Morphological Analysis of Hungarian in NooJ

1
Morphological Analysis of Hungarian in NooJ

• Peter Vajda
• Hungarian Academy of Sciences
• Research Institute for Linguistics

2
Summary

• Hungarian morphology
• Linguistic resources
• Some experiments with INTEX/NooJ
• The solution
• Examples
• Derivation

3
Hungarian morphology

• Agglutinative (and sometimes inflectional)
• The suffixes
• Can have many forms (vowel harmony)
• Can change the form of the stem (there are groups
  of variants)
• bokor (sg.) ? bokr ok (pl.) alma (sg.) ?
  almá k (pl.)
• Sometimes begin with a linking vowel
• plural -k / -ak / -ek / -ok / -ök
• A noun (adj., num.) can have 7-800 forms
• A verb can have 80 forms
• Orthography there are difficulties, when
  digraphs are doubled
• cs ? cscs ? ccs, gy ? gygy ? ggy

4
Nominal inflections

• 18 cases (nominative, accusative, dative
  grammatical relations which are expressed by
  prepositions in French/English)
• Expression of the possessives by suffixes
• Which mark the number, the person, the number of
  the possessed
• ház-a-m, ház-a-d, ház-a (my/your/his house)
• ház-a-i-m, ház-a-i-d, ház-a-i (my/your/his
  houses)
• Anaphorical possessive
• A ház Péteré ? The house is Péters A házak
  Péteréi ? The houses are Péters
• The maximal number of inflections can be five
• barát-ai-tok-é-i-t
• (I can see) those (things) of your friends

5
Verbal inflections

• Two tenses present, past
• three modes indicative, conditional, imperative
• definite and indefinite conjugations
• Néz-ek egy asztalt ? Néz-em az asztalt
• I watch a table ? I watch the table
• one special form where the subject is in 1st
  person and the object is in the 2nd
• néz-lek (I watch you)
• infinitive and conjugated infinitive (sometimes
  subjunctive in French)

6
The resources

• Dictionary of Hungarian inflections (Elekfi,92)
• A traditional description, profound and
  exhaustive
• Two dimensional classification
• Vowel harmony (3 classes) and
• complex features of the stems (stem-types,
  linking vowel, etc., 55 classes)
• Altogether 1700 different sub-classes
  (paradigms)
• systematic differences and similarities are
  hidden
• not convenient to use in finite-state transducers
• We have converted it into a database, where we
  can retrieve all the forms from

7
The experiments with INTEX/NooJ

• Brute-force method
• We created one graph per sub-class for testing
  INTEX
• 1700 sub-graphs
• 45000 paths in the graphs
• Using only dictionaries (.nod)
• Dictionary of stems (70000 words)
• ház,ház,NC2Astem1NW
• Dictionary of suffixes (one million entries)
• ()ak,lt1NC2Astem1gt0,1L,N1SanaPL
• ()am,lt1NC2Astem1gt0,1L,N1SanaPSe1
• ()at,lt1NC2Astem1gt0,1L,N1SanaACC
• ()at,lt1NC2A1stem1gt0,1L,N1SanaACC
• ()amat,lt1NC2Astem1gt0,1L,N1SanaPSe1ACC
  
• dictionary of lexical forms (which have a zero
  morpheme as suffix)
• ház,ház,NanaNOM

8
The linguistic solution

• transform the database into a grammar based on
  morpho-phonological features
• The grammatical features of stems and morphemes
  are in the dictionary
• The features of the stems and the suffixes can be
  unified

• Grammar
• We have to describe the order of the morphemes
• Introduce features which select from the
  allomorphs

9
The order of morphemes for nominals
10
The order of morphemes for nominals
barát-a-i-tok-é-i-t
barát,N PS PL ps_2 ps_pl
ANAPi ACC
11
(No Transcript)
12
Morpho-phonological features

• To introduce features we examine the allomorphs
• HÁZ HAJÓ
• HÁZ - A HAJÓ-JA
• ház,,Nnonj hajó,,Nj
• HÁZ - AT HAJÓ - T
• ház,,Nnonjacclink hajó,,Njaccnolink

13
The dictionary
14
(No Transcript)
15
The plural and the accusative kalap - ot (hat,
SGACC) kalap - ok - at (hats, PLACC)
16
Derivation

• Can change or leave the category (POS)
• Introduce new features
• kosár kosar - ak (pl.) basket
• kosar-as kosar - as - ok (pl.) basketball
  player
• Simple cases are handled by graphs
• Others are listed as lemmas in the dictionary

17
Assimilation and digraphs

• some suffixes (eg. val/vel) enforce total
  assimilation
• LÉC VEL ? LÉCCEL
• PÉCS VEL ? PÉCCSEL
• PLÉD VEL ? PLÉDDEL

18

• Conclusion
• We have adapted the traditional description
• We have described the inflectional morphology of
  Hungarian in NooJ grammars/dictionaries
• Handled some of the derivational morphology
• Objectives
• Find a simpler method for derivation
• Disambiguation
• Automatic methods to expand the dictionary
• Automatic delegation of features

19
Thank you