of these because those relationships are based on hierarchical structure, sharing
a ‘mother’ node, which all X and WP pairs in these structures do. Two additional syntactic rela-
tionships are illustrated by these trees. The phrasal constituents ZP, labelled Specifiers, have the
unique property of being direct ‘daughters’ of the XP. In addition, the WPs, as sisters specifically
of the heads of the phrase (XP), have a special status which we call ‘Complement.’14 In addition,
there are dominance relationships, denoting hierarchical positions. In the tree above, XP im-
mediately dominates both Spec and the highest X’ and dominates every node below that.15 Note
that in these X’ trees, unlike in the trees we have seen so far, there is only one head, X, and all the
other branches end in phrasal categories (ZP, YP, WP). In future, we will assume that every lex-
ical item projects a corresponding phrasal category, so D projects DP and Adj projects AdjP, for
example. This ‘bare’ X-X’-XP structure, not shown above, is the minimum structure of a head and
14The terms ‘specifier’ and ‘complement’ are simply convenient names given to these structural positions.
15‘Immediate’ domination is reserved for cases where no other nodes intervene.
5.5. Constituency and Structure Building 89

its phrasal category. In a sentence like ‘Bees sting.’, the V ‘sting’ projects V’ and VP and that is the
entirety of that lexical head’s structure/domain. The schematic trees above include additional
possible relationships above and beyond the minimum required structure.

5.5.1 Lexical vs. Functional Heads

The phrase structure of X’ Theory does not refer to particular syntactic categories, like noun,
determiner, and so on, but instead allows one to substitute a head of any type for the variable ‘X.’
The differences in phrases and sentences, then, must fall out, at least in part, from differences in
the properties of the heads of phrases. One major difference in heads seems to be whether they
contain a kind of ‘contentful’ semantics or have, instead, simply a ‘technical’ function. Examples
of this contrast, between lexical and functional heads are given in the following pairs:

(5.22)

Lexical Heads Functional Heads

N(oun) P(reposition)
V(erb) D(eterminer)
Adj(ective) Q(uantifier)
Adv(erb) M(odal)/I(nflectional)
Lexical heads are typically open class items. This means that they are categories which can be
added to freely (by borrowing words or by coining new words, for example). Functional heads,
on the contrary, belong to closed classes, or categories that do not normally allow the addition
of new members. In the syntax, both lexical heads and functional heads are necessary for an
interpretable structure – lexical heads carry all the real content and functional heads bring the
contentful pieces together into an interpretable whole. Let’s bring these bits together and see
what the structure of a complete sentence might look like.
We’ll start with an unordered selection of lexical heads – {see, boy, girl, stairs}. Notice that
if this was all we had to build our sentence with, there are too many combinatorial possibilities
and too many missing pieces of information for us to succeed at our task. The subject might just
as easily be ‘boy’ or ‘girl’; either the boy or the girl or both might be a particular boy or girl or
just any indeterminate boy or girl; the action may take place in the past, future, present, be only
hypothesized to take place; and so on. Clearly some more elements are required here but they
all appear to be functional elements. We also need to keep in mind the point made earlier that
‘sentencehood’ is basically the act of predicating something (some action, attribute, etc.) about
some entity. Since there’s only one verb in our list, it looks like the predication is going to involve
‘seeing’. To get something that is a reasonable sentence in English, we’re going to need determin-
ers for the NP’s and maybe a preposition to squeeze ‘stairs’ into the picture. Most importantly,
however, we need a functional element that will allow us to connect the two main constituents,
the subject and the predicate, to one another. This connection is going to tell us both how the
subject and predicate are related in time (past, present, or future) and may also tell us some-
thing about how the speaker regards the predication event, specifically as something that may
90 Chapter 5. Syntax

happen, can happen, should happen, and the like. The functional heads that do this work are
usually called ‘modals’ and ‘auxiliaries’ and we will classify them into a syntactic category, ‘I’,
(for ‘Inflection’). By adding the necessary functional heads, we are able to make well-formed
sentences like:

(5.23) a. The girl has seen the boy under the stairs.
b. The boy may see the girl on the stairs.
c. The girl on the stairs will see the boy.

Building sentences involves building X’-theoretic structure using some combination of lexical
and functional heads that must include an NP that serves as the subject, a VP that serves as the
predicate, and an IP that serves to relate the subject and predicate in the ways stated above. Here
are trees for the sentences in (5.23a) and (5.23c) above (sentence 5.23b has the same structure as
5.23a).

(5.24) IP

NP I’

DP N’
I VP
D’ N has
girl V’
D
the
V’ PP

V NP P’
seen
DP N’ P NP
under
D’ N DP N’
boy
D D’ N
the stairs
D
the
5.5. Constituency and Structure Building 91

(5.25) IP

NP I’

DP N’ I VP
has
D’ N PP V’
girl
D P’ V NP
the seen
P NP DP N’
on
DP N’ D’ N
boy
D’ N D
stairs the
D
the

With elements drawn from lexical and functional categories and the X’ schema we’ve dis-
cussed so far, we can create a multitude of sentences. There is really only one area where we still
fall a bit short and that is in our ability to produce sentences like those below.

(5.26) a. John may think that Sally has bought a car.

b. Sally may ask whether John did buy the car.
c. Sally has asked if John will buy her car.

These sentences are unlike previously-discussed sentences in two ways. First, if we go by our defi-
nition of sentences as predications, they each contain, in some sense, two sentences (or clauses).
In (5.26a), for example, John will do some thinking and Sally has done some buying.16 Second,
they each have a constituent between their clauses of a type that we have not yet encountered
– ‘that’, ‘whether’ and ‘if’. When we look around at other sentences, we see that these three lex-
ical items are very common and seem to introduce additional clauses which are subordinate to
the main clause – ‘that’ introduces declarative clauses and ‘whether/if’ introduce indirect ques-
tions.17 Following a long tradition, we’ll identify these items as members of a category ‘C’ (for
‘Complementizer’). CP’s almost always dominate IP’s and do always dominate main clause IP’s,
for reasons we will go into later. Because of this, we will represent IP’s as being dominated by
16As a rule of thumb, you will have the same number of clauses (IP’s) as you have number of verbs in a sentence.
17Note that the idea that these additional clauses are subordinate or embedded is supported by the fact that even
though the whether/if clauses are questions, the sentence as a whole is not a question. Specifically, the force of the
embedded clauses has no effect on the force of the main clause. Correspondingly, ‘Will John think that Sally has
bought a car?’ is a question even though the embedded clause is a declarative, not an interrogative.
92 Chapter 5. Syntax

CP’s from now on, regardless of whether the CP has an overt C such as ‘that,’ or not (‘John may
think Sally has bought a car.’ has no overt C in either CP, for example). Our final clause structure,
then, is outlined below, using sentence (5.26a) as a model.

(5.27) CP

C’

C IP
null
NP I’

N’
I VP
N may
V’
John

V CP
think
C’

C IP
that
NP I’

N’ I VP
N has
V’
Sally
V NP
bought
DP N’

D’ N
car
D
a

We have not seen examples of every possible syntactic category. Additional categories may
be introduced where relevant. And although all of the discussion has so far been based on En-
glish sentences, we assure you that we see the same categories used as the syntactic building
blocks of all human languages (and we will provide at least a few cross-linguistic examples in the
upcoming sections). As with phonology and morphology, when we find commonalities across
human languages that cannot be due to simple exposure to the data during acquisition, we at-
tribute those properties to UG, the innate endowment.
5.6. Syntactic Operations 93

5.6 Syntactic Operations

Up to this point, we have some evidence that suggests that the syntactic component is sensitive
to categories and uses items from these categories as building blocks for hierarchical arrang-
ments of constituents. The interpretation of content and function of lexical items plus the rela-
tionships established between constituents will be identified with an interpretation or meaning
for the string. So are we done? (Don’t get overly excited – you didn’t really think the chap-
ter would be this short, did you?) The only way to test and see whether we’re done is to keep
throwing new sentences at the framework we currently have and see whether the framework
can account for every sentence. If it can, we’re done. If not, we still have work to do.

5.6.1 Yes/No Questions

Cross-linguistically, we see only a couple of question types (in terms of information asked for).
There are questions which ask for additional (new) information such as:

(5.28) a. When did John arrive?

b. Who sat on the bee?
c. Where was the rhino going?
d. How many kumquats did you eat?

These questions are all asking for specific pieces of new information regarding the timing, play-
ers, or other setting information for a particular predication. Crucially, a felicitous answer must
contain the appropriate information. If, instead of “this morning,” you answer “yes” to (5.28a), it
will be considered deviant in the extreme.18 These questions – traditionally called WH questions
in spite of the many and obvious difficulties with the term – will be discussed at some length in
the next chapter.
Contrasting with the set of WH questions, are questions like those below.

(5.29) a. Will Sally go to school tomorrow? (cf. Sally will go to school tomorrow.)
b. Has Dan seen that movie? (cf. Dan has seen that movie.)
c. Should Irma sing that song? (cf. Irma should sing that song.)
d. Can Horace speak Old Church Slavonic? (cf. Horace can speak Old Church
Slavonic.)

Here, the only felicitous answers are ‘yes’ or ‘no’. If you try to give some information instead, it is
completely unacceptable. Correspondingly, these sorts of questions are called ‘yes/no’ questions
and although our framework can generate them, it cannot generate both the yes/no questions
and their declarative counterparts. Why not?
18Note that no one had to tell you this at any time during the acquisition process. Nor did anyone tell you that
there was only a very limited set of question types and what those types were.
94 Chapter 5. Syntax

The answer to this rests in part on an assumption about how the linguistic computational
system works. As with your average computer, the output is determined completely and irre-
vocably by the input. You cannot input X, Y, and Z and have it come out as A, B, and C on one
occasion, but on another occasion come out as D, E, and F. For the yes/no questions and their
declarative counterparts, this means that we cannot have the same input for both question and
declarative, even though it looks like we have the same lexical items (and therefore categories, at
least superficially). This is mildly distressing since one feels that question and declarative share
a great deal in common – i.e., everything except the question/non-question part – and should
therefore be the same. It does not seem as though the difference in placement of the ‘I’ category
items (will/has/should/can), before or after the subject, causes any difference in their interpreta-
tion, ‘will’ still gives us a future; ‘has’ a past; and so on. Moreover, and mighty strangely, the only
‘I’ that inflects for person and number (‘has’) still shows the same inflectional agreement with
the subject in the question as it does in the declarative sentence. We’re in a bit of a quandary
now because the inputs have to be different but there are other bits of evidence that say they
should be the same. If we focus on the ‘make them different,’ we might propose that two IP’s
can be stacked on top of one another in the case of questions and that it is the differentiating
presence of the higher IP that results in a different output type (question).

(5.30) IP

I’

I IP
has
NP I’
Dan
I VP
(null)
V’

V NP
seen that movie

Such a structure, while representing the difference in the order of constituents in the question,
would also open up many new and unpleasant options. For example, this structure suggests
that it would be perfectly grammatical to have two ‘I’ category items in the string. However, if
we plug in another ‘I’, we get a horribly ungrammatical sentence: *Has Dan has/can/will/should
seen that movie. It also gives rise to the possibility of two entirely separate subjects but no way
of determining which subject which ‘I’ should agree with. There are a number of other problems
that such a solution would give rise to, too many for us to reasonably adopt it.
As far as other possibilities go, we might consider that while there is no reason a priori that
the syntactic component should work like the phonological and morphological components, it
would not be surprising to find that it did. Following that route, we might consider modifying
5.6. Syntactic Operations 95

our framework to include two levels of representation mediated by some type of computational
device. In this way, we may be able to satisfy what appear to be the conflicting demands of ‘make
it different/same’. We could start with the idea that there is a base structure, generated just as de-
scribed in Section 5.5. However, we would now introduce a computational component, as well.
This component would take the base-generated form as its input and would produce a ‘compiled’
form as its output, whose representation would be sent on for pronunciation, equivalent to the
‘phonetic form’ in phonology, but called s-structure in syntax.19 If we want this type of system to
work on our question/declarative example, we need to invoke an operation/transformation that
will move the ‘I’ to a spot higher than the subject only when the force of the sentence is interrogative.
Let’s identify the landing spot for the ‘I’ first.
We asserted earlier that CP’s dominate IP’s, even in main (matrix) clauses that have no ob-
vious lexical items filling either the C position or the specifier of CP position. Here is a case
where we might offer some evidence for that assertion, as without the positions available in CP,
we would have no landing site for the ‘I’. However, CP has two possible positions, the specifier
and the head, so we will need to figure out which of these positions the ‘I’ moves to. The WH
question sentences can actually help us out with this. Note that a question like ‘Where was the
rhino going?’ has not only a WH word at the front but also the ‘I’ has moved ahead of the subject
(‘the rhino’), just as it does in a yes/no question. For this WH question, then, we will need two
positions higher in the tree than the subject, one for the WH word and one for the ‘I’ (‘was’).
Conveniently, we have the specifier of CP (Spec,CP) for the WH word and C for ‘was.’20
Finally, let’s consider how we will motivate the movement of the ‘I’ only in cases of an in-
terrogative. Recall that the input to the yes/no question must be different from the input to the
declarative. Once again, we might think about using a device that is used elsewhere, for example
in phonology. The rules of phonology refer to phonological features and we will invoke the no-
tion of feature here in the syntax as well. We already have invoked it implicitly by talking about
categories like N, V, C, and so on. Those can simply be viewed as features, as can the agreement
features that we see on subjects and ‘I’s of person and number. For our yes/no question example,
we are talking about a type of feature that will control the illocutionary force of the utterance –
essentially making it a question, an imperative, a declarative, and so on. Because this is about an
entire sentence, such features should be particular to C-category items since CP dominates the
clause (IP). Given that, we might expect that in our lexicon we have a C with a [+Q] feature, a C
with a [+Dec] feature, and so on. As with many functional categories, there will be members of
the category that are phonologically null and others that have phonological content. Consider
the embedded indirect question in ‘John asked if Mary had a dog.’ Under the present analysis,
the C, ‘if’, will have a [+Q] feature, making the embedded clause a question but the matrix clause
C will be [+Decl], making the overall sentence a declarative.21
19Similarly, the computations in syntax have their own term, transformations, because they ‘transform’ the base-
generated structure in various ways.
20There is additional support for these particular positions as landing sites for WH words and ‘I’s’, including what
seems to be a general property of the syntax that keeps heads and phrasal projections (XP’s) in their own ‘channels’.
Heads move only to other head positions and XP’s move only to XP-type positions.
21The names of the features are just convenient labels for us to use, just as the terms ‘noun’ and ‘verb’ are con-
96 Chapter 5. Syntax

All we need now is to make the ‘I’ want to move to C when the C has a [+Q] feature. The [+Q]
feature on C alone is not enough because it turns out that only some ‘I’s move to C[+Q], specifi-
cally, I’s of matrix clauses. In an embedded clause, no such movement is present (cf. the gram-
matical, ‘John asked if Mary was going to school.’ to the ungrammatical, *‘John asked if was Mary
going to school.’) The most straightforward solution seems to be to have a corresponding feature
on the ‘I’ itself – an I[+Q] – and to make the movement a process of ‘feature matching/checking’
where the C and I need to get together to match/check their [+Q] features. Moving I to C is one
way of getting them together to accomplish this.
The two trees below outline graphically the discussion above. Both trees show s-structure
(end of syntactic derivation) representations. A transformation that involves movement can be
denoted in several different ways, here we indicate the base position of the constituent with a
co-indexed t for ‘trace’. Arrows indicating movement from the base position to the landing site
are often used in addition.22

(5.31) CP

C’

C[+Decl] IP
(null)
NP I’
Dan
I VP
has
V’

V NP
seen that movie
venient labels.
22Co-indexation, by means of subscripted symbols, indicates identity in reference.
5.7. More Head to Head Movement 97

(5.32) CP

C’

C[+Q] IP
hasi [+Q]
NP I’
Dan
ti VP

V’

V NP
seen that movie

Once again, the crucial differences between the declarative in (5.31) and the question in (5.32)
are the features on C and I. (The internal structure of NP’s has been suppressed for convenience
as it is not directly relevant to the issue at hand.) Because the question transformation involves
the head of a phrase (I) moving to adjoin with the head of another phrase (C), we call this type
of movement head to head movement.
Before you dismiss these as outlandish wanderings – features and category items you never
hear or see, lexical items disappearing from one place and appearing in another place, and so
on – we should add that we do have some fairly satisfying support from various languages for
these hypotheses. For example, in Telugu, a Dravidian language spoken in South India, Yes-No
Questions are formed by adding a special question particle (the [ā] at the end of the sentence).
This particle is in exactly the position of a C (Telugu having the reverse order from English) and
acts in every way like the C[+Q] we posited for English, except this C does have phonological
content. (Any movement of I to C here is difficult to determine because in Telugu I and C are
next to one another typically anyway just due to the linear order.)

(5.33) Ravi guɖiki veɭɭāɖu / Ravi guɖiki veɭɭāɖā

Ravi temple-to went / Ravi temple-to went-Q
‘Ravi went to the temple’ / ‘Did Ravi go to the temple?’

There are many more examples of this kind, suggesting that the functional elements we’re propos-
ing here are not only ‘mentally’ real in human language but can also be found in overt phono-
logical form.

5.7 More Head to Head Movement

English is not the only language which has this type of transformation for question formation.
The example from French below reveals a very similar process.
98 Chapter 5. Syntax

(5.34) Il a mangé du pain / A-t-il mangé du pain?

he has eaten some bread / has-he eaten some bread
‘He has eaten some bread’ / ‘Has he eaten some bread?’

Unlike English, however, French also has yes/no questions where the V itself appears to be fronted
to C, such as in the sentences below.23

(5.35) Vous allez à l’église. Allez-vous à l’église?

you(pl) go to the-church/go-you(pl) to the-church
‘You are going to church./Are you going to church?

The corresponding English sentences are severely ungrammatical as the examples below indi-
cate (although they would have sounded OK about 500 years ago).

(5.36) a. *Go you to church?

b. *Sees he the car?
c. *Walk they the dog?

On one hand, it looks as though French questions are formed via I to C movement, as in English.
On the other hand, it looks like French questions are formed through some V to C movement in-
stead. Ideally, it would be nice to have a single analysis of French yes/no questions. It would be
even nicer if we could collapse yes/no questions in French and English together under a single
analysis. What does ‘nicer’ have to do with anything, though? From a scientific perspective, it is
preferable to have a unitary analysis, if at all possible, of phenomena that share many common
properties. It is also much more likely that we will find overarching patterns if we try to make the
broadest generalizations possible. Finally, and perhaps most importantly, in the ‘big picture’ of
linguistic theory, there is a well-supported claim, based on both acquisition and an examination
of empirical facts, that human languages are really very like one another. Moreover, we believe
that the ‘apparent’ differences, which are essentially all differences in lexical items, are super-
ficial and rather insignificant compared to the underlying structural similarities. The ease and
speed with which children acquire language, with no overt instruction, strongly suggests that
UG provides a substantial foundation for all language and therefore supports our claim. If we
approach the human knowledge of language from this perspective, we would have as our first
hypothesis, that French and English are ‘the same’ in some fundamental, structural sense. All
of this means that when we see common behavior either within French or across French and
English, we first look for a common source of that behavior.
There is another difference between French and English that looks like it will be relevant to
our analysis. Compare the grammaticality of the French and English sentences below.

(5.37) Il mange souvent des poires. / *Il souvent mange des poires.
he eats often Det. pears / He often eats Det. pears
‘He eats pears often’ / ‘He eats pears often’
23This is a slightly more formal/archaic way of making yes/no questions than generally found in casual speech.
5.7. More Head to Head Movement 99

(5.38) He often eats pears/*He eats often pears.

