Morphology 1

• Introduction
• Morphology
• Morphological Analysis (MA)
• Using FS techniques in MA
• Automatic learning of the morphology of a
language

NLP Morphology 1
 Morphology 2
• Morphology
• Structure of a word as a composition of morphemes
• Related to word formation rules
• Functions
• Inflection
• Derivation
• Composition
• Result of morphologic analysis
• Morphosyntactic categorization (POS)
• e.g. Parole tagset (VMIP1S0), more than 150 categories for Spanish
• e.g. Penn Treebank tagset (VBD), about 30 categories for English
• Morphological features
• Number, case, gender, lexical functions

NLP Morphology 2
 Morphology 3
• Morphologic analysis
• Decompose a word into a concatenation of
morphemes
• Usually some of the morphemes contain the meaning
• One (root or stem) in flexion and derivation
• More than one in composition
• The other (affixes) provide morphological features
• Problems
• Phonological alterations in morpheme concatenation
• Morphotactics
• Which morphemes can be concatenated with which others

NLP Morphology 3
 Morphology 4
• Problems
• Affixes
• Suffixes, prefixes, infixes, interfixes
• Inflectional affixes ≠ derivational affixes
• Derivation implies sometimes a semantic change not always
predictible
• Meaning extensions
• Lexical rules
• A derivativational suffix can be followed by an inflectional one
• love => lover => lovers
• Inflection does not change POS, sometimes derivation does
• Inflection affects other words in the sentence
• agreement

NLP Morphology 4
 Morphology 5

• Morphotactics
• Word formation rules
• Valid combinations between morphemes
• Simple concatenation
• Complex models root/pattern
• Language dependency regularity
• Phonological alterations (Morphophonology)
• Changes when concatenating morphemes
• Source: Phonology, morphology, orthography
• variable in number and complexity
• e.g. vocalic harmony

NLP Morphology 5
 Morphology 6
Morphemes
• 1 morpheme:
Evitar ( verb to avoid)
• 2 morphemes:
• evitable = evitar + able (adj: can be avoided)
• 3 morphemes:
• inevitable = in + evitar + able
(adj: cannot be avoided)
• 4 morphemes:
• inevitabilidad = in + evitar + able + idad
(noun: cannot be avoided)

NLP Morphology 6
 Morphology 7
Inflectional Morphology
• number
• house houses
• cheval chevaux
• casa casas
• verbal form
• walk walkes walked walking
• amo amas aman ...
• gender
• niño niña
NLP Morphology 7
 Morphology 8
Derivational Morphology
• Form
• Without change barcelonés
• Prefix inevitable
• Suffix importantísimo
• Source
• verb => adjective tardar => tardío
• verb => noun sufrir => sufrimiento
• noun => noun actor => actorazo
• noun => adjective atleta => atlético
• adjective => adjective rojo => rojizo
• adjective => adverb alegre => alegremente

NLP Morphology 8
 Morphological Analysis 1
Types of morphological analyzers
Formaries
• Dictionaries of word forms
+ efficiency
+ Languages with few variants (e.g. English)
+ extensibility
+ Possibility of building and maintenance from a
morphological generator
– Languages with high flexive variation
– derivation, composition
• FS techniques
• FSA
• 1 level analyzers
• FST
• > 1 level analyzers

NLP Morphology 9
 Morphological Analysis 2

Morphological analyzers of two levels

• General model for languages with morpheme
concatenation
• Independence between lingware and analyzer
• Valid for analysis and generation
• Distinction between lexical and superficial
levels
• Parallel rules for morphophonology
• Simple implementation

NLP Morphology 10
 Morphological Analysis 3

• Morphological rules
• Define the relations betweens characters
(surface) and morphemes and map strings of
characters and the morphemic structure of the
word.
• Spelling rules
• Perform at the level of the letters forming the
word. Can be used to define the valid
phomological alterations.
• Ritchie, Pulman, Black, Russell, 1987

NLP Morphology 11
 Morphological Analysis 4

• input:
• form
• output
• lemma + morphological features

Input Output
cat cat + N + sg
cats cat + N + pl
cities city + N + pl
merging merge + V + pres_part
caught (catch + V + past) or (catch + V + past_part)

NLP Morphology 12
 Morphological Analysis 5

reg_noun irreg_pl_noun irreg_sg_noun plural

fox sheep sheep -s
cat mice mouse
dog

reg_noun plural (-s)

0 1 2

irreg_pl_noun
Morphotactics
irreg_sg_noun
NLP Morphology 13
 Morphological Analysis 6

o
f
x
a
c t s
o
d g
ε
fog y
n
m e
cat e
e
dog o s
donkey u
mouse i
c
mice

Letter Transducers
NLP Morphology 14
 Morphological Analysis 7

upper level lexic cat + N cat

+ N + pl
lower level surface cat
cats

c:c a:a t:t +N:ε +pl:s

NLP Morphology 15
 Morphological Analysis 8
Using FST

• As a recognizer
• From a pair of input strings (one lexical and the other
superficial) determines if one is transduction of the other
• As a generator
• Generates pairs of strings
• As a translator
• From a superficial string generates its lexical translation