French/English bilinguals treat the grammatical French and English sentences above as meaning
exactly the same thing, in spite of the slight difference in word order. But what’s up with that
word order difference? Can it be related to our yes/no question issues? Consider the additional
examples below.
(5.39) *Il a mangé souvent des poires. / Il a souvent mangé des poires.
he has eaten often Det. pears / He has often eaten Det. pears
‘He eats pears often’ / ‘He eats pears often’
(5.40) He has often eaten pears/*He has eaten often pears.
Based on the differences in grammaticality, adverb placement in French is just like adverb place-
ment in English when there is an overt I (‘has’ in this case). Similarly, we noted earlier that yes/no
questions in French were just like yes/no questions in English when there was an overt I. Unlike
French, English never allows the adverb to appear between the verb and its complement in the
surface string and English never allows the verb to move to C in yes/no questions.
One more bit of data from negation should allow us to come up with a reasonable hypoth-
esis. We assume that negation, its own head, is fixed in the order it comes in the hieararchical
structure between I’ and VP. Moreover, that the head of negation in French is the ‘pas’ part of ‘ne
pas’.
(5.41) Il ne mange pas des poires. / *Il ne pas mange des poires.
he eats not Det. pears / He not eats Det. pears
‘He doesn’t eat pears’ / ‘He doesn’t eat pears’
(5.42) *Il n’a mangé pas des poires. / Il n’a pas mangé des poires.
he has eaten not Det. pears / He has not eaten Det. pears
‘He hasn’t eaten pears’ / ‘He hasn’t eaten pears’
Oddly, the examples with no overt I are fine with the verb preceding the negation head. This sug-
gests that either the verb has moved or the negation head has moved but since we see no general
evidence cross-linguistically that negation ever moves, it seems correct to rule such movement
out. That leaves verb movement.
The wrap-up here can now be brief. We have some evidence that verbs raise to I in French
when there is no overt I. This is another example of head movement. A V to I analysis would
account for not only the negation and adverb placement facts but also for the yes/no question
differences. On those occasions in French when there is no overt I, there is V movement to I, and
consequently yes/no questions will show the verb in C. Specifically, the first move is V to I and
the second move (question transformation) is I to C. Whatever is in I moves to C. If I is only an
auxiliary or modal, then the auxiliary or modal moves. If, on the other hand, the V has already
moved to I, then the V moves further to C. The happy result of such an analysis is that question
formation within French is uniform and question formation across French and English is also
uniform (and across many more languages).
100 Chapter 5. Syntax

5.8 Structures and Their Interpretations

We saw in Chapter 4 that morphologically complex words may be structurally ambiguous – have
more than one interpretation due to more than one possible structure. (Remember that that
ambiguity only exists for the listener – the speaker knows which structure he or she generated.)
Sentences can be structurally ambiguous in virtually the same way except that the constituents
involved are syntactic categories, not morphemes. Consider the following sentence.

(5.43) The zookeeper poked the gorilla with the banana.

This sentence is ambiguous – it has two possible interpretations. One of the interpretations is
that the zookeeper is using a banana as an instrument to poke the gorilla. The other interpreta-
tion is that the gorilla is holding a banana and the zookeeper is poking the gorilla (but we do not
know with what – his finger, a different banana, whatever …). Critically, you only had to read
or hear the sentence once (and not twice) to get these two interpretations. We can conclude,
therefore, that nothing about the sentence changes to allow a different interpretation from one
point in time to the next. On the other hand, there are many sentences, some much longer than
this one, which are not ambiguous, such as:

(a) The gorilla with the banana thought that the zookeeper poked him.

(b) The zookeeper with the banana poked the gorilla.

(c) The zookeeper said that the gorilla was poked with the banana.

How do we account for the fact that the original ‘zookeeper’ sentence is ambiguous while the
sentences immediately above, very like it in their choice of lexical items, are not?
Somehow, the ambiguous sentence must present the listener with choices that the other
sentences do not. It does not look like those extra choices are in the domain of the lexicon.
Sentence (b) above has the same lexical items in it but is unambiguous. (We will assume that
the phonologically null items are the same for the purpose of our discussion here.) The only
other possibility is that is has to do with how the structure can be/was built. If we consider what
kind of structure must be there to get the two interpretations of the original sentence, we see
that the critical difference is at what point the PP ‘with the banana’ is merged in the tree. If it is
merged with the N ‘gorilla’ (Tree A below), then it is interpreted as being a modifier of ‘gorilla’. If,
instead, it is merged with the V’ ‘poked the gorilla’ (Tree B below) is is interpreted as modifying
the ‘poking the gorilla’ activity as an instrument.
5.8. Structures and Their Interpretations 101

Tree A (VP only):

VP

V’

V NP
poked
D N’
the
N PP
gorilla
P’

P NP
with
D N’
the
N
banana

Tree B (VP only):

VP

V’

V’ PP

V NP P’
poked
D N’ P NP
the with
N D N’
gorilla the
N
banana

The surface order of words in the original, ambiguous sentence is the same regardless of whether
the PP has merged into the NP or the VP. The closely related sentence in (b) above presents the
listener with the same lexical items but with a surface order that does not allow them to come
up with two separate hypotheses about the location of the PP. Cases of structural ambiguity il-
lustrate the importance of understanding that trees indicate relationships between the elements
of a sentence that are crucial to the interpretation.
102 Chapter 5. Syntax

5.9 Summary
We have seen that, in a general sense, the syntactic module of the grammar is similar to the
phonological and morphological modules. In all three components, there is a more abstract
level of representation (d-structure for syntax), a computational component (where, in syntax,
transformations apply), and a less abstract representation (s-structure for syntax). A d-structure
representation is formed by taking elements from the lexicon and combining them in confor-
mance to X’ Theory, thus establishing certain types of constituents with particular relationships
to one another. The interpretation of the string is based on both the lexical items and the re-
lationships established between constituents. The d-structure representation is input to the
computational component where transformations may be triggered by particular elements in
the string. The computational system produces an output (the s-structure). A schematic of this
is given below.

Our model of human syntactic knowledge must capture speakers’ intuitions about constituency,
ambiguity and relatedness, as well as phenomena we have observed such as recursiveness and
creativity. As always, we look for the most economical representations and operations possible
which still account for the relevant data.
5.10. Exercises 103

5.10 Exercises
Questions (1-3) are based on the following sentences:

• Go away!

• Put the cat down!

• Stop shouting!

1. Sentences such as those above are:

(a) no problem – they are in line with everything we’ve said in this chapter about sen-
tence structure.
(b) a bit of a problem – they appear to be missing the subject NP and therefore a critical
part of the sentence is missing
(c) an insurmountable problem – they must be Martian sentences and we shouldn’t try
to account for them
(d) a bit of a problem – they have exclamation points.

2. Based on the goals of linguistics as well as general scientific principles, the first hypothesis
should be

(a) they are completely separate from other sentences and should have a completely
separate analysis.
(b) they are completely separate from other sentences and we do not need to provide
an analysis for them because they are not normal.
(c) they are, at some level, sentences just like all other sentences and we need to search
for a unified analysis
(d) they are, at some level, sentences just like all other sentences so we do not need to
analyze them further.

3. A reasonable follow-up step to your answer in (2) would be:

(a) Some sentences are base-generated with subjects and some are not.
(b) We will sweep commands under the rug.
(c) Some sentences undergo a transformation that deletes their subject under specific
conditions.
(d) We’ve said that all sentences have subjects and it’s obvious that these sentences have
‘you’ as their subject. We’re done.
104 Chapter 5. Syntax

Questions (4-6) are based on the Spanish data below (informally knows as cases of ‘pro-
drop’):

Voy al centro
go-1sg,pres to-the center
‘I am going to the center.’

Tenemos diez dólares

have-1pl,pres 10 dollars
‘We have 10 dollars.’

Trabajas en el casa
work-2sg,pres in the house
‘You(sg) work in the house’

4. What properties do the Spanish sentences show that are different from the commands in
(1-3)?

(a) They are declaratives

(b) The translations indicate that the subject is understood to be not just ‘you’ but also
‘we’ and ‘I’
(c) The verb forms indicate some subject agreement features, based on the glosses
(d) All of the above.

5. This Spanish data helps with English commands because

(a) now we can have one deletion transformation that covers all of them
(b) it is comforting to know that sometimes people don’t know what the subject is in
Spanish, either
(c) the absence of a surface subject is not limited to commands or to English
(d) if we have an analysis of Spanish, it will definitely work for English

6. What might make it difficult to unify the English command/Spanish pro-drop cases under
a single analysis?

(a) English and Spanish are two separate languages.

(b) English has only a few sentences like this but Spanish has an enormous number of
sentences.
(c) English and Spanish are genetically related.
5.10. Exercises 105

(d) English commands (with only an interpretation of ‘you’ as their subjects) and Span-
ish declaratives (with the ‘normal’ range of subject possibilities) might be hard to
find a single featural motivation for since they do not share a lot in common with
one another.

Questions (7-9) refer to the tree below. (Internal structure of DP’s suppressed for space
reasons.)

CP

C’

C IP

NP I’

N’ I VP
N V’
Harold
V NP
saw
DP N’
the
N PP
dog
P’

P NP
with
DP N’
the
N
binoculars

7. The syntactic relationships in this tree result in an interpretation where:

(a) Harold was wearing binoculars.

(b) Harold was using binoculars to see the dog with
(c) The dog had binoculars.
(d) No one had binoculars.
106 Chapter 5. Syntax

8. The NP ‘the dog with the binoculars’ is:

(a) the specifier of V

(b) the complement of V
(c) the sister of V
(d) both (b) and (c)

9. If the PP ‘with the binoculars’ was joined instead at an additional V’ immediately domi-
nated by the VP, the interpretation would change to

(a) Harold was wearing the binoculars

(b) Harold was using the binoculars to see the dog with
(c) The dog had binoculars
(d) No one had binoculars

10. One of the reasons linguist believe UG provides an extensive foundation of linguistic knowl-
edge is that

(a) human language, a complex system that is still not well understood, is acquired
rapidly and easily by very young children.
(b) lexical items are so similar cross-linguistically
(c) Languages are so completely different from one another.
(d) Children innately know what sounds are in the words of their language.
 Chapter 6
Semantics

Contents
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.2 Lexical Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.3 Sentential Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.3.1 Thematic Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.3.2 Interpretation of Functional Elements . . . . . . . . . . . . . . 121
6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

6.1 Introduction
Until now, we have concentrated mainly upon the computational aspects of the linguistic
system with little regard for the meanings of either words or utterances. However, the
linguistic system is used for communication (even if that communication is only you talk-
ing to yourself) and therefore is capable of sending and receiving messages. Semantics is
the study of meaning and the manner in which meaning is conveyed by the system. It is
important to note that this does not include the study of why individuals choose to convey
a particular meaning on a particular occasion. This ability lies outside the linguistic com-
putational system and falls more properly under the fields of psychology and sociology,
and perhaps several others.
An in-depth discussion of the intricate nature of formal semantics would require back-
ground knowledge of logic, philosophy, and possibly mathematics. Since this is beyond
the scope of an introductory course, we will restrict ourselves to brief discussions of lex-
ical meaning and sentential meaning.

117
118 Chapter 6. Semantics

6.2 Lexical Meaning

The presence of some ‘meaning’ component in lexical entries has been assumed since
Chapter 4 when we defined the morpheme as a ‘meaningful’ unit. In addition to ‘mean-
ing’, meanwhile, we saw in Chapters 4 and 5 that lexical entries must contain certain
information for the computational system to use critically in its operations. Part of this
information is categorial — whether the morpheme is a noun, a verb, an adjective, and
so forth. Another part of this information might be termed ‘functional’ — for example,
one morpheme might have the function of forming plurals, another might function to
form third person singular present tense forms of verbs. Now we will turn back to the
‘meaning’ component of lexical entries – information that is often closely associated with
properties of real world objects such as ‘dogs’ and ‘tables.’ For our purposes, the great-
est distinction appears to be between what linguists frequently term content or open class
morphemes, such as ‘dog’ and ‘table’ and function or closed class morphemes such as the
plural and third person singular present tense markers.1 Because the semantics of con-
tent morphemes is generally easier to discuss (although by no means well understood),
we will concentrate on members of that category.
We assume that the semantic system (like the phonological, morphological, and syn-
tactic systems) contains units that are specific to it. It does not, for example, make ref-
erence to articulatory features as does the phonological system.2 One of the goals of se-
mantic inquiry, then, is to make a model of the type and set of features which the system
operates on. One possibility is that there is a (probably finite) set of semantic features
available to human language. Every morpheme must be assigned some feature or combi-
nation of features from this set. How this might work is illustrated below.

(6.1)

girl boy baby puppy

+human +human +human -human
+female -female
-male +male
+young +young +young +young
-really little -really little +really little +really little
-4 legs -4legs -4legs +4legs

The above example is only meant to illustrate, in principle, how sets of semantic fea-
tures could serve to distinguish morphemes. Note that the particular features above are
1
The open vs. closed class refers to the ability (or lack thereof) to add new items to a category. It is
fairly trivial to borrow a noun from some other language, for example, or to coin a new noun. On the other
hand, it almost never happens that a preposition is borrowed or a new preposition coined.
2
This is not meant to imply that the semantic system cannot ‘see,’ or make reference to, any aspects of
the other systems. There is no doubt important interaction among all the systems involved in grammatical
computations. This issue is independent of whether each system has units particular to it.
6.2. Lexical Meaning 119

almost certainly not good representations of whatever the true semantic features are.
Actual semantic features are likely to be primitives — elements which are not able to
be broken down further into constituent parts. Note that we have actually used inter-
pretable morphemes and even combinations of morphemes (‘human,’ ‘really little’) as
the semantic features to define other morphemes, such as ‘boy’ and ‘baby.’3
In the most transparent cases, when the semantics of a word picks out an entity in the
real world, the word is said to denote that thing. So we might say that ‘dog’ denotes ‘ca-
nine’. There is another, more experience-oriented meaning that words may come to have
called connotative meaning. This meaning appears to vary much more across individuals
and seems to stem from an individual’s experience or pragmatic information about the
entity that the word denotes. So, for example, ‘dog’ may have the connotation for one
person of a ‘dirty, mangy, dangerous creature’ and for another person a ‘loyal, affection-
ate, playful companion’. In spite of the extreme differences in connotation, both people
would agree on the denotation of ‘dog’ as a ‘canine.’
Some of the features we have listed in 6.1 sound a bit implausible, particularly ‘+/-4
legs.’ We could, of course, have used ‘+/-canine’ and sounded a bit more sophisticated. We
deliberately did not do so for two reasons. The first is that it is just as unlikely that an in-
nate semantic feature system uses ‘+/-canine’ as ‘+/-4 legs.’ The ‘canine’ designation was
developed by biologists and zoologists to classify a particular species (Canidae) according
to certain anatomical (in this case, dental) features. It would be most surprising if this
kind of classificatory system was something that humans were innately endowed with,
particularly since it is, in this case, in English (and Latin). In fact, a very salient physical
property like ‘4 legs’ has considerably more appeal as a hypothesis. The second reason
is that the unlikely-sounding ‘+/-4 legs’ highlights the important distinction between a
model and the thing which is being modeled. There is no a priori reason for any part of a
model to either look like, or sound like, the thing it is modeling. Assuming that seman-
tic features (or any other units and operations we posit for the human linguistic system)
bear any resemblance to what is used by the real system itself is analogous to assuming
that numbers fall down because gravitational law is represented by an equation.
In spite of the above difficulties with semantic features, there are still many inter-
esting issues that we can consider. Note that in 6.1 we have not really listed a sufficient
number of features to distinguish a puppy from a kitten (or any other 4-legged ‘infant’).
The true feature system must be extremely detailed in order to capture such differences.
On the other hand, there are certainly some morphemes whose semantic features are far
less detailed. For example, the features for the morpheme ‘male’ (as in ‘It’s a male.’) can-
not be limited by a feature ‘human’ or by any feature which has to do with size or age. It
seems, then, that there may be a superset-subset relationship between semantic feature
bundles. It is also possible that these sets are organized hierarchically and are, therefore,
directly related to one another. If such relationships exist, it is doubtful that they are sim-
ilar to what we normally think of when we think of word relationships. However, people
3
The fact that we typically use other words for features indicates something about our current under-
standing of lexical semantics.
120 Chapter 6. Semantics

still use fairly traditional terms to describe how meanings of words relate to one another,
such as those below.

• synonyms –‘mean the same thing’

• antonyms – ‘opposite in meaning’

• homonyms/homophones – ‘sound the same’ (‘-nym’ also written the same)

• polysemous – ‘have multiple meanings’

• metaphor – ‘analogy’ (roughly)

In talking about how words are related to one another, it is often convenient to use such
terms.
Our traditional classification of word relationships, particularly synonyms, brings up
an interesting question. Can any two morphemes have identical semantic feature bun-
dles? That is, are any two morphemes truly synonymous? We believe that the answer
to this question is ‘no.’ One of the reasons that we do not think true synonymy exists is
because it would introduce a randomness into an operation that appears to be systematic.
Whatever way the human brain translates some ‘intention’ or ‘message’ into a selection
of morphemes and a syntactic representation, it seems as if it must do so based on some
criteria. How would we choose between two morphemes with identical semantic feature
bundles? Some choice must be able to be made on the basis of certain criteria. Two mor-
phemes, both of which met the criteria, would not allow for a differentiating choice.
There is also the challenging question of how humans actually construct semantic
feature bundles and associate them with some sequence of sounds. First of all, it seems
as if this process has no end point. Adding to the lexicon continues throughout one’s life.
Second, one is able to make ‘revisions’ to lexical entries. It frequently happens that you
thought a word had a particular meaning, found out that you were wrong, and subse-
quently altered the semantic features for that entry such that they were at least closer
to what you inferred the word meant from the new information you received about it.
Neither of these aspects appears to be present for adult speakers in the other parts of the
grammar — in the phonology, morphology, or the syntax. As we will discuss in the next
chapters, there appears to be some point at which you are ‘finished’ constructing your
grammar and after that point you cannot go back and revise it. New information received
after that time leads to the construction of an additional grammar rather than modifica-
tions of an existing grammar. The lexicon, then, and the semantic features it contains,
seems a different creature altogether than the computational components of syntax and
phonology, for example. Within the lexicon, though, it seems likely that there will be
a difference between content morphemes and functional morphemes in these respects.
People appear to be ‘finished’ with functional morpheme acquisition when they’ve com-
pleted their construction of the grammar.
6.3. Sentential Meaning 121

6.3 Sentential Meaning

When we interpret the meaning of sentences, we use more than just the semantic content
of individual words. As we saw in the previous chapter, the structural relationships in a
sentence are what tell us which words should be construed together (in a modification
relation, for example) and which words should not. In the two sections below, we will
give a couple of examples of how closely tied together structure and meaning can be.

6.3.1 Thematic Roles

In Chapter 5, we noted that the grammar would only let us build certain types of syntactic
structure and do certain types of operations. If we kept within those limits, however, the
assumption was that our sentences would be grammatical. It turns out that we still have
a few sentences that are troublesome (actually, more than a few...), some of which are
given below.

1a John poked the gorilla with the banana.

1b John imagined the gorilla with the banana.

2a John put the car in the garage.

2b *The boulder put the car in the garage.

2c *John put the car in an hour.

Let’s consider (1a-b) first. We already discussed sentences like (1a) and saw that, from
the listener’s perspective, such sentences were structurally ambiguous (it was possible to
assign two different syntactic structures to a single string). The problem arises in (1b).
We predict that (1b) should also be ambiguous but it is not. The only interpretation is that
there is a gorilla holding a banana. How could we have ‘lost’ an interpretation? At first
glance, it seem like the root of the problem must lie in changing the verb to ‘imagined.’
But remember that we said the syntax was ‘blind’ to the content meaning of words and
saw only their syntactic categories. Since both ‘imagined’ and ‘poked’ are V’s, it’s going
to be difficult to support a syntactic difference between the two. Most people would prob-
ably explain this anomaly by saying something like ‘Well you can poke something using
a banana but you can’t imagine something using a banana.’ This is really kind of the right
explanation but it would be nice if we could quantify it a little. Before we try to do this,
let’s look at the (2) examples.
The sentences in (2) differ, on the surface, fairly dramatically from the sentences in
(1), even to the point that some of them are ungrammatical. Very few changes seem to
have inspired that ungrammaticality, though. The difference between (2a) and (2b) is
122 Chapter 6. Semantics

only in the subject, ‘John’ vs. ‘the boulder.’4 In (2c), the only change is in the object of
the preposition (‘an hour’ vs. ‘the garage’). As with (1), it’s difficult to see how the syntax
can be responsible for the ungrammaticality since the syntactic categories are identical
in all (2) sentences. But peoples’ explanations for the ungrammaticality in (2b-c) have a
similar ring to the explanation for (1), for example, ‘Boulders can’t drive.’ and ‘You have
to make the car go into a place.’
It looks like there’s some additional factor here that we haven’t taken into account
yet. For the moment, let’s say it revolves around verbs, since that seems to be the locus
in our examples. One of the things we appear to know unconsciously is that the action
of some verbs requires different ‘players’ than the action of other verbs and that these
players must be present to complete the verbal event or predication. We describe these
as the thematic roles that are intrinsic to the semantics of the verb and we say that verbs
have thematic grids. An abbreviated list of thematic roles is below.5

• agent/actor – the one performing the action of the verb

• experiencer – the one ‘undergoing/feeling’ the action of the verb

• patient/theme – the one that the action of the verb is directed toward (‘patient’ if
affected/changed by it, ‘theme’ if not)

• instrument – the object used to perform an action

• location – the place where the action occurs

Subjects of sentences are usually agents or experiencers (the subject of ‘The girl feels
happy’ being an example of the latter). You can see some of the connection between
verb semantics and thematic roles if you compare ‘poke with a banana’ and ‘imagine with
a banana.’ The verb ‘imagine’ neither requires nor permits an instrument whereas the
verb ‘poke’ actually requires an instrument.6 This is the explanation for the the lack of
ambiguity in (1b). It is not a syntactic issue, per se, but is based rather on the listener’s
knowledge of the thematic roles associated with particular verbs, a semantically-based
distinction.
The (2) examples can now also be explained. The thematic grid for ‘put’ requires an
agent, a theme, and a location. Not all NP’s can easily be agents, though, since agents have
to be able to act/move. ‘Boulders’ are inanimate and therefore do not fit the profile for
an agent thematic role (or for an experiencer one, for that matter). So (2b) is a problem
because of a mismatch between thematic role and semantic properties of the NP filling it.
(2c), on the other hand, is a simple violation of the thematic requirements of ‘put.’ In this
4
Note that proper vs. common noun and presence/absence of ‘the’ cannot be the cause of this ungram-
maticality since ‘the woman’ works perfectly well.
5
‘Action’ is being used as a cover term here.
6
If there is no overt NP for the instrument, an instrument is still present at an abstract level and in-
cluded in the interpretation
6.3. Sentential Meaning 123

case, it is that there is no ‘location NP’. (This would also be true of a sentence like ‘*John
put the car.’) The presence of a ‘time NP’ has no bearing on the grammaticality as ‘John
put the car in the garage in an hour’ shows.
The interaction of thematic requirements (an essentially semantic property but one
which is not limited to verbs) with the syntax is a complex one. It may be considered
part of the ‘syntax-semantics interface’ along with the interpretation of reference and a
number of other topics.

6.3.2 Interpretation of Functional Elements

If it is possible that no two individuals interpret the same content words in exactly the
same way, is it possible that what appear to be the same sentences are also interpreted
differently by different people? If the latter were true, it would seriously compound
the difficulties in interpreting messages. In our model of syntactic structure, we as-
sume that there are a number of syntactic categories and relationships that are language-
independent — only trivial differences exist between surface word order across languages.
We have made basically the same hypotheses for other modules of the grammar (phonol-
ogy and morphology), that is, that there exists a set of universal features and operations
available to human language. Our discussion of the semantic features for content words
follows these assumptions in that we have posited a universal set of features. However,
there is no guarantee that one individual’s grammar will have the same semantic feature
bundle as another individual’s grammar for a particular lexical item. We would still like
to believe, though, that the semantic component of the grammar processes input in a
systematic fashion, as do the other components.
Note that semantic interpretations must involve more than just adding together the
features of a set of content morphemes, otherwise strings like ‘boy dog big tall’ would
be perfectly acceptable (i.e. interpretable) semantically. Additional information includ-
ing which properties are being assigned to which entities and the force of the utterance
(statement, question, command) must be supplied from other sources. Moreover, notions
such as ‘who’s doing what to whom’ and whether the utterance is a question or a state-
ment appear to be somewhat independent of the interpretation of the content words in
an utterance. Consider the following examples.

(6.2) a. The glor strued the zoobie at the blag

b. What zimz kaddle ording un ploget
c. Jore hof ap reeg trudle egdil

While 6.2 (a) and (b) are not stellar examples of clarity, they can be assigned some par-
tial meaning. 6.2 (a) has the benefit of having what appears to be perfectly fine syntactic
structure, which may increase the feeling that it is almost interpretable. One feels that if
one only knew what a ‘glor,’ a ‘zoobie,’ and a ‘blag’ were and what sort of action ‘struing’
was, then the sentence would be perfectly clear. The syntactic structure of 6.2 (b) is not
124 Chapter 6. Semantics

at all obvious. What is obvious about (b) is that it begins with ‘What’ and as soon as you
read or hear ‘what,’ you are fairly certain that this is a question of some kind. In 6.2 (c),
on the other hand, it is clear that we have just made up a bunch of stupid-sounding words
and tossed them together on the page (no easy task...). There is no sense in (c) of even
minimal interpretation possibilities.
We are going to discuss rather briefly the nature of words like ‘what’ and the kind of
role such elements may play in the interpretation of utterances. Our basic assumptions
will be the following. The first is that the semantic component will produce, for a given
set of features, the same output (interpretation) every time and that this is true not only
within individual grammars but for all grammars. The second is that there is some set of
semantic units which are functional elements that all humans possess innately, some or
all of which are active in all adult grammars. For instance, all languages have questions,
as well as statements, and we assume that there is some semantic feature or features
(possibly very abstract) which mark these two types of utterances for the different in-
terpretations. Thirdly, we believe it is possible, in principle, to separate the process of
interpreting the formal properties of utterances based on functional elements from the
semantic features of the content words involved in those utterances. (This was, at least
in part, the exercise involved in 6.2).
We are hopeful that examining WH words will help to illustrate the points we have
made above. In English, WH words are all of the interrogative pronouns, most of which
begin with the sequence <wh...> and one of which begins with just <h>, i.e., <how>.7
All languages have some such set of morphemes (not necessarily words) and we conve-
niently refer to all such sets as WH words, independent of the language in question (and
all phonological evidence to the contrary). There is a noticeable pattern in the place-
ment of WH words in English — if there is only one WH word, it always occurs first in the
sentence.8

(6.3) a. What did the child see?

b. Who did the child see?
c. Where did the child go?
d. Why did the child go?
e. How did the child fall?
f. Which child did Susie see?
g. *Did the child see what?
h. *Did the child see who?
7
There are other words which look and sound just like these but do not have the same function or
distribution, such as the set of relative pronouns. We are only interested in the interrogative pronouns
here.
8
The actual data on WH questions is far more complex than these examples suggest but this subset will
be sufficient to illustrate our point.
6.3. Sentential Meaning 125

i. *Did the child go where?

j. *Did the child go why?
k. *Did the child go how?
l. *Child did Susie see which?

You may be able to think of apparent counterexamples to the first set (3a-f) above.
(The ungrammatical examples in 6.3 are there for comparison purposes only and we won’t
refer to them again.) For instance, it is perfectly grammatical to say ‘The child saw what?’
This type of question is known as an ‘echo question.’ Echo questions have a different
interpretation than the acceptable types of WH questions in 6.3 and are used in different
contexts. Note that ‘What did the child see?’ is a request for new information in the sense
that the listener has not yet heard what it is that the child saw. On the other hand, the
echo question ‘The child saw what?’ is a request to repeat information that has already
been presented but which, for some reason (usually either mishearing or disbelief), the
listener did not process.

(6.4) • The child saw a Martian.

• The child saw what?

Echo questions also have a different intonational pattern than their simple WH counter-
parts. The placement of the WH word in echo questions is relevant to our discussion, as
well, and we will come back to it later on. For the moment, we will concentrate on the
simple WH questions in 6.3 .
It is mildly curious that all of the Wh words in 6.3 seem to have to be at the front of
the sentence. (We know that it is unlikely to be due to some phonological property of
the sound sequences of those words because they can occur elsewhere, as echo questions
show.) It is even more curious when we look at WH questions in other languages. Many
languages have WH questions which are like those in the Telugu examples, below.

(6.5) Pilla ēmiʈi ʧūsindi

child what saw
‘What did the child see?

(6.6) Pilla ēvaru ʧūsindi

child who saw
‘Who did the child see?’

In Telugu, the WH words do not all occur at the front of the simple question. In fact,
their distribution is instead just like the WH words in echo questions in English, allowing
for the difference in Telugu word order. Critically, however, the interpretation of the
126 Chapter 6. Semantics

Telugu simple questions above is not as echo questions but is identical to the English ‘What
did the child see?’ and ‘Who did the child see?’, respectively.9
The critical question to ask at this point is whether we are counting apples and or-
anges (i.e. things that cannot be counted together because they are completely unre-
lated to one another) or whether we have, at some very abstract level, essentially an
utterance with the same function and force in both English and Telugu. Certain pieces
of information may be relevant in answering this question. First, the similarity between
the placement of the WH word in English echo questions and Telugu WH questions. Sec-
ond, children at the acquisition stage in an English environment do not produce strings
such as 6.3 (g-l), nor any strings which have the WH word other than at the front or in
the echo question position. Third, speakers of Telugu (and languages like Telugu with
respect to WH questions) find the ‘switch’ to English WH questions completely trivial.
Fourth, Telugu-English bilinguals accept the Telugu questions and their English equiva-
lents in 6.5 as being ‘the same.’
This is hardly a scientific approach to the question. The information that we have
asked you to consider in the preceding paragraph is largely anecdotal and not substanti-
ated with empirical evidence. However, a more careful consideration is well beyond the
scope of an introductory text. As it is, we can only say that all of these things are sug-
gestive of a hypothesis which treats the simple English WH questions and their Telugu
equivalents as being the same at some level of interpretation. If we adopt such a hypoth-
esis, we must still account for the fact that the surface forms are different–somehow we
have to make two strings which appear different be the same.
If you recall, in our discussion of allophones and phonemes in Chapter 3, we stated
that allophones are just surface representations of an underlying mental representation,
the phoneme. The case of WH words might benefit from a similar (though not identical)
analysis. One possibility is that the features of a WH word such as ‘what’ are relevant
to two aspects of interpretation. One of these is that ‘what’ is a ‘stand-in’ for some NP.
In this sense, ‘what’ is ‘NP-like’ and needs to be in a position that NP’s can be in. This
requirement would be satisfied by the position of ‘what’ in the Telugu sentence or in the
English echo question case. In both of these cases, ‘what’ can be replaced by an NP, such
as ‘the dog’. At the same time, however, ‘what’ appears to have a feature that causes an
utterance to be a question instead of a declarative. Note that the question feature seems
to affect the entire utterance, not just one word. This can be compared to the English
echo question where the question feature of ‘what’ does seem to affect just one word. It is
possible that, in order to have an effect over the entire utterance, ‘what’ may need to be
at the beginning (or perhaps at the very end) rather than in the middle somewhere. From
a syntactic point of view, this would place it ‘over’ the rest of the sentence in hierarchical
structure terms. It seems, then, as if ‘what’ needs to be in two places at once or perhaps
have two copies, one for each position. If the distinction between English and Telugu is
truly only a superficial one, then it might simply be that both languages require a copy
9
Telugu does have echo questions but they are indicated by a postposition which is not present in these
examples.
6.4. Summary 127

of ‘what’ at the beginning and both require a copy of ‘what’ in an NP position and English
pronounces the ‘what’ at the beginning whereas Telugu pronounces the ‘what’ in the NP
position. At an abstract, non-phonetic level then, English and Telugu will be the same
(and, by hypothesis, any case of a ‘what’ question in any other language as well), it is only
a matter of which ‘phonetic’ choice is made (based on some assumption that phonetic
redundancy is avoided) that causes them to appear different.
Note that the task of acquiring language will be much less difficult under a scenario
where this type of variation across languages is just a superficial property. This is partic-
ularly true of the components of the grammar which are inherently more abstract, like
syntax and semantics. Correspondingly, one might assume that these more abstract com-
ponents are innately more specified than other components for the simple reason that it
would otherwise be considerably more difficult to deduce their operations from surface
structure.
The main point of this discussion of WH words cross-linguistically is to show an aspect
of sentential interpretation that takes place at the level of the functional elements of a
sentence.

6.4 Summary
The level at which meaning exists in language is undoubtedly the most abstract and thus
the least amenable to an empirical approach. Our very brief discussion of it here is in-
tended primarily to offer support for the view that universal properties of language (UG)
exist in semantics as they do in the other parts of the grammar that we have discussed.
In addition, we sought to show that semantic units (in particular ‘functional’ units) and
computations are a critical part of constructing and interpreting sentences.
128 Chapter 6. Semantics

6.5 Exercises
1. When we say that a word ‘denotes’ something, we mean that:

a) the word has come to be associated with a certain set of properties drawn
specifically from our own personal experiences
b) the word picks out an entity in the physical world (roughly)
c) the word can be broken down into smaller parts
d) the word has several homophones

2. Thematic roles

a) ‘complete’ the action/event semantics of the verb.

b) are inherent in syntactic categories
c) are inherent to the semantic features of individual lexical items
d) ‘complete’ the word status of lexical items.

3. Sentential interpretation requires:

a) interpreting the semantics of individual lexical items only

b) interpreting the structural relationships between syntactic categories present
only
c) interpreting the lexical and structural properties of the entire string
d) interpreting the thematic roles present only.
 Chapter 7
Sociolinguistics

Contents
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.2 Traditional Dialectology . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.3 Problems with Traditional Dialectology . . . . . . . . . . . . . . . . . . 131
7.4 The Sociolinguistic Revolution . . . . . . . . . . . . . . . . . . . . . . . 133
7.5 Standard and Non-Standard . . . . . . . . . . . . . . . . . . . . . . . . 136
7.6 Gender and Linguistic Variation . . . . . . . . . . . . . . . . . . . . . . 137
7.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

7.1 Introduction
Up to this point in the course, we have been concerned almost exclusively with the nature
and structure of the grammar — the linguistic ‘knowledge’ that resides in the mind/brain
of an individual speaker. We have discussed the various types of knowledge that appear
to be involved (phonological, morphological, syntactic, etc.) and posited ‘modules’ of the
grammar to account for how these different aspects of linguistic knowledge appear to
function. For the rest of the course we will focus on several interesting questions which
relate to these individual knowledge states, including:

• Why, if such knowledge is a property of individuals, do people from the same area,
class, gender, ethnicity, etc. seem more similar to one another than to others with
respect to this knowledge? How does it come about that this is the case?

129
130 Chapter 7. Sociolinguistics

• Why, given that such knowledge is a property of individuals, does it appear (from
our historical records) that people who produced output using English-type gram-
mars five hundred or a thousand years ago had such different knowledge states
than those who have English-type grammars do today? Why does no one speak
‘Old English’ or ‘Middle English’ today? What do these terms mean?

• What does it mean to say that two ‘languages’ are related? Does it mean anything,
now that we recognize that ‘language’ in the traditional sense is not a very mean-
ingful scientific concept?

• How did the knowledge state that an individual has in his or her brain/mind get
there? What is the learning mechanism? How much of what we know, when we
know a language, is learned and how much is part of our genetic inheritance as
humans?

The first set of questions forms the central focus of the field known as ‘sociolinguistics,’
which will be our primary concern in this chapter. The second set of questions forms the
core of the study of historical, or diachronic, linguistics which will be treated in the first
part of Chapter 8. The third set of questions, those regarding linguistic classification, are
the subject matter of comparative linguistics and will be dealt with in the second part of
Chapter 8. Finally, the fourth set of questions represents the central focus of the field of
language acquisition. The study of language acquisition allows us to see what contribu-
tion the field of linguistics can make to the important general philosophical debate on the
so-called ‘nature’ vs. ‘nurture’ issue. We hope it will become clear to you in the course of
your reading that the sets of questions given above are in fact intimitely related to one
another — in spite of their apparent diversity.

7.2 Traditional Dialectology

As any of you can doubtless attest from your own experience, the precise form of in-
dividual grammars is not uniform — different people speak differently, in part because
their grammars are not the same. It is of some interest to note that the differences that
we observe in the way people speak are not randomly distributed across the population.
Instead, speakers from similar geographic areas, for example, tend to speak in a similar
manner. At the same time, however, speakers from roughly the same geographic area
(e.g., Boston) may still sound quite different from one another. These differences often
depend upon such factors as social ‘class,’ age, ethnicity, gender, religion, occupation, etc.
The study of the relationship between ‘social’ factors, broadly construed, and linguistic
variables (for example, whether or not one pronounces the [r] of words like ‘farm’) is
known as sociolinguistics. The study of the relationship between geography and linguistic
variables is generally known as dialectology. In this chapter we will consider the nature
and major results of these two disciplines.
7.2. Traditional Dialectology 131

The fact that people spoke somewhat differently in different places was apparent to
linguists (and probably to everyone else who thought about it for a moment) from early
on. In the 19th century the primary focus of linguistic research was historical linguistics —
i.e., the study of the relationships between languages and how these languages came to
look the way they did. ‘Language’ was of course not interpreted in the modern theoreti-
cal sense (in which ‘language’ equals ‘grammar’) and thus a certain amount of effort was
invested in determining the answer to questions such as ‘Is Dutch a different language
from German, or a dialect of the same language?’ We recognize the futile nature of such
questions today, but in an attempt to deal with such matters in the 19th century an in-
vestigation of geographical variation in the relatively local sense — i.e., variation within a
single ‘language’ (socio-politically conceived) — was begun. It was quickly discovered that
the study of dialects in use in particular localities was of considerable interest to linguists’
primary questions of the day (remember, these were historical questions, essentially). As
an example of why this was the case, take a look at the following page (taken from the
Linguistic Atlas of England).
132 Chapter 7. Sociolinguistics
7.3. Problems with Traditional Dialectology 133

This map divides England into various areas based on the way people in these areas
say the verb and pronoun in the question ‘Are you married?’ As you can see, responses to
the inquiries of dialectologists on this matter were quite diverse, but included such truly
archaic forms as the bist thou of ‘area 10’ (compare German bist du ‘are you?’).
The lines which divide the area studied into ‘dialect regions’ within which some par-
ticular form (or, in this case, set of forms) predominates are known as isoglosses. How
are we to interpret the ‘isoglosses’ found on such dialect maps? Take another example.
As one drives south from Michigan to southern Indiana, one crosses a major midwestern
American isogloss, the so-called pin/pen-line. South of this line, the vowels of ‘pin’ and
‘pen’ are identical (because of a historical phonological event whereby the [ɛ] of ‘pen’
raised to [ɪ] before nasals), while north of the line the vowels of these (and phonologi-
cally similar) words are distinct. Let us get some basic facts straight first. It is not the
case that, as you cross this line, the vowels of ‘pin’ and ‘pen’ suddenly merge in your own
speech (if they hadn’t before) — i.e., the isogloss doesn’t control the behavior of humans,
it doesn’t cause the phenomenon it attempts to describe. So the isogloss clearly does not
mean ‘everyone south of this line merges ‘pin’ and ‘pen’ and everyone north of it keeps
them distinct.’ Instead, isoglosses usually1 represent the mid-point in the distribution
of a variable. Excluding people temporarily present on a particular side of the isogloss
— i.e., considering only long-term residents, born where they now reside — in general
we will find a point north of the pin/pen-line where everyone keeps the vowels of these
two words distinct. At a point some distance south of the isogloss, we will reach a point
at which all lifetime residents of the area show in their speech merger of the vowels of
‘pin’ and ‘pen.’ The isogloss represents the midpoint between these two extremes, and of
course we generally find, as we approach the line, that the distribution of people showing
or not showing the merger in question approaches 50-50.
Isoglosses, as we saw from the first example, are not restricted to phonological varia-
tion. They can separate speakers by morphological, syntactic, or lexical features as well.
For example, as one approaches the pin/pen-line, one also runs into the famous bag/sack-
line. North of the bag/sack-line a little paper container, open on the top, is called a ‘bag.’
South of it, this same object is called a ‘sack.’ That is, the same object has two different
names in the two regions. Lexical variation is widespread — many of the differences be-
tween American and British English involve their respective lexicons (‘trunk’ vs. ‘boot;’
‘cookie’ vs. ‘biscuit’; ‘truck’ vs. ‘lorry’; and so on) although differences in many other areas
exist as well.