NLP Morphology 16
 Morphological Analysis 9
reg_noun irreg_pl_noun irreg_sg_noun plural
fox sheep sheep s
cat m o:i u:ε ce mouse
dog g o:e o:e se goose

reg_noun +pl:s

+N:ε
0 irreg_sg_noun 1 4 2
+sg:ε
+N:ε
2 5 +sg:ε

irreg_pl_noun +N:ε +pl:ε

3 6

NLP Morphology 17
 Morphological Analysis 10

lexical level f o x +N +pl

morphotactics
intermediate level f o x ^ s
spelling rules
superficial level f o x e
s

NLP Morphology 18
 Morphological Analysis 11

o
f
x
a
c t +pl:^s
+N:ε
o
d g
+sg:ε
y
n
m e
e +sg:ε
fog o s
cat u e +pl:ε
dog o:i +N:ε
donkey +u:ε c
mouse e
mice +N:ε
NLP Morphology 19
 Morphological Analysis 12
Spelling rules

name description example

consonant doubling single letter consonant
beg/begging doubled before -ing/-ed
e deletion silent e dropped before
-ing/-ed
make/making
e insertion e added after -s,-z,-x,-ch,-sh
before -s watch/watches
y replacement -y changes to -ie before -s, to
i before -ed try/tries
k insertion verbs ending with voyel +c
add -k panic/panicked

NLP Morphology 20
 Morphological Analysis 13
Spelling rules: e-insertion

ε:e ⇔ [xsz]^:ε ___ s#

⇒ decomposition
/⇐

ε:e ⇒[xsz]^:ε ___ s# ε:ε /⇐ [xsz]^:ε ___ s#

NLP Morphology 21
 Morphological Analysis 14

epenthesis

+:e <=> {< {s:s c:c} h:h> s:s x:x z:z} --- s:s

context

<=>
C: {...}
=> context restriction
V: {a,e,i,o,u,y}
<= surface coercion
C2: {...}
=: whatever
example: box + s
box e s
NLP Morphology 22
 Morphological Analysis 15

e-deletion

e:0 <=> = :C2 --- <+:0 V:= >

or <C:C V:V> --- < +:0 e:e >
or <c:c g:g> --- < +:0 {e:e i:i} >
or l:0 --- +:0
or c:c --- < +:0 a:0 t:t b:b>
mov e + ed
mov ed

agre e + ed
agre ed

NLP Morphology 23
 Morphological Analysis 16

a-deletion

a:0 <=> <c:c e:0 +:0> --- t:t

redu c e + a t ion
redu c t ion
... left context focus right context ...

NLP Morphology 24
 Morphological Analysis 17

lexical level f o x +N +pl

Lexicon-FST

intermediate level f o x ^ s

spelling rules FST1 FST2 ... FSTn

superficial level f o x e s

NLP Morphology 25
 Morphological Analysis 18

Lexicon-FST Lexicon-FST Lexicon-FST

•
FSTA

FST1 .. FSTn FSTA= FST1 ∧ ... ∧ FSTn

.

intersection composition
NLP Morphology 26
 Automatic morphology learning 1

• Problem
• Paradigm stem + affixea
• Obtaining the stems
• Classification of stems into models
• Learning part of the morphology (e.g. derivational)
• Two approaches
• No previous morphologic knowledge is available
• Goldsmith, 2001
• Brent, 1999
• Snover, Brent, 2001, 2002
• Morphologic knowledge can be used
• Oliver at al, 2002
NLP Morphology 27
 Automatic morphology learning 2

• Automatic morphological analysis

• Identification of borders betwen morphemes
• Zellig Harris
• {prefix, suffix} conditional entropy
• bigrams and trigrams with high probability of
forming a morpheme
• Learning of patterns or rules of mapping
between pairs of words
• Global approach (top-down)
• Golsdmith, Brent, de Marcken

NLP Morphology 28
 Automatic morphology learning 3

• Goldsmith’s system based on MDL

(Minimum Description Length)
• Initial Partition: word -> stem + suffix
• split-all-words
• A good candidate to {stem, suffix} splitting in a word
has to be a good candidate in many other words
• MI (mutual information) strategy
• Faster convergence
• Learning Signatures
• {signatures, stem, suffixes}
• MDL

NLP Morphology 29
 Automatic morphology learning 4

• Semi-automatic morphological analysis

• Oliver, 2004
• Starts with a set of manually written morphological
rules
• TL:TF:Desc
• lemma ending
• form ending
• POS
• Lists of non flexive classes , closed classes and
irregular words
• Corpora
• Serbo-Croatian 9 Mw
• Russian 16 Mw
NLP Morphology 30