7.3 Problems with Traditional Dialectology

Since traditional dialectologists were in the business of providing evidence for scholars
who were working specifically on historical linguistic themes, it was most valuable for
1
We exclude fairly uninteresting cases in which the isogloss represents an actual major physical barrier,
like the Atlantic ocean.
134 Chapter 7. Sociolinguistics

them to find, as they went from village to village, the oldest informants they could. Af-
ter all, someone who is 90 years old learned their first language 88 years ago or so, while
someone who is twenty only learned the language 18 years ago. If one is interested in
history, the 90 year old informant thus has things to offer that a 20 year old cannot.
Moreover, it was important that the informants selected by the dialectologist be long-
term residents, preferably from birth, of the village they were being taken to represent.
Neither of us (authors) would be a good choice of informant, for example, since neither
of us currently resides in our place of origin and, as yet, neither of us has adopted the
manner of speaking of those currently around us. In addition, if you could find some-
one who had basically never talked to anybody outside of the village, you were likely to
get the ‘purest’ data for how that village spoke long ago, since you have minimized the
possibilities of linguistic contamination from outside the village.
These two factors — age and long-term residency — had, as an effect, that dialectol-
ogists selected old, lifetime village residents who had had little contact with anybody
outside the village. This meant, given life in the modern world, that the person selected
to represent a given community on a dialect was not actually very representative of the
village itself. That is, if you went to the village, you would not necessarily find that the
average villager (who wasn’t so old and maybe travelled around a bit…) talked the way the
dialect atlas indicated the residents of the village would—since the dialect atlas based its
characterization of the village on the speech of quite old, native born homebodies (rather
than the average ‘man on the street’ or ‘woman on the street’).
In addition, in communities which showed internal diversity along some social di-
mensions (note that this includes virtually every community), such as class or ethnicity,
it was difficult to pick an individual who might be representative of the group as a whole.
Everyone already knew, as we know now, that people from different levels of social struc-
ture talk differently, the working classes talking quite distinctly from the upper classes.
In addition, it was known that if there was, for example, a sizable black community in
a given town, that it was likely that the members of this community would not talk like
the non-black residents. In keeping with the somewhat classist and racist assumptions of
social science at the time, dialectologists generally sought middle class whites as ‘repre-
sentative’ of the community under study. This skewing in the informant selection process
had, needless to say, a corresponding effect on the accuracy of the dialect to geographic
area mapping.
In cases of extreme internal racial, ethnic, and class diversity — as with most major
cities in North America, for example — it was felt that no individual could represent the
community. This problem was dealt with in a relatively straightforward, if decidedly non-
scientific manner: cities were skipped. So the fact that each neighborhood in London, for
instance, has its own dialect is nowhere reflected in the Linguistic Atlas of England. The
same decision was made regarding major North American cities.
7.4. The Sociolinguistic Revolution 135

7.4 The Sociolinguistic Revolution

In keeping with the increasing urbanization of the population in the late 1950’s and early
1960’s, and with the demand for ‘social relevance’ of science that accompanied major so-
cial upheavals such as the civil rights and women’s liberation movements in North Amer-
ica, the shortcomings of traditional dialectology came to be felt as fatal flaws to the enter-
prise. Questions that traditional dialectology, because of the nature of its data collection
techniques (arising in part from its goal as servant to the historical linguistic enterprise),
simply couldn’t ask came to be of major concern. What do people in this place actually talk
like (not just the old, kind of quiet and immobile ones)? What happens — linguistically
— when someone moves across an isogloss? What relationship does linguistic variation
have to such social variables as ethnicity, age, gender, class, etc.?
Clearly a different approach to the study of linguistic variation was required if we were
to make some progress on understanding these issues. Primarily through the efforts of
a single dedicated scholar — William Labov, now at Penn — the field of sociolinguistics,
which attempts to understand the relationship between social variables and linguistic
ones, began to flourish.
Labov is perhaps most famous for his well-developed sense of experimental design (he
was originally trained as a chemical engineer, which may have helped). For example, he
was interested in getting federal funding for a full-scale investigation of the sociolinguis-
tics of New York’s Lower East Side. As you may know, to get federal funding for research
you generally have to do a ‘preliminary study’ to show that there is a good likelihood that
your research, if funded, is going to be fruitful. Unfortunately, Labov had no money to
undertake such a study. In spite of this, he devised a research project which, with no
money and only his own effort, would yield interesting and valuable results. This now
famous study is known as the New York Department Store Study.
The project was simple in its conception. Labov took three socially-stratified depart-
ment stores on the Lower East Side: Saks Fifth Avenue (an upper to upper-middle class
store), Macy’s (a middle to upper-working class store) and Klein’s (like a K-Mart — an
essentially working class store).2 He followed the same procedure for each of the three
stores. He went up to the fourth floor to see what was there. Whatever goods were there
on the fourth floor formed the basis of his survey questions for people in that particular
store. He then proceeded to other floors, going up to people and asking, for example,
‘where are the women’s dresses?’ The person asked would respond (if they knew) ‘the
fourth floor.’ This gave Labov two chances to note down whether or not postvocalic /r/
was being pronounced by people in these various stores — postvocalic /r/ was the lin-
guistic variable he was interested in.
2
Labov walked through the stores after he had collected his linguistic data establishing empirical tests
for the class differentiation. Some of these were physical aspects such as how much merchandise was out
where the customer could see and touch it (in Klein’s all of it was — the place was pretty junky-looking,
in Macy’s most of it was, in Saks one generally had to tell a salesperson what one was interested in and
they brought out samples). There was a lot of open space to mill about in Saks, but in Klein’s one wandered
through narrow isles through stacks and stacks of goods for sale.
136 Chapter 7. Sociolinguistics

Saks Macy’s Klein’s

estimated age (Upper Middle) (Lower Middle) (Working)
15-30 67 21 10
35-50 26 26 0
55-70 13 39 4

Figure 7.1: Percentage of speakers with two /r/’s in casual ‘fourth floor.’

The stroke of brilliance in Labov’s study was the following. After getting a response
to his question which indicated that the person he had asked knew where the women’s
dresses (or whatever) were, he said ‘excuse me?’ This conveyed to the speaker that the
message had not gotten through, and so they repeated it, being more careful, slower,
and more attentive to the act of speaking. For each person he approached in this way,
then, he ended up with four instances of postvocalic /r/: two ‘casual’ and two ‘careful’
realizations.
One of the tables showing some of the results of this study is printed in Figure 7.1.
This table reveals how the realization of /r/ (it’s presence in both ‘fourth’ and ‘floor’)
correlates with Labov’s estimate of the age of the speaker.
The results are quite interesting, both because they provide us with a snapshot of
how people from various classes actually spoke in day-to-day interaction on New York’s
Lower East Side in 1966, and because they give us some idea of the correlation between
the robust presence of postvocalic /r/, social class, and age in that community. As you
can see from the chart, older upper middle class speakers realize /r/ in percentages more
like that of the youngest group of working class speakers — while the youngest group of
upper class speakers are as far away from the youngest group of working class speakers
as anyone. How can we explain this?
As is apparent if you watch American movies from the 1930’s and 1940’s (most of which
were shot in New York) and contrast the speech heard in those movies with that we take
as characteristic of New York today, the status of so-called r-lessness has changed con-
siderably during this period. In general, the well-to-do speakers of the movies of the 30’s
and 40’s are r-less, while the well-to-do of today’s New York are r-ful. In addition, we
know from historical records regarding educational policy in New York that r-lessness
was explicitly taught as the norm in New York schools until just after the Second World
War (which ended in 1945). Older upper middle class New Yorkers in Labov’s study spoke
in keeping with the sociolinguistic situation which was in place when they learned to
speak: that is, they were generally r-less. Younger upper middle class speakers acted in
accordance with the current sociolinguistic situation: they were generally r-ful.
Another interesting result of Labov’s study, not clearly indicated by Figure 7.1, is that
there were differences in some individual’s realization of postvocalic /r/ in casual vs. for-
7.4. The Sociolinguistic Revolution 137

mal speech samples. Specifically, if there was a difference (for a particular speaker), it was
always the case that the casual sample was r-less and the formal sample was r-ful. A sensi-
ble question to ask at this point would be ‘How can one grammar produce both r-less and
r-ful output?’ We hope that this query was inspired by the fact that you know that phono-
logical rules cannot be turned on and off at will. The answer to the question depends upon
one’s view of what’s going on with r-lessness. Under one scenario, r-lessness is produced
by phonological rule. In this case, there is only one possible answer to the question, in our
opinion, and that is that the individual is bidialectal. This is, of course, identical to being
bilingual. Since there is no way to define ‘language’ in a linguistic framework and also no
way to define ‘dialect,’ the terms bilingual and bidialectal reduce to the same thing.
The second possible scenario involves not the grammar but the lexicon, along with
some type of ‘conscious’ cognitive processing. This second scenario develops as follows.
You are exposed to another dialect. You notice, after a little while, not only that they talk
funny but that there seem to be some consistent differences in phonology, morphology,
and syntax between the way you talk and the way they talk. This is usually realized by
thinking something like ‘When I say X, they say Y.’ Critically, these realizations proceed
on a case-by-case basis initially (and often permanently). So, for example, one of the au-
thors of this book is not from Michigan. This one has noticed that when she would say
‘I have run up and down stairs all day.’ many people from Michigan would instead say
‘I have ran up and down stairs all day.’ She has filed this information away as a single
case of correlation between something she says and something ‘they’ say. It is possible
(though unlikely) that this and other forms like it are produced by rule in the grammar of
many Michiganders. If so, there is a chance that this author will, given sufficient time, ex-
posure, and incentive, become bidialectal and will internalize this rule along with many
others. On the other hand, it is possible (and probable) that this difference is the result
of the author’s having a list of irregular past participle forms in her lexicon which does
not match the list of irregular past participle forms in the lexica of many Michiganders. If
this is true, it is highly unlikely that this author will ever be able to produce the Michigan
forms with any consistency. To do so would require that the author hear and consciously
remember the Michigan correlate for every past participle on her list and, for each utter-
ance, consciously translate her output into the correct Michigan form. A process like this
correlation one described has nothing to do with the systematic workings of the gram-
mar. The output of correlating and translating will be slower than normal and also riddled
with odd, performance-related mistakes.
One type of mistake that this ‘correlation technique’ frequently results in is some-
thing that has been termed hypercorrection. Hypercorrection is the result of some general
awareness on the part of a speaker that there exists a correlation between certain things
they say and certain things speakers of another dialect say, but the speaker is not too
sure of the particulars. For example, a characteristic of Brooklyn dialects was the [bojd]
pronunciation for [br̩d], that is, in many lexical items where the standard had a syllabic
[r̩], Brooklyn dialect had the diphthong [oj]. The attempts on the parts of those Brook-
lyn speakers who were not bidialectal to produce the standard resulted in forms such as
138 Chapter 7. Sociolinguistics

[kr̩l] for <coil>, [ʤr̩n] for <join>, and [tr̩lɪt] for <toilet>. It is easy to see how such a
situation could come about. A Brooklyn dialect speaker was aware of some [oj]-[r̩] differ-
ence between their dialect and the standard. Unfortunately, the lexicon of the Brooklyn
speaker included both words had [oj] in them for all speakers (such as the three above) and
words that had [oj] in them for only Brooklyn speakers (e.g., [bojd]) due to some historical
change that affected only that community. The problem facing the Brooklyn speaker was
‘Which words do I translate into [r̩] forms?’ In these sticky cases, speakers often opt for
the ‘overdoing it is better’ strategy and end up ‘hypercorrecting’ — ‘fixing’ forms that do
not need ‘fixing.’
It is difficult to determine exactly which of the scenarios described applied to the
individuals in Labov’s study who showed both r-fulness and r-lessness. Most probably,
some of the respondents were bidialectal and others were simply ‘translating’ (more or
less successfully). We have not yet discussed the reason why, in the case of someone
producing both forms, it was systematically a change to r-fulness from r-lessness in the
careful speech forms. We will turn to that question and related issues in our discussion
of standard and non-standard forms below.

7.5 Standard and Non-Standard

When we described the Broolyn dialect feature earlier, we used the term ‘standard’ for
the non-Brooklyn comparison dialect. So just what is the ‘standard’ and where did it come
from? Well, it is not something defined with respect to Brooklyn, New York, such that ev-
eryone who lives outside that borough (but no one who lives in the borough itself) speaks
the standard. The very notion ‘standard’ is predicated on the non-linguistic assumptions
that there is some grammar plus lexicon that the majority of people share (this relates to
the usual idea that whatever most people do is the ‘common’ or ‘standard’ thing to do) and
that, by virtue of some completely independent social metric (independent of the sheer
numbers formula cited above), one ‘dialect’ is to be held as the ‘purest’ incarnation of the
language. It is easy to see that neither of these assumptions is going to lead to a single,
definable entity that can be, for example, ‘standard English’ and, indeed, no such entity
actually exists. First of all, no two individuals share exactly the same knowledge state
with respect to language. Prescriptive grammars and the educational system attempt to
remedy this situation, in part, but they focus on only a tiny subset of natural language
features for ‘correction’ (and are, in any event, consistently unsuccessful). As far as the
‘purity’ goal is concerned, it is, of course, linguistically impossible to evaluate different
forms for ‘purity.’ Moreover, even a non-linguistic approach is going to run into logical
problems with the idea that some forms are better or more pure than others since what
is considered better or more pure seems to be quite changeable over time and situation.
It turns out that describing what is ‘standard’ involves talking about where a standard
comes from. At any particular point in history, the standard tends to be the dialect of the
social group which at that time possesses prestige, socio-economic status, and power rel-
ative to other social groups. Note that it may not be enough simply to have enormous
7.6. Gender and Linguistic Variation 139

prestige, for example, because money and sheer numbers in the right places are impor-
tant, too. The standard for British English (called Received Pronunciation (RP)) is not the
dialect of Her Majesty Queen Elizabeth II, but rather the dialect of a group with slightly
less prestige (in terms of social class). (RP is typically exemplified by BBC broadcasters.)
In the US, the current standard was drawn from the dialects of white, upper middle class
people who were from other than Atlantic coastal or southern areas (broadly defined).
Interestingly, the US model seems to be the US counterpart of a BBC broadcaster, Walter
Cronkhite. This is still, of course, very vague. On occasion, it is easier to rule out forms
than to describe what might be included in anything that is called ‘standard.’ Even here,
the ground is murky. ‘Non-standard’ forms come in two types — those that are stigma-
tized and those are are ‘non-stigmatized.’ Stigmatized dialect features suffer from guilt by
association. The existence of stratification within society, while dependent upon many
different variables, means that some social groups are held to be on the bottom relative to
others. These groups tend to be stigmatized and, as a result, the dialect features particu-
lar to the group tend to be stigmatized, as well. After some time, the direction of causality
is lost and anyone who has the stigmatized dialect features is, him/herself, stigmatized
regardless of whether they happen to be a member of the stigmatized social group. This
is completely illogical but probably no more so than having social stratification in the
first place. On the other hand, ‘non-stigmatized’ forms which are ‘non-standard’ may
be noticed but used with impunity (socially speaking). The fact that, in some dialects (in-
cluding the current standard), the word for a cooking vessel is [pat] while in other dialects
it is [pɔt] is unremarkable. In cases such as these, the forms which are, strictly speaking,
non-standard will be accepted as standard (just a ‘different’ standard).
The existence of a (somewhat normative) standard is another facet of the interaction
of social variables and linguistic features. The majority of sociolinguistic studies focus
upon standard and non-standard (frequently stigmatized) linguistic features with respect
to various social variables. The ‘direction’ of correlation is most often similar to that of
the Department Store study where people tend to ‘switch up’ (toward the standard) if
they can. In the following section, however, we will see that this is not always the case.

7.6 Gender and Linguistic Variation

Another study which showed some correlation between linguistic forms and social class
was conducted by Trudgill (1996). In this study, five different groups in Norwich, Eng-
land were examined for the production of [ɪn] (a non-standard, somewhat stigmatized
form) and of [ɪŋ] as possible final sounds in such words as ‘running,’ ‘jumping,’ and so on.
The five groups were: Lower Working Class (LWC), Middle Working Class (MWC), Upper
Working Class (UWC), Lower Middle Class (LMC), and Middle Middle Class (MMC). Data
from both casual speech and reading (i.e. very careful speech) were collected.
140 Chapter 7. Sociolinguistics

Group Casual Reading

LWC 0 34
MWC 5 56
UWC 13 85
LMC 58 90
MMC 72 100

Figure 7.2: Percentage [ɪŋ] forms produced.

These results indicate that speakers of higher socio-economic classes (LMC and MMC)
produce a much greater percentage of [ɪŋ] forms in casual speech than do speakers of
lower socio-economic classes. Speakers in all groups shifted ‘upward’ (toward the stan-
dard form) in careful speech contexts, with the most dramatic shift being found in the
UWC group. Interestingly, it appears to be characteristic of the UWC (and the LMC in sur-
veys of US informants) that they show this type of greater shift toward the standard. One
hypothesis put forth to account for this is that these social groups suffer from higher lev-
els of ‘social status insecurity’ and therefore have a greater incentive to emulate linguistic
variables associated with social groups above them in the social ladder.
The data in the Norwich study might have illustrated an additional relationship be-
tween linguistic variables and social variables if more finely tuned social variables were
noted — particularly, the gender of the informants.3 In culture after culture, we observe
that women consistently speak like men of the next higher socio-economic class. More-
over, when asked about their own behavior (e.g., if asked ‘how often to you say [ɪn] instead
of [ɪŋ] in words like running, jumping, playing?) women consistently underreport stigma-
tized forms (like [ɪn]), claiming to use such forms only rarely, while men tend to overreport
the use of stigmatized forms, claiming to say [ɪn] far more often than observations of their
behavior in informal situations actually indicates is the case. These gender behavior dif-
ferences (in terms of forms actually used, not forms reported by respondents to be used)
manifest themselves well before individuals reach adulthood. In a study of the casual
speech of twelve boys and twelve girls aged three to ten in a New England village, Fischer
(1958) noted the following results.

[ɪŋ] [ɪn]
Girls 10 2
Boys 5 7

Figure 7.3: Number [ɪŋ] and [ɪn] Forms Produced By Boys and Girls Aged 3 to 10.

3
Note that, unlike some other social variables — age, income or occupation, etc. — gender has been,
until recent medical breakthroughs, essentially immutable.
7.7. Summary 141

The explanation for this phenomenon is not entirely clear. The role of women as
the primary bearers of responsibility for ‘social propriety’ has been suggested, as has the
general riskiness of their social status relative to men in the same social class. Another
suggestion is that there is a ‘masculinity factor’ associated with working-class lifestyles
and, correspondingly, with the linguistic forms of the working class. This factor may be
strong enough to compete successfully against the prestige factor of the linguistic forms
associated with more generally prestigious social groups.

7.7 Summary
If we ask ourselves why linguistic variables appear to match up with, for example, social
class the way that they do, an explanation does not seem difficult to come up with. For
the most part, people learn their language from members of their own social class and,
on the whole, they spend more time talking to those in their own class, or (in a more fine-
grained analysis), members of immediately adjacent classes, than they do to people from
wildly different social sets. In this sense, the explanation for the relationship between
class and linguistic variables is the same as that for the relationship between geographic
area and linguistic variables: people usually talk to and learn their linguistic forms from
people who live nearby. One obviously tends to sound like those to whom one speaks and
from whom one learns to speak. Class, in this sense, is just a measure of ‘social space,’
in which the working classes are ‘more distant’ from the upper classes than are the mid-
dle classes. ‘Social space’ regulates (mildly) social interaction, just as geographical space
regulates (mildly) social interaction.4 Variables, like r-lessness, spread from individual
to individual in these spaces — geographical and social. After all, it is extremely unlikely
that someone who is r-less in New York was born to r-ful parents and just came up with
the idea of being r-less on his or her own. An r-less speaker with r-ful parents has almost
certainly acquired his r-less dialect by learning it from someone who was already r-less
(a schoolmate, for instance). The study of sociolinguistic variables is the study of diffusion
of linguistic forms or rules throughout the social network.
Why would someone adopt the linguistic forms of another? Obviously, some (if not
all) speakers seem to think that they need an additional linguistic system at some point
— that their original system was not adequate for all their needs. (Note that neither the
decision nor the implementation is conscious. No one suddenly says to him/herself ‘Boy,
I really need another linguistic system.’) What is the motivation for this?
These are difficult questions, though in general there appears to be a single, relatively
clear answer: prestige. We learn new forms as part of a desire to be more like those who
we think are ‘cool.’ We may adopt various other properties of these prestige individuals,
as well: the way they dress, the way they walk, the kinds of things they worry about, etc.
Successful adoption of these features requires sufficient contact that we can generate a
4
By ‘mildly,’ we merely mean not in a deterministic manner — you can talk to someone from another
place or another class, it’s just that on average you are less likely to.
142 Chapter 7. Sociolinguistics

coherent theory of what system they are using to determine what to wear, how to walk
a certain way, and so on. Similarly, in the linguistic dimension, successful learning of
the linguistic system of another requires considerable contact. As pointed out above, the
bulk of our contact is with those who live in our area and who are members of our own
(or closeby) social classes or groups. Every social similarity makes interaction more likely
(same gender, same ethnicity, same religion, same class, etc.), and thus opens the door to
the learning of a new linguistic system.
Sociolinguistics is thus about the diffusion of linguistic variables through the social
network. As such, it can be very revealing as to the precise nature of that network. The
origins of the social network are not, of course, the concern of linguists, but what of the
origins of linguistic variables? That is, where did the first r-less speaker get his or her
r-lessness? This question is the proper concern of historical linguistics, to which we now
turn.
7.8. Exercises 143

7.8 Exercises
1. Which of the following statements is true?

a) Isoglosses are lines which show absolute divisions between sets of speakers
with certain speech features.
b) Isoglosses are lines which indicate natural barriers on maps.
c) Isoglosses are lines which represent mid-points in the distribution of some
speech feature.
d) Isglosses are lines which keep speakers with one speech feature from talking
to speakers with another speech feature.

2. Traditional dialectologists focussed on non-urban areas because

a) urban areas didn’t show any dialect differences.

b) speakers in urban areas were too linguistically diverse for there to be a single
representation for them.
c) speakers in urban areas were too linguistically uniform for it to be an inter-
esting exercise.
d) it was difficult to find elderly speakers in urban areas.

3. Which of the following is true?

a) The dialect features that you are born with determine what social class you
will be in.
b) Whatever dialect you are born with, you maintain throughout your life.
c) Your dialect features will probably be similar to dialect features of other mem-
bers of your social class.
d) Social variables never correlate with linguistic variables.

4. Which of the following is true?

a) Having stigmatized dialect features is a clear indication of intellectual inferi-

ority.
b) Having stigmatized dialect features means that you speak only the standard
form of the language.
c) The ‘stigmatization’ of dialect features is the result of some dialects being bet-
ter than others from a linguistic point of view.
d) The ‘stigmatization’ of dialect features is a social phenomenon, not a linguistic
one.
 Chapter 8
Historical Linguistics

Contents
8.1 Language Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
8.1.1 Phonological Change . . . . . . . . . . . . . . . . . . . . . . . 145
8.1.2 Morphological Change . . . . . . . . . . . . . . . . . . . . . . 147
8.2 Comparative Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . 149
8.3 Relative Chronology and Subgrouping . . . . . . . . . . . . . . . . . . . 162
8.4 Cultural Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

8.1 Language Change

Where do the linguistic variables, which diffuse through social networks, come from?
Popular media and old fogies will tell you that linguistic change is the result of the de-
cay, laziness, and stupidity of the younger generation. They’ve said that throughout all
of recorded human history (and no doubt earlier than that). While we would be hard
pressed to deny claims that we’re getting lazier, stupider, or more decayed for our own
generation (let alone younger ones), it seems unlikely that such factors have anything to
do with language change. Instead, it appears that change is one of the essential properties
of ‘language’ (in its sociopolitical sense) – we have never observed a linguistic tradition
on earth without noting that change (and the variation which results from the diffusion
of change) has taken place. Change cannot therefore be something which happens to lan-
guage as a result of the accidental properties of its speakers (laziness, etc.), which might
not be present in a given population or under certain climatic conditions (which tend
to be relatively constant for non-nomadic populations). If the universality of change is

145
146 Chapter 8. Historical Linguistics

to be coherently explained, the fact of change must be traced to something inherent in

language itself.
It is not difficult to see what feature of language it is that is responsible for universal
change. As we pointed out in the previous chapter, the diffusion of sociolinguistic vari-
ables is generally the result of learning. You talk in a manner similar to that of the people
from whom you learn to talk. The reason for this similarity is to be sought in the mech-
anism whereby you come to have a linguistic system: while some aspects of that system
are innate (as we’ll see in the next chapter), much of the details of a particular grammar
(e.g., its entire lexicon) must be learned. This learning takes place on the basis of input
which you receive as a child – and this input comes from those with whom you interact.
However, as we noted at the beginning of this book, the evidence for the underly-
ing linguistic system that the child receives is seriously defective in a variety of ways.
The data is full of performance errors – misspeakings, hesitations, false starts, changes of
intended sentence mid-stream, and other lapses – and presented to you not in the sound-
proof booth that linguists are fond of doing phonetic research in but in an environment of
passing busses, squawking pigeons, annoyingly noisy siblings, people talking with their
mouths full, etc. It’s a wonder you can extract any coherent system from the awful data
you get (a mystery we’ll address when we turn to acquisition).
It is hardly surprising, then, that, on the basis of the noisy and performance-influenced
data which your mother (or whoever) presented you with as a child, you failed to con-
verge precisely on the same knowledge state as was in her mind when she was speaking
to you. You inevitably ended up with a grammar which was (slightly) different than that
of the preceding generation because language is an imperfectly replicating system. This im-
perfect replication is a function of the fact that neither you, nor your mother for that
matter, had any form of direct explicit access to the grammatical knowledge in your
mother’s brain. She could therefore give you no explicit assistance in your challenging
learning task (being quite unaware herself of the properties of her grammar – after all,
if we all knew what our grammars were like, we’d all get perfect scores on the exams in
this class…).
So, the grammar which you construct on the basis of the preceding generation’s out-
put will differ in subtle ways from the grammars of that generation. The grammar of your
linguistic descendants will also differ from yours in small ways. And the grammar of your
descendants’ descendants will differ again. Over time, obviously, the differences between
generations at some considerable remove from one another will be quite dramatic. For
example, many generations ago (e.g., in the fourteenth century) the words corresponding
to our [haws] ‘house’, [maws] ‘mouse’ and [laws] ‘louse’ were pronounced [hu:s], [mu:s]
and [lu:s]. Small changes from generation to generation since that time have resulted in
these words having, for most North Americans, rather different pronunciations.
These examples tell us two important things: first, one type of change is sound change;
and second, sound change is systematic and regular. In each of the examples cited above,
the [u:]’s of Middle English are now pronounced [aw] by us—and this is quite generally
the case for Middle English [u:]’s. This regularity allows us to write ‘sound change’ rules,
8.1. Language Change 147

much as we can write phonological rules which reside in the grammars of speakers (like
the rule which aspirates voiceless stops in English). Examine the data below concerning
the development of h from Old to Modern English.

Old English Modern English gloss

hluːd lawd loud
hrɪŋ rɪŋ ring
hnʊtʊ nət nut
hrævn revn̩ raven
hnappian næp nap
hand hænd hand
heːla hil heel
hwiːnan wajn whine

What this data shows (we’ll ignore all of the development except those of h) is that Old
English h was sometimes lost and sometimes preserved in Modern English. In fact, it
should be clear that the developments can be specified quite precisely: initial h was lost
before another consonant, but preserved when it preceded a vowel.1 We can write this
as a ‘sound change’ of the form:

• h > Ø before consonants

It is important to bear in mind that such rules have a very different status than the
rules of theoretical phonology which we discussed earlier. Whereas phonological rules
reside in the mind of speakers, regulating the computational processes which govern the
generation of phonological forms, sound changes express the relationship between ear-
lier and later grammars (which are assumed to form part of a linguistic tradition). These
rules are thus not aspects of any one person’s cognitive system.
Change (which is the ultimate source for variation) is not limited to the sound system
of a language. We find changes in the lexicon (e.g., the loss of ‘davenport’ as the word for
‘couch’ or ‘sofa’ for many North Americans), in morphology (e.g., the replacement of ir-
regular plurals like ‘kine’ and ‘brethren’ by regular plurals ‘cows’ and ‘brothers’), syntax,
and semantics. A complete survey of these various types of change is beyond the scope
of this introductory textbook. We will focus our attention on phonological and morpho-
logical change – areas which are relatively well-understood in historical linguistics.

8.1.1 Phonological Change

As we saw in the discussion of the development of Old English /h/ above, one of the most
striking aspects of phonological change is its regularity. Like phonological rules in the
1
This statement is not true – as you might expect – for all dialects of English. We are using our own
dialects as the standard example. For example, some dialects have lost h before all consonants except [w] –
speakers of these varieties of English still say things like [hwajn] for ‘whine’. Other dialects, e.g. Cockney
English, have lost h in initial position in all words, both before consonants and before vowels.
148 Chapter 8. Historical Linguistics

grammar, phonological changes apply whenever their conditions are met. Sometimes,
these conditions are very general – in such a case, the segment which changes will un-
dergo the change regardless of what phonological environment it is in. Examine the data
from the Mafea language of Vanuatu in the table below.2 The data given in not sufficient
to enable you to figure out everything that happened, so we direct your attention to the
development of Proto-Oceanic *k.3

Proto-Oceanic Mafea English

*kaɣati ati bite
*keli eli dig
*vinsiko v̈isio meat
*pakiwa p̈aio shark
*kuɣurita uita squid

As you can see from this data, Proto-Oceanic *k is lost in Mafea in every case – both word-
initially and intervocalically.4 This is an example of an unconditioned sound change – the
affected segment shows the change in all environments in which it is found. One could
write the change as follows:

• *k > Ø

Note the lack of any statement of a conditioning environment for the change.
By way of constrast, examine the data regarding the development of Proto-Oceanic
*p in Mafea presented in the table below.

Proto-Oceanic Mafea English

*pepe p̈ep̈e butterfly
*apu apu fire
*puto puto navel
*ponse poso paddle
*piri p̈iri seed
*pakiwa p̈aio shark

We see that Proto-Oceanic *p has two distinct reflexes in Mafea. Since sound change is
regular, these two developments must be conditioned in some way. An examination of the
2
The IPA symbol ɣ is used to designate a voiced velar fricative. The Mafea sounds designated by p̈
and v̈ are the ‘apico-labial’ stop and voiced fricative, respectively. ‘Apico-labial’ designates the very rare
articulation in which the tip (apex) of the tongue is placed against (or, in the case of the fricative, very near)
the upper lip.
3
‘Proto-Oceanic’ is the name given to the reconstructed ancestor of most of the languages of Oceania.
The method of linguistic reconstruction will be introduced to you later in this chapter. Hypothetical forms
reconstructed for proto-languages are marked with an asterisk.
4
Proto-Oceanic did not have any word-final consonants, so there is no evidence regarding Proto-
Oceanic *k in this position.
8.1. Language Change 149

data above reveals that the conditioning factor is the roundness of the following vowel:
Proto-Oceanic *p becomes Mafea p̈ when the following vowel is not round, remaining p
when the following vowel is round. We can write a formal statement of the change to
account for these developments as follows:

• *p > p̈ / V[−round]

Note that no change statement need be posited to get Proto-Oceanic *p to come out p in
Mafea since there’s no change involved in this development, obviously.
In general, the types of events we see in examples of sound change are very much like
the types of processes we see in phonological rules. For example, assimilatory changes
are very common both as synchronic phonological rules (e.g., English vowel nasalization
before nasal vowels) and as diachronic phonological events, as can be seen from the data
in the table below.

Proto-Romance Portuguese English

*kantare kãtar to sing
*lɔŋga lɔ̃ga long
*infɛrnu ĩfɛrnu hell
*dɛnte dɛ̃te tooth

In Portuguese, vowels have become nasalized before nasals (and nasals after vowels are
subsequently lost). The first step in this change is assimilatory, just like the English syn-
chronic vowel nasalization rule.

8.1.2 Morphological Change

Interestingly, it is not the case that whenever the sounds associated with a particular
meaning change, we call it ‘sound change.’ For example, in earlier forms of English, the
plural of ‘cow’ was ‘kine,’ the only plural of ‘brother’ was ‘brethren,’ and the only com-
parative of ‘old’ was ‘elder.’ The change of earlier English ‘kine’ to Modern English ‘cows’
is clearly not sound change (e.g., ‘mine’ did not change into ‘mows’, ‘line’ did not change
to ‘lows’, etc.). It seems clear that the change of ‘kine’ to ‘cows,’ or of ‘brethren’ to ‘broth-
ers,’ is related in some way to the fact that /-z/ is the regular plural morpheme in English.
Remember, from our discussion of the ‘wug’-test, how we account for an ‘irregular’ plural
like ‘oxen’ (as opposed to a regular plural, like ‘dogs’). The lexical entries of ‘ox’ and ‘dog’
include the following information:

• /dɔɡ/ dog, [+N], [pl: ]

• /aks/ ox, [+N], [pl: /aksn̩/]

150 Chapter 8. Historical Linguistics

In these entries we see that ‘dog’ includes a phonemic specification of the underlying
phonological form (/dɔg/), a representation of the meaning (dog), a syntactic category
designation ([+N] for nouns), and an indication that there is no idiosyncratic plural form
([pl: ]). By contrast, for ‘ox’ we find a phonemic representation (/aks/), a semantic rep-
resentation (ox), a syntactic category designation ([+N]), and an indication that there is
an idiosyncratic plural associated with this lexical item ([pl: /aksn̩/]). In the ‘wug’ test,
since the subject of the experiment is assumed not to have a lexical entry ‘wug’ at all un-
til the experimentor presents them with the word, it seems certain that the subject does
not have any ‘idiosyncratic’ plural information stored with this word. Having no idiosyn-
cratic information stored with a given noun means, of course, that its lexical entry looks
just like that of ‘dog’ with respect to the plural. Like ‘dog’, such forms get the ‘default’
English plural marker /-z/, thus /wəgz/ ‘more than one wug’ is generated.
Looked at this way, we can envision the change of ‘cow:kine’ to ‘cow:cows’ in the fol-
lowing way:

• /kaw/ cow, [+N], [pl: /kajn/] → /kaw/ cow, [+N], [pl: ]

In this scenario, the idiosyncratic plural information stored with the lexeme ‘cow’ changes
from having an entry ([pl: /kajn/]) to lacking one ([pl: ]). Since all lexemes which lack
idiosyncratic designation of their plural get the regular /-z/ plural in English, /kawz/ will
be generated after the change.
How did the lexeme come to lack its idiosyncratic plural form? We don’t know for
certain in this particular case, but several possibilities are readily available. Perhaps the
learner responsible for this innovation never saw or heard anything about more than
one cow until he or she was thirty years old. Not having heard /kajn/, s/he can hardly
have stored it in his/her lexicon! Or perhaps a learner heard /kajn/ but assumed it was
a generic word for ‘cattle’ that had nothing to do with ‘cow.’ This, too, would lead to him
or her having no /kajn/ in the plural designation of ‘cow’ and thus lead him or her to say
‘cows’ in this function.
The case of ‘brethren’ is a little more complicated. Imagine that the original situation
was that there was a word ‘brother’ which meant simply ‘male sibling’. It would have had
a lexical entry like:

• /brəðr̩/ male sibling, [+N], [pl: /brɛðrən/]

Imagine further that this word could be used metaphorically to refer to any pair of male
associates with a particularly close relationship, such as, for example, male members of
a monastic order. Eventually, this metaphorical use became so regular that it was taken
to be simply a different word altogether. At this point there were two words ‘brother’ in
English, one referring to a male sibling, one to a male member of a religious order. When
the original lexical entry split in two, it had to be determined by the learner which one
of the forms the irregular plural was associated with, or whether it was associated with
both. Obviously, some learner got this ‘wrong’, assigning the irregular plural ‘brethren’
8.2. Comparative Reconstruction 151

only to the religious use of the term, leaving the plural designation for the ‘male sibling’
use of the term blank. A blank entry under ‘[pl: ]’ means that the regular /-z/ plural will
be used. We can designate the result of this change to the lexical entry cited above as
follows:

• /brəðr̩/ male sibling, [+N], [pl: ]

• /brəðr̩/ male member of a religious order, [+N], [pl: /brɛðrən/]

The vast majority of morphological changes involve ‘regularization’ of forms of this

type. The process is sometimes called ‘analogy,’ based on the idea that the new plurals are
created by a form of reasoning usually designated by that name. The form of reasoning
works like this

• /ston/:/stonz/::/kaw/:X

which is usually read ‘stone is to stones as cow is to X.’ If you solve this ‘equation’ for X
you will get the form ‘cows.’ Unfortunately, this theory can be taken as implying that
learners create lexical items by this form of (non-linguistic) reasoning. We have shown
above that there are probably more plausible linguistic mechanisms which can account
for the phenomenon of morphological change.

8.2 Comparative Reconstruction

It is often the case, when we examine data from a set of languages, that they show similar-
ities in the forms of morphemes. There are three possible explanations for such similari-
ties: (1) the similarities may be due to simple chance – after all, only so many phonological
segments exist in human language, and, by chance alone, we might expect some relatively
small number of morphemes to have similar segments in them when we compare two lan-
guages; (2) the similarities may be due to borrowing – languages sometimes borrow words
from one another, and when they do so, the words will end up looking similar or even
identical to one another; or (3) the similarities may be due to the fact that the languages
being examined were, at some earlier historical period, the same. This last will occur when
differences arose due to separate sets of changes taking place in the history of each lan-
guage since the time of their common origin. This last possibility is known as the genetic
hypothesis – where ‘genetic’ refers not to some physiological connection through the
genes, but rather to the ‘origins’ meaning of ‘genetic’ seen in words like ‘genesis.’ Lan-
guages which show genetic similarities, as opposed to similarities due merely to chance
or borrowing, are known as ‘related’ languages. Thus, to claim that English is ‘related’ to
Hindi (as it is) is to claim that English and Hindi have evolved by differentiating changes
from a common ancestor.
152 Chapter 8. Historical Linguistics

Historical linguistics provides us with a technique for determining attributes of the

common ancestor of a set of related languages: the comparative method. We will demon-
strate this technique by examining a set of data for four related Polynesian languages:
Hawaiian, Maori, Samoan and Tongan. The data we will be considering is presented in
the table below.

Comparative Polynesian Data

Hawaiian Maori Samoan Tongan English
manu manu manu manu bird
awa awa awa awa channel
niu niu niu niu coconut
pua pua pua pua flower
mimi mimi mimi mimi urinate
peʔa peka peʔa peka bat
muli muri muli mui behind
kani taŋi taŋi taŋi cry
au au au ʔau current
kuna tuna tuna tuna eel species
walu waru walu walu eight
iʔa ika iʔa ika fish
kae tae tae taʔe excrement
lono roŋo loŋo oŋo hear
lau rau lau lau leaf
ʔuku kutu ʔutu kutu louse
umu umu umu ʔumu oven, earthen
walu waru walu wau scratch
kapu tapu tapu tapu taboo
ako ato ato ʔato thatch, roof
lua rua lua ua two
lua rua lua lua vomit
lena reŋa leŋa eŋa yellow, turmeric

It should be obvious on even superficial consideration by you that the forms in this table—
which are representative of the kinds of similarities found throughout the lexicon of these
languages—show a remarkable degree of similarity. Contrast, for example, the first four
columns with the form you see in the English gloss in the fifth column.5 For row after
row, the first four columns are more similar to one another than any of them is to the
fifth column. The ultimate explanation for this fact is to be found in the fact that the first
four columns provide data from a set of related languages (in a sense we will make clear
soon). The fifth column—the English data—by contrast is taken from a language which is
not related to the other four.
5
The only English form which really comes close is ‘taboo’, which, as it turns out, is a loanword from a
Polynesian language!
8.2. Comparative Reconstruction 153

As mentioned above, the similarities seen in the first four columns are pervasive for
these languages—they hold of literally thousands of additional lexical items, which we
don’t list here merely for reasons of space. This similarity is too extensive for it to be
attributed to borrowing—quite aside from the fact that these speakers are separated from
one another by hundreds, and in some cases thousands, of miles of open ocean (making
‘borrowing’ words from each other quite a challenge!). In addition, the similarities we
see in the data are too widespread for them to be due to mere chance. After all, chance
similarity between the forms cited in the first four columns should show the same pattern
as the relationship between any of those columns and the English data (for which there
is just as high a probability of accidental similarity), yet the data looks very different for
the first four columns.
This leaves us with a genetic explanation for the similarities—Hawaiian, Maori, Samoan
and Tongan must have once been the same language, the present-day differences be-
tween them being due to a set of phonological changes which have taken place in each lan-
guage since they went their separate ways (much as North American English and British
English have come to differ from one another).
Imagine that you were asked to formulate a scientific hypothesis about what the com-
mon ancestor of Hawaiian, Maori, Samoan and Tongan (call it ‘Proto-Polynesian’) might
have sounded like. Let us start with the first row of data in the table above:
Hawaiian Maori Samoan Tongan English
manu manu manu manu bird

What might you guess the Proto-Polynesian word for ‘bird’ probably was? it seems that
hypothesizing that this word was anything other than *manu would be pretty silly (and
certainly wouldn’t be justified by any evidence from the data you’ve been given)!
Let us think about how we came up with the idea that the Proto-Polynesian word was
probably *manu—indeed, that it would be irresponsible, given the data, to hypothesize
that it was anything else. If we break down our reasoning into its component parts, it
seems like our brains did something like this:

Hawaiian Maori Samoan Tongan Proto-Polynesian

m m m m *m
a a a a *a
n n n n *n
u u u u *u

The relationships represented in this table are known as ‘sound correspondences’. The ta-
ble can be interpreted as follows: an m in Hawaiian corresponds to an m in Maori, Samoan,
and Tongan; an m in Maori corresponds to an m in Hawaiian, Samoan and Tongan; an m
in Samoan corresponds to an m in Hawaiian, Maori, and Tongan; and an m in Tongan
corresponds to an m in Hawaiian, Maori, and Samoan. These correspondences are to be
explained by the fact that the Proto-Polynesian *m has not undergone any change (at
154 Chapter 8. Historical Linguistics

least in this phonological context—on which more below) during the historical devel-
opment from Proto-Polynesian times to modern Hawaiian, Maori, Samoan, and Tongan.
Since none of these languages changed the original Proto-Polynesian *m, they still all
have /m/ at the start of the word for ‘bird’. Again, it would be very unscientific to as-
sume anything else: we find evidence only for /m/ at the start of the word for ‘bird’ in all
of the Polynesian data we have.
We can extract from this reasoning a simple principle:

• Principle I: If all the languages being compared have the same segment in a given
correspondence, reconstruct that segment for the ancestral proto-language.

Note that this principle merely captures the logic which you yourself used when you were
asked what you would hypothesize the Proto-Polynesian word for ‘bird’ was. It represents
a binding principle: if all of the languages being compared have a given segment in a
correspondence set, you are not allowed, in reconstructing the proto-language, to posit
any other segment.
If we examine more of the data, we will discover more correspondence sets in which
we find the same segment in all of the ‘daughter’ languages of Proto-Polynesian.6 Exam-
ine the forms below:
Hawaiian Maori Samoan Tongan English
awa awa awa awa channel
niu niu niu niu coconut
pua pua pua pua flower

Just as in the case of Proto-Polynesian *manu ‘bird’, it seems clear that the most responsi-
ble scientific hypothesis we could make from this data is that the Proto-Polynesian word
for ‘channel’ was *awa, the word for ‘coconut’ was *niu and the word for ‘flower’ was *pua.
Note that in addition to knowing more about these individual words, we also know more
about the segments of Proto-Polynesian. We established earlier that that language must
have had an *a, an *u, an *m and an *n (otherwise it couldn’t have had a word *manu!),
we can add the following ‘sound correspondences’ to our knowledge set now:

Hawaiian Maori Samoan Tongan Proto-Polynesian

w w w w *w
i i i i *i
p p p p *p

Importantly, the hypotheses we have generated regarding Proto-Polynesian, like all good
scientific hypotheses, have predictive power. For example, if we assume nothing special
is going on, and I tell you that the Hawaiian word for ‘urinate’ is /mimi/, what would you
6
We will pretend here that these are ‘all’ of the Polynesian languages, because we don’t want to bury
you in data from dozens of additional languages.
8.2. Comparative Reconstruction 155

predict the Tongan word for ‘urinate’ to be? Well, since Hawaiian /m/ comes from Proto-
Polynesian *m, and Hawaiian /i/ comes from Proto-Polynesian *i, we can hypothesize
from Hawaiian /mimi/ that the Proto-Polynesian word for ‘urinate’ was *mimi. If it was,
since Proto-Polynesian *m comes out /m/ in Tongan, and Proto-Polynesian *i comes out
/i/ in Tongan, the Tongan descendent of Proto-Polynesian *mimi should be /mimi/. Look
back at the original table of data which we started our discussion of reconstruction with:
there’s the Tongan word for ‘urinate’, /mimi/, just as predicted.
The predictions we can generate in this way are not limited to the data in the table
we gave you. If we tell you that the Samoan word for ‘anoint the hair with coconut oil’
is /pani/, can you tell us what the Maori word for this is? Yes, you can, because Samoan
/pani/ comes from Proto-Polynesian (from now on we’ll abbreviate this to PPn) *pani (as
the sound correspondence tables tell us), and PPn *pani will give Maori /pani/! Again,
this is correct. From the information that PPn *pani means ‘anoint the hair with coconut
oil’ can you tell me what the Chinese word is? No, you can’t: you have no ability to say
anything about Chinese from data on Proto-Polynesian, because Chinese isn’t a Polyne-
sian language, and isn’t included in the sound correspondence charts! The sound corre-
spondence charts and the Proto-Polynesian reconstructions are scientific theories about
a part of the world: the Polynesian languages and their phonological (and lexical) rela-
tionships. Since Chinese, French, and English aren’t Polynesian languages, these theories
tell us nothing about those languages.
Okay, you might say, this is too easy: all the words we’ve looked at have been exactly
the same in all of the Polynesian languages, and in Proto-Polynesian as well. If this were
generally true of these languages, the data in the ‘Comparative Polynesian’ table would
show the same form in every column for every word. But it doesn’t, as can be seen from
examples such as:

Hawaiian Maori Samoan Tongan English

peʔa peka peʔa peka bat
iʔa ika iʔa ika fish

This data gives us one additional ‘across-the-board’ sound correspondence, which allows
for a trivial PPn reconstruction:
Hawaiian Maori Samoan Tongan Proto-Polynesian
e e e e *e

but what are we to do about this sound correspondence:

Hawaiian Maori Samoan Tongan Proto-Polynesian

ʔ k ʔ k

It seems clear enough that there are two possibilities here: either PPn had *ʔ in the words
for ‘bat’ and ‘fish’ (like Hawaiian and Samoan do) or it had *k in these words (like Maori
156 Chapter 8. Historical Linguistics

and Tongan do). Notice that any other hypothesis, like one that held that PPn had *h in
these cases, would require every language to have changed the original form, whereas the
*ʔ or *k hypothesis require changes in only two of the four languages. This makes these
hypotheses, since they require less to have happened for us to have ended up with the
data we have, more economical. But it is hard to see how we could possibly pick between
them. Indeed, from the evidence seen by you so far, there is no way to select between
the two theories. The responsible thing to do in this case is not pick! We will reconstruct
instead a segment which we will write *(ʔk), which is to be interpreted as meaning either
glottal stop or *k, but we can’t (yet) tell which. We can hope that later data will help us
to resolve the question. Thus we have at this point:

Hawaiian Maori Samoan Tongan Proto-Polynesian

ʔ k ʔ k *(ʔk)

added to our list of sound correspondence.

The next form we cannot at present explain involves daughter languages with /l/ or
/r/. We extract from the original table of data all the straightforward (i.e., not confused
by other changes we don’t yet understand) data regarding these segments:

Hawaiian Maori Samoan Tongan English

muli muri muli mui behind
walu waru walu walu eight
lau rau lau lau leaf
walu waru walu wau scratch
lua rua lua ua two
lua rua lua lua vomit

The facts regarding /l/ and /r/ are complicated. There are two correspondence sets in
the data:
Hawaiian Maori Samoan Tongan Proto-Polynesian
l r l Ø
l r l l

The first question that arises concerns how many Proto-Polynesian segments we need
to posit in order to account for these correspondence sets. You might think the an-
swer to this is obvious—two different sound correspondences, therefore two different
PPn segments—but things are not that clear. For example, look at the following table of
lexical correspondences between the speech of a typical anglophone midwesterner and
a typical working-class Bostonian:
8.2. Comparative Reconstruction 157

Ypsilanti, MI Boston, MA gloss

farm fam ‘farm’
ʧarm ʧam ‘charm’
kar ka ‘car’
rɛd rɛd ‘red’
krajd krajd ‘cried’

From this data, we can set up two sound correspondences involving /r/:

Ypsilanti, MI Boston, MA
r Ø
r r

If two different correspondence sets means two different proto-language segments, we

would need two segments in ‘Proto North American English’ for the sound correspon-
dences above. But evidence indicates strongly that in fact there was a single segment, *r,
in all of these words. If sound change is regular, how can *r come out in two different
ways in the Boston dialect?
We can see the answer to this question if we examine the phonological context of the *r:

Proto North American English Outcome of *r in Boston

*farm Ø
*ʧarm Ø
*kar Ø
*rɛd r
*krajd r

Approaching this question is much like approaching a phonology problem. Sound changes,
like phonological rules, do not have to affect a segment in all phonological environments:
they can be conditioned by context. For example, in the table above, all of the *r’s which
got lost in Boston were post-vocalic (i.e., after vowels). None of the *r’s which survived in
Boston were post-vocalic. So we can write a sound change for Boston like this:

(8.1) *r > Ø / V

which means simply that *r was lost after vowels. Since only some instances of *r were
lost (those after vowels), while others were retained (those not), we get two distinct corre-
spondence sets between the Ypsilanti dialect (where all *r’s were retained) and the Boston
dialect. Two correspondence sets, but only one original segment!
The general principle we use in a case like this is the following:

• Principle II: Minimize the number of segments reconstructed for the proto-language.
158 Chapter 8. Historical Linguistics

This principle implies that if we can derive a sound correspondence by the use of a con-
ditioned sound change, we should do so. To determine whether we can do so, as we have
seen, we need to look at the phonological environments in which we find the segments
involved in the correspondence.
Of course, the fact that sometimes two similar-looking correspondence sets can be
collapsed into a single proto-language segment doesn’t mean that they always can. The
key question to ask is whether, without giving up on the idea of the regularity of sound
change, we can posit a conditioned sound change (or a set of conditioned sound changes)
that would have given rise to multiple correspondence sets even though we start out
with only a single segment. Let us return to the Polynesian case and see how the matter
stands. If the two sound correspondences are to be linked up to a single PPn segment,
then we would need to be able to write conditioned sound changes, as we did above for
*r in Boston, which would give rise to the two correspondence sets. Look again at the
relevant sound correspondences, repeated here for your convenience:

Hawaiian Maori Samoan Tongan Proto-Polynesian

l r l Ø
l r l l

It is only the Tongan data which is responsible for the fact that we have two rows here,
rather than one: if we took Tongan out, there would just be the one l = r = l set for
Hawaiian, Maori and Samoan, respectively. So the place that the conditioned change
would have happened, if that is the right analysis, would be in Tongan, where we would
sometimes get *l and sometimes get Ø from our PPn segment. So, if we could write a rule
whereby PPn *l went to Ø in some well-defined environment in Tongan, we would be all
set. But look at the l/r-data for Polynesian. You will find pairs like ‘eight’ and ‘scratch’,
and ‘two’ and ‘vomit’, repeated here for your convenience:

Hawaiian Maori Samoan Tongan English

walu waru walu walu eight
walu waru walu wau scratch
lua rua lua ua two
lua rua lua lua vomit

For the data to involve a conditioned sound change on the way to Tongan, we would
need to figure out an deletion-triggering environment which the proto-segment was in in
words like ‘scratch’ and ‘two’ (since Tongan has Ø in those words), but was not in in words
like ‘eight’ and ‘vomit’. But if we reconstruct ‘eight’ as *walu and ‘scratch’ as *walu as well,
i.e., if we assume there was just one PPn segment in these cases, it will not be possible
to write such a rule. The *l of ‘eight’ and the hypothesized identical *l in ‘scratch’ are
in exactly the same phonological environment (wa u). Ditto for ‘two’ and ‘vomit’. Since
we cannot write a sensible conditioned phonological development that will systematically
8.2. Comparative Reconstruction 159

give rise to the sound correspondences involving /l/ and /r/ in the Polynesian languages,
there must have been two segments in the proto-language.
Since there were two segments, and since we find two segments in the daughter lan-
guages (/l/ and /r/) it seems like one of the correspondence sets should represent Proto-
Polynesian *l and the other should represent Proto-Polynesian *r. The possibility that
the proto-language had Ø for one of these segments—the one reflected as Tongan Ø–can
usually be safely left to one side: it is unlikely that a word with no segment in it in these
diverse phonological environments would have given rise to a bunch of /l/’s and /r/’s in
the daughter languages. Indeed, we can elevate this to a principle:
• Principle III: Deletion is highly preferred over insertion.
But if one of the segments was *l and one was *r, which was which? Again, notice that
in Hawaiian and Samoan we have only /l/ for both PPn *l and PPn *r, no matter which
words had *l and which ones had *r. Similarly, in Maori we have only /r/, regardless of
which words had *l and which ones had *r. That is, PPn *r, whichever words it was in, has
become /l/ in Hawaiian and Samoan, and PPn *l, whichever words it was in, has become
/r/ in Maori. These languages, therefore, can’t help us figure out which words had *l and
which words had *r: they treat all words the same with respect to this contrast.
But Tongan does not: Tongan treats one of the segments as an *l in that, by positing
no change, we can get Tongan to /l/ in words like ‘eight’ and ‘vomit’. Unlike the other
languages, Tongan treats the other PPn segment, the one in ‘scratch’ and ‘two’, differently
from *l. It is the simplest and most economical hypothesis to assume that the segment
that Tongan—the only language which distinguishes the two—treats as /l/ was *l and the
segment that Tongan treats as not /l/, was *r. Thus:
Hawaiian Maori Samoan Tongan Proto-Polynesian
l r l Ø *r
l r l l *l
Let us summarize now what we know so far. We have the following sound correspon-
dences:
Hawaiian Maori Samoan Tongan Proto-Polynesian
m m m m *m
a a a a *a
n n n n *n
u u u u *u
w w w w *w
i i i i *i
p p p p *p
e e e e *e
ʔ k ʔ k *(ʔk)
l r l Ø *r
l r l l *l
160 Chapter 8. Historical Linguistics

which allow us to fully reconstruct (except for the *ʔ/*k issue, which remains unresolved)
the following forms from our initial data set:

Hawaiian Maori Samoan Tongan Proto-Polynesian gloss

manu manu manu manu *manu bird
awa awa awa awa *awa channel
niu niu niu niu *niu coconut
pua pua pua pua *pua flower
mimi mimi mimi mimi *mimi urinate
peʔa peka peʔa peka *pe(ʔk)a bat
muli muri muli mui *muri behind
walu waru walu walu *walu eight
iʔa ika iʔa ika *i(ʔk)a fish
lau rau lau lau *lau leaf
walu waru walu wau *waru scratch
lua rua lua ua *rua two
lua rua lua lua *lua vomit

The forms which we still must account for are thus:

Hawaiian Maori Samoan Tongan English
kani taŋi taŋi taŋi cry
au au au ʔau current
kuna tuna tuna tuna eel species
kae tae tae taʔe excrement
lono roŋo loŋo oŋo hear
ʔuku kutu ʔutu kutu louse
umu umu umu ʔumu oven, earthen
kapu tapu tapu tapu taboo
ako ato ato ʔato thatch, roof
lena reŋa leŋa eŋa yellow, turmeric

The Polynesian data for ‘cry’ provides us with two new correspondence sets. The
vowels are of course clear from what we know already, but the two consonants in the
word give the following:

Hawaiian Maori Samoan Tongan Proto-Polynesian

k t t t
n ŋ ŋ ŋ

Let’s deal with the latter of the two first. Clearly there are two possible segments we
could reconstruct for PPn: *n (as in Hawaiian) or *ŋ (as in Maori, Samoan and Tongan).
Of course, you will recall, we already have a PPn *n (*manu ‘bird’, *niu ‘coconut’) and that
8.2. Comparative Reconstruction 161

*n does not become *ŋ in Maori, Samoan and Tongan. So the segment involved in our new
correspondence set must be either *n or *ŋ, but it can’t be *n, so it must be *ŋ. We know,
therefore, that PPn *ŋ has become Hawaiian /n/ by a sound change:

(8.2) PPn *ŋ > /n/ in Hawaiian

Since nothing happened to *ŋ on the way to Maori, Samoan or Tongan, it is still an /ŋ/ in
those languages.
What of the k = t = t = t correspondence? Clearly we should consider either *t or *k
as possible reconstructions (there is no evidence for anything else), but, again, how could
we select? Much like the *(ʔk) situation, there would appear to be no easy way to decide,
so we must, at least until we see additional evidence, reconstruct *(tk), i.e., a segment
indeterminate as to whether it was *t or *k. Additional data will help us, shortly. But note
right now, that although we can’t tell whether PPn *(ʔk) was a *ʔ or a *k, and we can’t
tell whether PPn *(tk) was a *t or a *k, we know that they weren’t both *k! The word for
‘eel species’ must be reconstructed *(tk)unu and the word for ‘louse’ *(ʔk)u(tk)u. If both
*(tk) and *(ʔk) were actually *k, these reconstructions would be *kunu and *kuku. But
the word-initial *k, which in both cases sits in the phonological environment between the
word boundary to its left and an *u to its right, shows two different outcomes everywhere.
In ‘eel species’ we get k = t = t = t, but in ‘louse’ we get ʔ = k = ʔ = k. There is no way
to get a single PPn segment (*k) to show up two different ways in the daughter languages
when it is in the very same phonological context. So, we don’t know exactly what *(tk)
and *(ʔk) were, but we do know that they were two different segments.
Let’s turn now to the next piece of data, the word for ‘current’, which presents us with
a new correspondence set:

Hawaiian Maori Samoan Tongan Proto-Polynesian

Ø Ø Ø ʔ

Recall at this point ‘Principle III’ above: deletion (or loss) is to be preferred over insertion
(or addition). Notice, moreover, that since we have reconstructed a word-initial *a in the
word for ‘channel’, there is apparently no rule which inserts /ʔ/ before word-initial /a/
in Tongan. So, where did Tongan *ʔ come from? Well, if it didn’t get inserted, in must
be preserved from an earlier segment. All of the evidence we have about that segment
says that it was a *ʔ (since the other Polynesian languages tell us nothing about it all, and
Tongan says it was a glottal stop). Therefore, we must reconstruct PPn *ʔ. The word for
‘current’ was PPn *ʔau, and we need a sound change of the following type:

(8.3) PPn *ʔ > Ø in Hawaiian, Maori, and Samoan

This reconstruction, as trivial as it is, turns out to have serious implications for the
work we have done so far. You will doubtless recall that we have segment which we pro-
visionally reconstructed as *(ʔk), since we could not determine whether it was originally
162 Chapter 8. Historical Linguistics

a *ʔ or a *k. But we now know what happened to a PPn *ʔ: it was lost everywhere but in
Tongan. But our segment *(ʔk) wasn’t lost in any of the daughter languages, so it cannot
have been PPn *ʔ. And if it was either *ʔ or *k, and it wasn’t *ʔ, well, it must have been *k,
by simple logical deduction. We thus have:

Hawaiian Maori Samoan Tongan Proto-Polynesian

Ø Ø Ø ʔ *ʔ
ʔ k ʔ k *k

and the rule:

(8.4) PPn *k > ʔ in Hawaiian and Samoan

But that’s not all: since we now know what happened to PPn *k, we can take a look at an-
other of our ‘indeterminate’ reconstructions, *(tk). You will recall that this reconstruc-
tion is to be interpreted as ‘either *t or *k, but we can’t tell which’. But now we can tell
which, because the correspondence set for which we reconstructed *(tk), which has *k
in Hawaiian and *t everywhere else, is not what happened to PPn *k, which we have just
seen became /ʔ/ in Hawaiian, not /k/, and became /ʔ/ in Samoan, not /t/! Therefore,
our earlier reconstruction *(tk) was not *k, but was either *t or *k: so, it must have been
*t! This gives:

Hawaiian Maori Samoan Tongan Proto-Polynesian

k t t t *t

and a sound change:

(8.5) PPn *t > k in Hawaiian

We can now provide a complete account of the data we started with, reconstructing
each word for Proto-Polynesian, each segment for each correspondence set, and provid-
ing a list of all changes which took place between the time of Proto-Polynesian and now
in each of the daughter languages.7

7
The data we provided does not tell the entire story for Proto-Polynesian. There will be some additional
data given in the ‘exercises’ which follow this chapter.
8.2. Comparative Reconstruction 163

Reconstructed Word-forms:
Hawaiian Maori Samoan Tongan Proto-Polynesian gloss
manu manu manu manu *manu bird
awa awa awa awa *awa channel
niu niu niu niu *niu coconut
pua pua pua pua *pua flower
mimi mimi mimi mimi *mimi urinate
peʔa peka peʔa peka *peka bat
muli muri muli mui *muri behind
kani taŋi taŋi taŋi *taŋi cry
au au au ʔau *ʔau current
kuna tuna tuna tuna *tuna eel species
walu waru walu walu *walu eight
iʔa ika iʔa ika *ika fish
kae tae tae taʔe *taʔe excrement
lono roŋo loŋo oŋo *roŋo hear
lau rau lau lau *lau leaf
ʔuku kutu ʔutu kutu *kutu louse
umu umu umu ʔumu *ʔumu oven, earthen
walu waru walu wau *waru scratch
kapu tapu tapu tapu *tapu taboo
ako ato ato ʔato *ʔato thatch, roof
lua rua lua ua *rua two
lua rua lua lua *lua vomit
lena reŋa leŋa eŋa *reŋa yellow, turmeric
164 Chapter 8. Historical Linguistics

Correspondence Sets:

Hawaiian Maori Samoan Tongan Proto-Polynesian

p p p p *p
k t t t *t
ʔ k ʔ k *k
Ø Ø Ø ʔ *ʔ
m m m m *m
n n n n *n
n ŋ ŋ ŋ *ŋ
l r l l *l
l r l Ø *r
w w w w *w
a a a a *a
e e e e *e
i i i i *i
o o o o *o
u u u u *u

Sound Changes:

(8.6) PPn *ʔ > Ø in Hawaiian, Maori, and Samoan

(8.7) PPn *k > ʔ in Hawaiian and Samoan

(8.8) PPn *t > k in Hawaiian

(8.9) PPn *ŋ > n in Hawaiian

(8.10) PPn *r > l in Hawaiian and Samoan

(8.11) PPn *r > Ø in Tongan

(8.12) PPn *l > r in Maori

8.3 Relative Chronology and Subgrouping

It is important to realize that the ‘sound changes’ listed above are intended to represent
not simply formal relationships between Proto-Polynesian and the daughter languages,
but actual historical events which took place, in real time, during the development from
Proto-Polynesian times (when all of these languages were the same) til the present (at
which time they have diversified via change events). As such, it might be sensible to ask
whether we can know anything about when the changes happened. In a case such as the
8.3. Relative Chronology and Subgrouping 165

Polynesian one, there are essentially no relevant historical records which would allow
us to date these sound changes in absolute terms, however, it does turn out to be the
case that the internal logic of the developments demands that some of the changes must
have taken place before others. This is called a relative chronology of the changes: change x
preceded change y (i.e., is relatively earlier), but we don’t know in what year either actually
happened.
Examine these three changes, e.g.:

• PPn *t > k in Hawaiian

• PPn *k > ʔ in Hawaiian and Samoan

• PPn *ʔ > Ø in Hawaiian, Maori, and Samoan

and imagine that the events took place in the order given above, i.e., that the order above
is the correct relative chronology. Let’s take two PPn words, *kutu ‘louse’ and *taʔe ‘ex-
crement’ and see how they should show up, if this ordering is correct, in Hawaiian, Maori,
and Samoan:
stage Hawaiian Maori Samoan Hawaiian Maori Samoan
PPn *kutu *kutu *kutu *taʔe *taʔe *taʔe
after *t > k in H *kuku *kutu *kutu *kaʔe *taʔe *taʔe
after *k > ʔ in H and S *ʔuʔu *kutu *ʔutu *ʔaʔe *taʔe *taʔe
after *ʔ > Ø in H, M, and S uu kutu utu ae tae tae
actual H, M, and S forms: ʔuku kutu ʔutu kae tae tae

When you compare the forms expected, if this were the order of events, with the actual
forms, you will see that this cannot have been the order in which the events took place:
both Hawaiian forms and one of the Samoan forms is wrong. By contrast, if we hypothe-
size that the order of events was the opposite:

• PPn *ʔ > Ø in Hawaiian, Maori, and Samoan

• PPn *k > ʔ in Hawaiian and Samoan

• PPn *t > k in Hawaiian

we would get the following:

stage Hawaiian Maori Samoan Hawaiian Maori Samoan

PPn *kutu *kutu *kutu *taʔe *taʔe *taʔe
after *ʔ > Ø in H, M, and S *kutu *kutu *kutu *tae *tae *tae
after *k > ʔ in H and S *ʔutu *kutu *ʔutu *tae *tae *tae
after *t > k in H ʔuku kutu ʔutu kae tae tae
actual H, M, and S forms: ʔuku kutu ʔutu kae tae tae
166 Chapter 8. Historical Linguistics

The resulting forms match those we actually observe in the languages. It is, of course,
not possible to give even a relative chronology for all of the changes in this data (and
this is fairly generally the case). The change of PPn *ŋ to /n/ in Hawaiian, for example,
simply doesn’t interact with any other events, and thus could, in principle, haven taken
place at any time within the history of Hawaiian itself. To give a complete solution to a
sound change or reconstruction problem, however, you must state any ordering which is
actually necessary (just as with synchronic phonology problems).
Another question that can be asked about the above ‘sound changes’ concerns cases
such as:

• PPn *ʔ > Ø in Hawaiian, Maori, and Samoan

Does this represent one event, or three events (one in Hawaiian, one in Maori, and one
in Samoan)? While in individual cases it can be difficult to determine, we are most inter-
ested here in making clear and explicit just what the question itself means, rather than
on giving you the tools necessary to decide in every case. In what sense could a change—
an actual event in the history of Hawaiian, Maori, and Samoan— be a single event, rather
than three? You will recall that the changes we have posited represent the step-wise
differentiation of an originally unified proto-language. That is, there was a time when
Hawaiian, Maori, Samoan and Tongan were a single language (this is, after all, what it
means to say the languages are ‘related’). The differentiation of these languages was of
course gradual; brought about by the changes we have discovered (plus others, not part
of the problem set we are considering here). Note that a given event cannot differenti-
ate the proto-language into four distinct daughters. The first change which made these
daughters different from one another can have only split the original undifferentiated
language into two sets: those that show the change, and those that do not. Let us imag-
ine for a moment that the *ʔ > Ø was that first differentiating change (we know that it
has to have preceded some of the other changes, given the relative chronology we estab-
lished moments ago). If we ask what division this event introduced into Proto-Polynesian
it seems clear that this change gave rise to two groups: one group eventually developed
into Tongan, the other group eventually developed into Hawaiian, Maori, and Samoan.
We can represent these relations in a tree:
Proto-Polynesian

*ʔ > Ø Proto-Tongan
Proto-H-M-S
This tree means that Proto-Polynesian differentiated into two groups via the change *ʔ
> Ø, the ancestor of Tongan (‘Proto-Tongan’), which did not undergo the change, and the
ancestor of Hawaiian, Maori, and Samoan (‘Proto-H-M-S’), which did.
A change which we know followed the loss of glottal stop was the change of *k > ʔ in
Hawaiian and Samoan. Again, one could ask whether this was one event, or two. If it was
8.3. Relative Chronology and Subgrouping 167

a single event, then obviously the ancestor of Hawaiian and the ancestor of Samoan were
still ‘the same entity’ at that point, but they were no longer ‘the same entity’ as Maori
or Tongan, since the latter do not show the change. Notice that if Hawaiian and Samoan
were a single entity, to the exclusion of Maori and Tongan, it would also make sense to
place their shared change of *r > l at that stage. This gives us:
Proto-Polynesian

*ʔ > Ø Proto-Tongan
Proto-H-M-S

*k > ʔ Proto-Maori
*r > l
Proto-H-S
Finally, the last of the three changes for which we have a ‘relative chronology’ is that
of *t > k, which took place only in Hawaiian. Note that the change of *ŋ > n is also found
only in Hawaiian. We can also add to our tree the sound changes which took place only
in Maori and only in Tongan, giving us a complete picture of how the changes relate to
the family tree:8
Proto-Polynesian

*ʔ > Ø Proto-Tongan
Proto-H-M-S
*r > Ø
Tongan
*k > ʔ Proto-Maori
*r > l
Proto-H-S *l > r
Maori
*t > k Samoan
*ŋ > n
Hawaiian
A tree of the above type is called a ‘subgrouping’ tree, showing the steps in the differenti-
ation of Proto-Polynesian to these four daughter languages. You can tell what happened
to a given language by working your way up the tree. So, e.g., starting with Maori, we see
8
It is worth pointing out at this juncture that this is not necessarily the correct tree for the Polynesian
language family; it is the correct tree for the data we have considered, which is (1) only a subset of the
available phonological data (there may also be relevant morphological, syntactic and lexical evidence) and
(2) only a subset of the actual Polynesian languages.
168 Chapter 8. Historical Linguistics

that it is characterized by the change of *l > r (in the ‘Maori’ node in the tree), then, as we
move up we reach the ‘Proto-H-S-M’ node, which has the change *ʔ > Ø. The next node
up the tree is Proto-Polynesian itself: therefore, the changes which made Maori different
from PPn are *l > r and *ʔ > Ø.
Subgroups are often given meaningful names, when they figure prominently in dis-
cussions of the history of languages. For example, Proto-Polynesian is itself a well-named
subgroup of the Proto-Oceanic language family, itself a subgroup of the Proto-Austronesian
language family, etc.

8.4 Cultural Reconstruction

One final note regarding our reconstructions. They tell us much about the linguistic
structure of Proto-Polynesian, but they also provide potentially interesting evidence of a
less directly linguistic type. For example, imagine we wanted to know where the Proto-
Polynesian people lived. Some of the reconstructed lexical items could be quite help-
ful in identifying likely homelands: e.g., *niu coconut and *tuna eel species tell us the
Proto-Polynesians likely lived somewhere where coconuts were present and the partic-
ular species of eel designated by *tuna resided. So, probably not the North Pole! Other
reconstructed lexical items tell us about technological developments among the Proto-
Polynesians (*hama ‘outrigger for canoe’, e.g.) or social structure (kinship systems, names
for social rank and status, etc.). In some cases, a fairly rich understanding of the flora,
fauna, physical geography, and social life of the speakers of a reconstructed language can
be established.

8.5 Summary
The very complicated example which we have just worked through reveals an important
phenomenon that we mentioned earier in the chapter: sound change is regular. The sim-
ilarities between the Polynesian languages we have looked at can be explained by starting
with a well-defined set of Proto-Polynesian forms and applying to these an ordered set of
regular changes, different for each of the languages. It is the regularity of change which
makes it possible for us to demonstrate that languages are related and thus that they fall
into groups of relatedness, called ‘language families.’ Some of the properties of the lan-
guages ancestral to the set of languages within a family can be determined by the use of
the comparative method, which thereby reveals aspects of the history of human linguistic
(and cultural and historical) systems which we could not discover in any other way.
8.6. Exercises 169

8.6 Exercises
A. Additional Polynesian Comparative Data
Examine the data below. Consider the discussion of PPn in the chapter. Establishing
any new sound correspondences and sound changes required to get the data, think up an
account for this new data. An account includes (1) a set of Proto-Polynesian reconstruc-
tions, (2) any new sound correspondences the data reveal, and (3) any new sound changes
you need to posit to produce the data. Of course, if any ordering is required for the sound
changes, you need to figure that out as well.

Hawaiian Maori Samoan Tongan English

ani aŋi aŋi aŋi blow
hamu hamu samu hamu eat scraps
maka mata mata mata eye
ʔai kai ʔai kai food, eat
ake ate ate ʔate liver
limu rimu limu limu moss, seaweed
nuku ŋutu ŋutu ŋutu mouth, beak
piko pito pito pito navel
ihu ihu isu ihu nose
ama ama ama hama outrigger
ʔiako kiato ʔiato kiato outrigger boom
ua ua ua ʔuha rain
ala ara ala hala road
kai tai tai tahi sea
ʔili kiri ʔili kili skin
nalu ŋaru ŋalu ŋalu wave

a. One of the new correspondence sets is:

1. Ø : Ø : Ø : h, for which one must reconstruct PPn *h

2. Ø : h : Ø : h, for which one must reconstruct PPn *h

3. Ø : Ø : h : h, for which one must reconstruct PPn *h

4. h : Ø : Ø : h, for which one must reconstruct PPn *h

b. Another one of the new correspondence sets is:

1. h : s : s : h, for which one must reconstruct PPn *s

2. h : h : s : s, for which one must reconstruct PPn *s

3. h : h : s : h, for which one must reconstruct PPn *s

170 Chapter 8. Historical Linguistics

4. s : h : s : h, for which one must reconstruct PPn *s

c. Given these reconstructions, one needs sound changes such as:

1. *s > h in Hawaiian, Maori, and Tongan

2. *h > Ø in Hawaiian, Maori, and Samoan

3. Both (1) and (2).

4. No sound changes are required.

d. The reconstruction of the PPn words for ‘rain’, ‘nose’ and ‘sea’ must be:

1. *ʔusa, *ihu and *tahi

2. *ʔuha, *ihu and *tasi

3. *ʔuha, *isu and *tahi

4. *ʔuha, *ihu and *tahi

B. Yet More Polynesian Comparative Data

In the data below you see evidence for the development of a segment reconstructed
as PPn *f. Your task is to figure out (1) what are the correspondence sets relating to this
reconstruction and (2) what, if any, conditioned sound change is involved.

Hawaiian Maori Samoan Tongan English

ahi ahi afi afi fire
haa hw aa faa faa four
hua hua fua fua fruit
hulu huru fulu fulu hair
heʔe hw eke feʔe feke octopus
hw ai fai fai rayfish
hiku hw itu fitu fitu seven
haku hw atu fatu fatu stone
niho niho nifo nifo tooth
honu honu fonu turtle
mahana mahana mafana mafana warm
8.6. Exercises 171

a. The correspondence set(s) involving PPn *f are:

1. h : h : f : f only

2. h : hw : f : f only

3. h : h : f : f and h : hw : f : f

4. h : h(w) : f : f only

b. The conditioned sound change involved in the development of PPn *f is:

1. *f > Maori hw before front vowels; all other PPn *f became h

2. word-initial *f > Maori hw before round vowels; all other PPn *f became h

3. word-initial *f > Maori hw before non-round vowels; all other PPn *f became h

4. *f > Maori h before a; all other PPn *f became hw

c. There is also an unconditioned development of *f in Hawaiian. It is:

1. PPn *f > k in Hawaiian.

2. PPn *f > hw in Hawaiian.

3. PPn *f > s in Hawaiian.

4. PPn *f > h in Hawaiian.

d. Given that in Exercise A we posited a rule which said that PPn *h got lost in Hawaiian,
Maori, and Samoan, does the rule you established in (b) and (c) need to be ordered relative
to that loss?

1. Yes, the development of *f to /h/ in Hawaiian and Maori needs to precede h-loss.

2. Yes, the development of *f to /h/ in Hawaiian and Maori needs to follow h-loss.

3. Yes, the development of *f to /h/ in Hawaiian and Maori needs to both follow and
precede h-loss.

4. No, the two ordering of the two events cannot be determined.

172 Chapter 8. Historical Linguistics

C. Fijian
Here are a bunch of words in Fijian which correspond to PPn words we have recon-
structed. Fijian is not considered to be a Polynesian language, because it cannot be de-
rived systematically from Proto-Polynesian, but it is related more distantly to the Polyne-
sian languages. Examine all of the evidence below to find what indicates that, in spite of
its similarity to the Polynesian languages, Fijian cannot be a Polynesian language, if Proto-
Polynesian is correctly reconstructed. (Note: Fijian [m b], [n d], and [ŋ g] are pre-nasalized
voiced stops. [n r] is a prenasalized [r].)

Fijian PPn gloss

manu *manu bird
niu *niu coconut
m
beka *peka bat
muri *muri behind
taŋi *taŋi cry
n
duna *tuna eel sp.
walu *walu eight
n
dee *taʔe excrement
mata *mata eye
ika *ika fish
vaa *faa four
vua *fua fruit
vulu *fulu hair
roŋo *roŋo hear
n
rau *lau leaf
kutu *kutu louse
ðama *hama outrigger
uða *ʔuha rain
vai *fai rayfish
sala *hala road
vitu *fitu seven
vatu *fatu stone
tam bu *tapu taboo
vonu *fonu turtle
rua *rua two
lua *lua vomit
reŋa *reŋa yellow, turmeric

a. Is the fact that Fijian has both prenasalized and plain voiceless stops corresponding
to Proto-Polynesian plain voiceless stops a problem for considering Fijian a Polynesian
language?
8.6. Exercises 173

1. Yes, there is no way to write systematic sound changes which will account for when
Fijian has a prenasalized stop and when it has a plain voiceless stop, so they can’t
derive from the same PPn segment.

2. Yes, because prenasalized stops are weird and so they must be old.

3. Yes, because Fijian has more kinds of consonants than Proto-Polynesian did.

4. No, no problem.

b. Is the fact that Fijian has both /s/ and /ð/ corresponding to PPn *h a problem for con-
sidering Fijian a Polynesian language?

1. Yes, there is no way to write systematic sound changes which will account for when
Fijian has /s/ and when it has /ð/, so they can’t derive from the same PPn segment.

2. Yes, because /ð/ is weird and so is must be old.

3. Yes, because Fijian has more kinds of consonants than Proto-Polynesian did.

4. No, no problem.
 Chapter 9
Language Acquisition

Contents
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.2 The Earliest Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.2.1 Perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.2.2 VOT Contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.2.3 Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.3 Later Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.4 Acquiring a Lexicon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.5 Theories of Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.6 The Critical Period Hypothesis . . . . . . . . . . . . . . . . . . . . . . . 182
9.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

9.1 Introduction
The process of language acquisition sits at the very center of the major concerns of mod-
ern linguistic theory. It is fundamental both to the problem of characterizing the nature
of grammar (some part of which is assumed to be an innate endowment of humans, an-
other part learned) and to an understanding of language change (and therefore, of lan-
guage variation).
While linguistic research has made considerable progress in the study of certain as-
pects of acquisition, the picture is far from complete. This is partly because children,
especially young children, are rather difficult to work with, and partly because experi-
mentation on human subjects is not always possible in general. (Oddly, parents will not

175
176 Chapter 9. Language Acquisition

generally allow their children to undergo useful experiments which involve procedures
like injecting radioactive fluid into their children’s brains so that we can get an image of
the activity in there, solely in the interest of scientific curiosity!) In addition, children
are even worse than adults at conveying anything useful about what’s going with them
linguistically. In the early stages they can’t talk (and we are unable to extract information
from bubbles and various gastric noises). Later on, they talk after a fashion, but it turns
out they generally have no idea what you’re talking about when you ask them perfectly
reasonable questions (like: ‘Is that an allophone of this?’). Consequently, with the excep-
tion of some nifty non-invasive experiments for speech perception, most of our data on
acquisition comes from the same sources as our data for adult grammars – production
and comprehension – with all of the associated drawbacks. For reasons we will examine
shortly, using children’s production is even more fraught with uncertainty than using
adult production and children’s comprehension data is not appreciably better.
Before discussing acquisition data and methodology in more detail, we will outline
a few points about knowledge of language which are relevant to the question of which
aspects of that knowledge are innate (part of UG), and which aspects must be learned
from the environment. An innate component is suggested by the following:

• Aspects of linguistic knowledge appear to be present at birth.

• The development of language mirrors the development of other, innate abilities.

• The system is too complex to be acquired by children in the time that we know they
acquire it.

• The data that children receive is insufficiently robust to allow them to construct a
grammatical system without relying on some innate knowledge.

A learned portion is suggested by the very obvious fact that children acquire what-
ever language (or languages) they are exposed to in their environment. It is not quite
as straightforward as simply memorizing vocabulary items, of course, but it seems to be
true that a child will be unable to acquire language if he or she is not in some language
environment.

9.2 The Earliest Stages

9.2.1 Perception
While infants are not known for their stimulating conversational repartee, they do demon-
strate some surprisingly sophisticated, and language-specific, apparently, knowledge of
speech sounds. To study early (from within a few days or even hours of birth to about 4
months) infant speech perception, a particular type of experimental procedure was de-
veloped – the High Amplitude Sucking test (HAS). This test takes advantage of one of the
9.2. The Earliest Stages 177

few activities babies of this age are capable of, as well as of the intriguing fact that babies
appear to suck faster on pacifiers when they’re interested in or excited by something than
when they are bored. In the HAS experiment, a pacifier is attached to a pressure trans-
ducer which records sucking rate. A baseline sucking rate is established for each infant
and subsequently the infant is exposed to some stimulus. When first exposed to the new
stimulus, the sucking rate increases, but as the same stimulus is played repeatedly, the
sucking rate drops off (the infant becomes habituated). When the ‘old’ stimulus is re-
placed with a new stimulus, the sucking rate increases (and then drops off again after a
certain period of time).
Using this technique, experimenters have tested infant perception of a number of dif-
ferences between natural language speech sounds. For example, when infants are played
the sound [ba] in a HAS experiment and then the stimulus is changed to [pa], infants re-
act with increased sucking rate, indicating that they perceive [pa] to be a new stimulus
(thus they distinguish between [ba] and [pa]). This same type of experiment has been
performed on infants from a range of different language environments and the stimuli
have included a range of natural language speech sounds. The results of these exper-
iments suggest that infants from the earliest ages are already sensitive to the possible
sound distinctions in language. Moreover, their discrimination, at this early age, is not at
all dependent upon the language environment that they have been in. For example, En-
glish environment infants are as capable as Hindi environment infants at distinguishing
between sounds that are present in Hindi but not in English and vice versa.
Results such as these suggest that the ability to perceive distinctions in natural lan-
guage speech sounds is innate rather than learned. On the other hand, one might per-
fectly well argue that all that these experiments show is that infants have very acute hear-
ing and that they are not discriminating between speech sounds, merely between sounds.
In order to try and resolve this question, experimenters have run additional tests based
on what we know of adult perception of certain distinctions.

9.2.2 VOT Contrasts

In the HAS experiment described above with [ba]/[pa], the difference that infants are per-
ceiving is one of voicing on the initial consonant. The true description of what ‘voicing’
consists of for stops is a little more complicated. It turns out that the real cue for the lis-
tener is not simply whether or not there is vocal fold vibration but, instead, exactly when
that vibration starts with respect to when you release your lips (in the case of a bilabial
stop). For a listener to perceive a bilabial stop as voiced (i.e. as [ba]), vocal fold vibration
must start anywhere between about 10 milliseconds before (roughly at the start of the
bilabial closure for [b]) to 10 milliseconds after the lips are released. In the case of [pa],
speakers usually wait until about 20 milliseconds into the [a] before they start voicing
(that is, 20 milliseconds after the release). What we have been describing here is Voice
Onset Time (VOT). VOT is literally when voicing starts with respect to the release of the
stop closure (in the case of bilabials, that is when you open your lips). Again, for bilabials,
178 Chapter 9. Language Acquisition

if you start voicing anytime between 10 milliseconds before the stop release (-10 msec.
VOT) and 10 milliseconds after (+ 10 msec. VOT), an adult listener will hear [ba], while
if you start voicing sometime after 20 milliseconds (+ 20 msec. VOT) after the stop re-
lease, adults will hear [pa]. Experimenters based a number of acquisition experiments
upon these observations of adult behavior, as well as on one additional factor which we
will talk about at the same time as the experiment itself. These experiments were geared
toward answering the question that arose from the perception discrimination experi-
ments described earlier, i.e., is infant reaction simply accurate audition or is it the result
of knowledge particular to the language faculty.
The experiment proceeds in a similar fashion as the simple [pa]/[ba] experiment but
with some fine tuning. First, a series of [ba]’s is played to the infant but the VOT of the
[ba]’s is different. The first set contains [ba]’s which start voicing at 10 milliseconds prior
to release. When the infant’s sucking rate decreases, a second set of [ba]’s which start
voicing 10 milliseconds after release is played. Critically, the infants show no interest in
this 20 millisecond change of voicing – they do not increase sucking rate when the second
type is introduced. Similarly if you play a series of [pa]’s, the first set of which has voicing
starting 20 milliseconds after the stop release and the second set with voicing starting
40 milliseconds after the stop release, the infant sucking rate continues to fall as if there
was no change in stimulus. Note that, in both cases ([ba]’s and [pa]’s), there was a 20
millisecond difference in VOT between the first and second sets.
One more set of stimuli, which differ only by 20 milliseconds, is now added to the pic-
ture. The infant is played a [ba] with voicing which starts simulaneously with release. Then
a second stimulus is introduced in which voicing starts 20 milliseconds after release. In
this case, the infant does react as if there is a new stimulus and the sucking rate increases.
Now in this and the previous two cases, voicing was shifted by exactly 20 milliseconds but
the the infant treated two of the 20 millisecond shifts as if nothing new happened, unlike
the third shift. Since the stimuli were equidistant from one another, the only explanation
for the fact that the child treats only one pair of them as if they were different is if the
child is particularly sensitive to the difference between 0 millisecond delay (after stop
release) and a 20 millisecond delay (after stop release) – precisely the difference between
a [b] and a [p] in adult language. This phenomenon is called ‘categorical perception’ and,
unlike the first results cited for simple [pa]/[ba], cannot be explained by simply saying
that the stimuli were acoustically different. (Note, again, that every shift was between
acoustically different tokens of 20 msec. but only 1 of these shifts was treated by the
infants as making a difference.) These results seem to indicate something about the pres-
ence of knowledge specific to language in infants. To emphasize this, HAS experiments
have been done using pure tones, rather than speech sounds. No categorical difference
of the type described above exists for the tone-based experiments.1
1
Nothing is ever clear-cut, unfortunately. Chinchillas are also very good at categorical perception ex-
periments, whatever that fact means.
9.3. Later Stages 179

9.2.3 Production
From the production perspective, there is little going on at the earliest ages. Most of the
sounds produced by infants up to about 3-5 months fall into the attractively labelled ‘veg-
etative noises’ category. After this time, however, infants begin to make sounds which are
somewhat more related to speech sounds. At this point, their production is called ‘bab-
bling.’ The babbling sounds are not linguistically significant in that they are not produced
consistently with respect to naming anything, however listeners hear these sounds as
speech-like. While babies in this stage produce a very wide variety of speech-like sounds,
the majority of the consonant sounds they make are from the set [h d b m t g w n k]. Once
again, this set is language environment independent – all babies babble similarly regard-
less of the sounds heard in the particular environment. Given which sounds are most
‘popular’, as indicated by the set above, it seems likely that babies are producing most of-
ten the sounds which require the least sophisticated motor coordination. Sounds which
require fine motor control and airflow regulation (many fricatives, for example) are typ-
ically only produced in an adult-like fashion much later (possibly much much later). In
general, the babbling stage is felt to be a type of ‘practice’ for the real thing later on.

9.3 Later Stages

From about 12 to 18 months, children start producing some recognizable speech (all things
considered).2 Their earliest utterances consist of single words. As a result, this is called
the ‘one word stage.’ The assumption is that children actually intend something like a
sentence but can only manage a little part of it. The latter is due, in our opinion, to gen-
eral limitations that it is well known children have, such as limited short-term memory,
limited motor coordination, limited attention spans, and so on. If one is limited to one
word, a good strategy is to pick the word that will be key in getting your message across
– this appears to be what children do.
Interestingly, comprehension far exceeds production at this point in time. In general,
production does not catch up to comprehension for a number of years although the gap
becomes narrower as time wears on. (One could conceivably make the point that pro-
duction never actually catches up to comprehension, of course, given adult performance
vs. competence.) Children at the one-word stage are able to interpret at least simple sen-
tences relatively well even though they cannot produce such long utterances themselves.
In a frequently depicted experiment, English environment children at the one-word stage
are placed in front of two TV screens. One of the screens will show Big Bird tickling Cookie
Monster and the other will show Cookie Monster tickling Big Bird. (Success is predicated
upon familiarity with these characters, obviously.) A loudspeaker will announce some-
thing like ‘Look! Big Bird is tickling Cookie Monster!’ and the children turn toward the
appropriate screen and concentrate on it for some time. This shows that the children
2
Note that an awful lot must have gone on to get to this point from the earlier stage. The details of all
of this (those that we’ve managed to figure out, anyway) are too many to include here.
180 Chapter 9. Language Acquisition

already have some relatively advanced knowledge of the syntactic structure of English
where the subject (the one doing the action) comes first in a simple sentence. This knowl-
edge is certainly not displayed in their very limited production at the time, supporting the
position that production should not form the basis for determining children’s language
development (a development of the mind, not the articulators).
A few months after the one word stage, the next stage in production occurs. This is
the ‘two word stage’ characterized by – you guessed it – utterances of only two words. In
this case, the strings are frequently, although not necessarily, composed of a noun and a
verb.

• Allgone sock.

• Baby chair.

• Doggie bark.

• No eat.

• Throw ball.

Again, these two word strings are assumed to represent more complete utterances (sen-
tences) but due to the limitations mentioned above, only a couple of key words are there.
There is no three word stage to follow the two word stage. Instead, some months later,
utterances which have varying numbers of words but virtually no function words begin
to be produced. Because this is similar to the style of old-time telegraph messages (where
all but the most important words were omitted for reasons of cost), this stage is called the
‘telegraphic stage.’ This stage might include strings like the following.

• Man ride bus today.

• Me wanna show Mommy.

• Me put it back.

• No do that again.

Beyond this point, no very clear distinctions can be made in terms of stages of pro-
duction. The number of vocabulary items soars and there is constant progress toward
approximating adult speech. By the time children are about 5 years old, their grammars
appear little different from adult grammars.3
3
Note that we said ‘grammars’, not ‘articulation’, ‘vocabulary’ or ‘rhetorical ability.’
9.4. Acquiring a Lexicon 181

9.4 Acquiring a Lexicon

Building up a lexicon is, in some ways, not only the most fundamental but also the most
difficult part of acquiring language. For necessary reasons, we believe that UG provides
much of the structure of the computational system, so that it doesn’t have to be learned.
However, the lexicon must just as obviously be acquired through experience. Even if one
assumes (as we do) that there is an innate set of semantic features available in UG, the
tricky question is: How do children (and adults) form hypotheses about which semantic
features should be grouped together to make up a coherent semantic representation for a
lexical entry? There is evidence from production and comprehension that children in the
acquisition stage do not ‘zero in’ on an adult-like representations the first time around.
Interestingly, children seem both to overgeneralize and undergeneralize. An example of
the former is a child who refers to any number of (non-dog) animals as ‘dog.’ An example
of the latter is a child who only refers to his or her own dog as ‘dog.’ There are a number
of possible explanations for both of these phenomena, only some of which have anything
to do with the accuracy of children’s semantic representations. (We will leave these to
you as a ‘thought exercise’ for now and return to it in question form at the end of the
chapter.) In spite of these apparent false starts, the vast majority of children converge on
something like the adult semantic representation of ‘dog’ as far as we know. In addition,
however, they all perform what seem like magical leaps in the area of semantic represen-
tation. When they finally converge on ‘dog,’ they are able not only to identify virtually all
and only the disparate members of the canine species as ‘dogs’ but also to identify stuffed
animals that look like dogs as ‘dog,’ photographs, drawings and paintings of dogs as ‘dog,’
and things that bark in the night as ‘dog.’ Notice that the physical reality of stuffed toys,
pictures, and noises are quite different both from one another and from a live dog. The
evidence of ‘doghood’ must come from a sufficient number of abstract semantic features
that one can remove a good many features and still be left with something that is ‘dog.’
This very brief discussion of semantic features brings up the fundamental question of
whether or not any two individuals have exactly the same featural representation for a
particular morpheme in the same language, let alone for what we think of as an equiva-
lent morpheme in some other language (cf. ‘dog’ and the French equivalent ‘chien’). The
assumption that there is a universal semantic feature set available does not entail the fur-
ther assumption that everyone constructs identical semantic feature bundles. It seems
likely, once again, that functional morphemes and content morphemes may be differ-
ent in the domain of acquisition. Functional morphemes tend to be much more circum-
scribed, something that would make them easier to acquire. For content morphemes, on
the other hand, experience must play a large role in what features are chosen to represent
a particular morpheme. Since no two individuals have identical experiences, one might
hypothesize that there exist a number of differences between semantic feature bundles
across individuals. This is probably more likely in the case of abstract notions, such as
‘like,’ ‘despair,’ ‘acquaintance,’ and ‘friend’ than for concrete objects such as ‘dog’ or ‘ta-
ble,’ but may well be true of all content morphemes. However, if such differences exist,
182 Chapter 9. Language Acquisition

they seem to be, for all practical purposes such as casual conversation, insignificant.

9.5 Theories of Acquisition

Over the years, a number of theories have been put forth to account for the fact that chil-
dren acquire language. One of these theories was the imitation theory. It held that children
learned language by imitating the people speaking around them. This theory has an obvi-
ous disadvantage in that it seems to predict that we should only be able to say things that
we’ve heard others say (if all our knowledge is due to imitation, what else could we do?).
As the examples from the two word and telegraphic stages suggest, children are doing
something other than imitating people around them. Few adults produce sentences such
as ‘Man ride bus today’ or ‘Allgone sock.’ Even the more ‘subtle’ aspects of language devel-
opment cannot be due to imitation. For example, children’s morphological development
shows that they at some point deduce the existence of regular, productive morphology
and incorporate the appropriate rules into their grammar. This leads to output such as
the following.

Earlier stage Later stage

dogs dogs
cats cats
mans men
childs children
kicked kicked
goed went
runned ran

It is only at a later stage that lexical exceptions are ‘recorded’ (stored as exceptions in the
lexicon) and then produced appropriately. There is no explanation for non-adult forms
in the first column under a theory which says that children imitate those around them
(i.e. adults).
Another theory, the reinforcement theory, popular with behavioral psychologists ear-
lier in this century, held that children made some noise, and if they received positive re-
inforcement from their caregivers for making this sound, they made it some more. Even-
tually, through the process of continual adult reinforcement, children came to say words
and eventually sentences. An interesting dialogue between parent and small child reveals
some problems with this theory.

Child: Nobody don’t like me.

Parent: You mean ‘nobody likes me.’

Child: Nobody don’t like me.

9.5. Theories of Acquisition 183

Parent: Say ‘nobody likes me.’

Child: Nobody don’t like me.’

Parent: Say ‘nobody.’

Child: Nobody.

Parent: Say ‘likes.’

Child: Likes.

Parent: Say ‘me.’

Child: Me. Nobody don’t like me.

Note first that any normal (i.e., non-linguist) parent would respond to this child’s pathetic
assertion of lack of self-worth with a supportive ‘Oh, come on, everybody likes you.’ or
some such statement. This dialogue thus reveals two difficulties for the ‘reinforcement
theory.’ First, parents generally reinforce children not for the form of their assertions but
for the content. For example, when, as a child, one of the authors said ‘Toni is stupid’ (Toni
is his sister), he did not get a nice pat on the head from his mother because of his proper
use of the verb ‘to be.’ He got, instead, some negative reinforcement in the form of a swat
on the behind. On the other hand, when he said ‘I wuves you mommy’ he got a big hug,
in spite of the ill-formedness of his utterance (and the fact that he had his eye on some
more candy). Reinforcement will never explain the acquisition of the grammar.
This dialogue is equally bad for the ‘imitation theory.’ The parent offers several, care-
fully presented models for the child to imitate and the child stubbornly persists in saying
it his way as opposed to imitating. And again, who could the child be imitating when
he says ‘nobody don’t likes me’? This particular child was born into an environment in
which there were no speakers of any dialect which produced such forms. Instead, what
such dialogues seem to reveal is the same aspect of acquisition that the morphology ex-
amples revealed. At any particular point, the child has a grammar – it’s just a different
grammar than an adult grammar. The child’s motivation is neither to get some particu-
lar reinforcement nor to accurately imitate an adult, but rather to express him/herself,
using the grammar he has constructed.
Regardless of what kind of input they receive, a child learning an English type gram-
mar will produces sentences like ‘nobody don’t like me.’ This can be explained if this
type of output reflects a necessary intermediate step between the innate linguistic sys-
tem, present in all speakers at birth, and the fully-acquired English-type grammar. The
innateness hypothesis, as this theory is called, has many distinct advantages over its ear-
lier competitors. Theories which have the child starting out with nothing in their heads
(so-called tabula rasa or ‘clean slate’ theories) not only can’t explain the HAS results on
VOT, they would also never predict dialogues such as the one cited above. The fact that
184 Chapter 9. Language Acquisition

sentences with this structure are uttered by every acquirer of English tells us something
important: there is an innate linguistic system. The role of caregivers and others within
hearing distance of the child is far less obvious than many people supposed.4 Speakers
in the child’s environment (caregivers, siblings, and so on) certainly do provide the child
with all important data, but they cannot influence what the child does with the data they
receive – this appears to be regulated by some unconscious, internal acquisition mecha-
nism of the child.
If we take the facts above together with the obvious complexity of human grammars
(as you’ve seen in this course) and the relatively poor quality of the data that children
have to work with (consider all of the performance errors in natural speech), a clear con-
clusion may be drawn. Children are really good at learning languages because they are
preprogammed to do so. They are born with innate knowledge of precisely what kinds of
things they might find in a grammar and their learning task is more a matter of weeding
out elements they don’t need (because it isn’t in the particular language they’re being ex-
posed to) than a matter of discovering facts of great complexity on the basis of really bad
evidence. While the individual lexical items of the language to which they are exposed
must be learned, the structural possibilities for all human languages are present in their
minds at birth – which we term UG.
It seems appropriate to end this section by referring you back to the waveform and
spectrogram at the beginning of Chapter 2. Those show the physical reality of language
– no discrete sounds, no words, no sentences, no meaning – just air pressure variation.
The human acquirer gets air pressure variations and ends up with a mental system of in-
credible complexity before they can successfully tie their shoes. An innate component to
knowledge of language seems a logical necessity given this.

9.6 The Critical Period Hypothesis

The developmental stages described in this chapter are stages that every human child
goes through (barring severe and specific neurological abnormalities and assuming nor-
mal exposure to language data in their environment). While none of the ages are abso-
lute for any one stage, the order of acquisition is essentially the same for all children and
the ages at which certain aspects of language develop are approximately the same. In
this sense, language acquisition is similar to walking. All children learn to walk in pretty
much the same way at pretty much the same time – speeding up the process through out-
side intervention is not possible, as far as we know.5 Many innate ‘abilities’ are on-line, in
some sense, at birth. Human hearing is an example. Others develop into their adult-like
state soon after birth, such as vision. Still others, such as walking, arguably, take a rather
longer time to become adult-like. When developmental time after birth is required, it is
4
The pendulum has swung back and forth on whether caregiver talk (‘baby talk’, ‘motherese’) has pos-
itive or negative effects on acquisition. The tremendous variation across cultures in this domain, however,
suggests that any possible effects would be insignificant.
5
It is interesting to speculate on whether a child will walk if they are not surrounded by ‘walkers.’
9.7. Summary 185

also sometimes the case that environmental input is crucial to proper development. Vi-
sion, in at least some species, is like this. In such cases, there is said to be a ’critical period’
for development.6 This brings up the interesting question of whether there is a ‘critical
period’ for acquiring language. That is, does language, specifically, one’s first language,
have to be acquired within a certain time frame?
There is no humane experiment one can perform in order to help answer this ques-
tion. The only information that might be relevant comes from a handful of tragic cases
in which children have been deprived of linguistic stimulus during the normal develop-
mental years. The most famous and most carefully studied of these cases was of a girl,
Genie, who was deliberately deprived of almost all normal human interaction through
her first fourteen years. A linguist, Susan Curtiss, worked with Genie for some period of
time after she was rescued from her home environment. Although Genie made consid-
erable progress, her acquisition was ‘spotty’ and much more successful in the realm of
vocabulary than of structure. Her linguistic development was unlike that of a child from
a normal environment and remained far from adult-like. Few conclusions can be drawn
about the notion of a critical age for acquisition, though, due to the existence of many
other variables such as the impact of psychological trauma and neurological problems.
In general, the multitude of factors involved in cases such as these do not allow us to get
a clear picture of what might be going on.

9.7 Summary
Human knowledge of language is a system of great complexity and we are far from un-
derstanding all of its intricate mechanisms. On the other hand, how hard can it be if a
three year old can do it? (Note again that, for a three year old, tying one’s shoe is an in-
surmountable obstacle.) The answer appears to be ‘Not hard, if you’re human.’ but ‘Not
likely, if you’re not.’ Humans also seem to treat bipedal locomotion as a trivial activity
whereas an attempt to fly by flapping one’s arms is doomed to failure. All species do what
they are genetically programmed to do. And all species have some mechanism that allows
them to communicate amongst themselves. The human linguistic system is a property of
the human genetic code. Language, in the sense it has in linguistics – a mental system
made up of both genetic and experiential components – is thus a defining property of
what it means to be human.

9.8 Exercises
1. If a child in the early stages of acquistion refers only to his/her dog as ‘dog’, what
might this indicate?
6
The ‘critical’ part may be the maturation of some aspect of the organism, some input from the envi-
ronment, or both.
186 Chapter 9. Language Acquisition

a) The child’s grammar takes ‘Spot’ and converts it to ‘dog’ by regular, phono-
logical rules.
b) The child has taken ‘dog’ to refer to the very specific set of attributes of his/her
particular dog (i.e., hair of a particular length and color, bulgy brown eyes,
wart on the chin, and so on).
c) The child will never converge on some adult-like meaning of ‘dog.’
d) The child initially assumes that the sound [d] refers to hairy objects, and [g]
refers to 4 legged objects.

2. Which of the following is true?

a) Some morphemes have meaning and some do not.

b) Some morphemes have ‘content’ semantics and some ‘functional’ semantics.
c) All morphemes have both ‘content’ semantics and ‘functional’ semantics.
d) The semantics of simple morphemes are derived during the course of a deriva-
tion.

3. If a child in the early stages of acquisition refers to a bunch of ‘non-dog’ animals as

‘dog,’ this indicates that:

a) the child’s knowledge of language includes only phonological features at this

point, not semantic features.
b) the child has heard everyone else refer to those other animals as ‘dogs.’
c) the child cannot distinguish between ‘non-dog’ animals and dogs.
d) the child has overgeneralized — later he/she will refine their semantic feature
set for ‘dog’ to be more specific.

4. During the course of language acquisition

a) production is constantly ahead of comprehension

b) comprehension and production proceed at the same pace
c) comprehension outstrips production by far at the earlier stages
d) comprehension starts to lag behind production after awhile