Issues of Valency in Prague Dependency Treebank: Creating Valency Lexicon of Verbs

1
Issues of Valency in Prague Dependency Treebank
Creating Valency Lexicon of Verbs

• Markéta Lopatková
• Center for Computational Linguistics
• MFF UK, Prague

CIL XVII, Prague, July 26, 2003
1
2
Motivation

• traditional linguistics
• source of data for linguistic research
• verification of theoretical criteria set up
• natural language processing
• lemmatization
• morphological tagging
• syntactic analysis
• word sense disambiguation
• semantic analysis
• machine translation
• building other resources
• language acquisition

CIL XVII, Prague, July 26, 2003
2
3
Syntactic vs. semantic approach I.

• Levin Verb Classes (Levin, 1993)
• hypothesis syntactic features of verbs are
  semantically determined
• method syntactic behavior ? semantic classes
• alternation a change in the realization of
  the argument structure of a verb
• conative alternation
• Edith cuts the bread ? Edith cuts at the bread
• classes verbs which undergo certain types of
  alternations

CIL XVII, Prague, July 26, 2003
3
4
Syntactic vs. semantic approach II.

• PropBank (Palmer et al., 2001)
• layer of semantic annotation in PennTreebank
• argument structure for verbs
• arguments Arg0, ... Arg5
• modificators ArgM (LOC, TMP, EXT, PRP, ADV)
• He was drawing diagrams and sketches for his
  patron.
• Arg0 he
• Rel drawing
• Arg1 diagrams and
  sketches
• Arg2-for his patron
• He keeps st in the fridge.
• Arg0 he
• Rel keeps
• Arg1 st
• Arg2-in the fridge
• (also Hajicová, Kucerová, 2002)

CIL XVII, Prague, July 26, 2003
4
5
Syntactic vs. semantic approach III.

• FrameNet (Fillmore, 2002)
• it groups lexical items with parallel semantic
  characterization
• the structure and particular components
  correspond to semantic roles of the common
  semantic frame
• verbs, nouns, adjectives, prepositions
• Communication
• Speaker Message Addressee Topic
  Medium
• Tom communicates with Kim
  about the festival.
• Tom communicates with Kim
  by letter.
• Tom communicates the message to me.
• Reciprocality
• Protagonists Prot-1 Prot-2
• Tom fought with Kim.
• Tom and Kim fought.

CIL XVII, Prague, July 26, 2003
5
6
Syntactic vs. semantic approach IV.

• LCS Database (Lexical Conceptual Structure)
  (Dorr, 2001)
• semantic representation
• semantic structure semantic content
• verb cut down lexical item (act_on loc (
  thing 1) ( thing 2) (( on 23) loc
  (head) (thing 24))
• (cutingly
  26)
• (down/m))
• sentence United States cut down (the) quota.
• (act_on loc (us) (quota)
• (( on 23) loc (head) (thing 24))
• (cutingly 26)
• (down/m))
• logic arguments (ag, exp, th, src, goal, info,
  perc, loc,poss, time, prop)
• logic modifiers (mod-poss, ben, instr, purp,
  mod-loc, manner, mod-prop)
• cut down _ag_th,mod-loc(on)

CIL XVII, Prague, July 26, 2003
6
7
Prague Dependency TreeBank

• based on
• Functional Generative Description (FGD) (Sgall et
  al., 1986)
• dependency-oriented
• stratificational
• level of underlying representation
• (tectogrammatical level) (described in
  Hajicová et al., 2000)
• valency theory (esp. Panevová, 1994)

CIL XVII, Prague, July 26, 2003
7
8
Valency in FGD I.

• complementations
• inner participants vs. free modifications
• obligatory vs. optional
• valency frame
• Matka.ACT predelala detem.ADDR loutku.PAT z
  Kašpárka.ORIG na certa.EFF.
• Mother re-made a puppet for children from a
  Punch to an imp. (Panevová)
• V Praze.LOC se sejdeme na Hlavním nádraží.LOC u
  pokladen.LOC. (Panevová)
• In Prague we will meet at Main Station near a
  booking-office.

obligatory optional
inner participants
free modifications
CIL XVII, Prague, July 26, 2003
8
9
Valency in FGD II.

• a middle position
• syntactic criteria are used for the
  identification of Actor and Patient (Actor is the
  first inner participant, the second is always a
  Patient)
• other inner participants (Addressee, Origin and
  Effect) as well as free modifications are
  determined in accordance with semantic
  considerations
• concept of shifting (Panevová, 1974-75)
• Origin
• Actor Patient
  Addressee
• Effect
• Kniha.ACT vyšla. (Panevová) The book appears.
• Chlapec.ACT vyrostl v muže.PAT. (Panevová) A
  boy grew up to a man.

CIL XVII, Prague, July 26, 2003
9
10
Valency in FGD III.

• valency of autosemantic words
• verbs (Panevová, from the seventies)
• 5 inner participants - Actor, Patient, Addressee,
  Origin, Effect
• app. 45 free modifications
• shifting of cognitive roles for inner
  participants
• nouns (esp. Panevová, 2000, Reznícková,
  manuscript)
• verbal complementations
• spec. nominal complementations - Identity,
  Partitive, Appurtenance, Restrictive and
  Descriptive Attribute
• adjectives (Panevová, 1998)
• verbal complementations
• spec. adjectival complementations

CIL XVII, Prague, July 26, 2003 10
11
Valency structure on TR level of PDT

• the core of annotation on the tectogrammatical
  level
• problem of consistency ? valency lexicon
• verbs
• two branches
• lists of verbs with their complementations being
  created and used by annotators (PDT-VALLEX)
• complex valency lexicon (VALLEX)
• nouns
• the theoretical aspects and methodology are
  refined now (Reznícková, manuscript)
• lists of nouns with their complementations
• adjectives
• lists of adjectives with their complementations

CIL XVII, Prague, July 26, 2003 11
12
Valency lexicon of verbs PDT-VALLEX

• lists being created and used by annotators
• valency frames of verbs in their particular
  meanings, as they appear during annotation, the
  lexeme as a whole is not analyzed
• the information specifying elements of frames
• functor - i.e. name of complementation
• type - obligatory / optional
• possible morphemic form(s)
• example(s)
• it serves for consistency of annotation
• approx. 4 700 verbs with 7 150 valency frames
  (i.e. 1,5 frames per verb)
• dát to give ... ACT(1obl) ADDR(3obl)
  PAT(4obl)
• dát nekomu knihu to give sb a book

CIL XVII, Prague, July 26, 2003 12
13
Valency lexicon of verbs VALLEX

• complex information on the whole verb lexeme in
  all its meanings (Lopatková, Žabokrtský, 2002)
• the information on particular valency frames,
  corresponding to its meanings (described with
  gloss(es) and example(s))
• the information specifying elements of frames
• functor - i.e. name of complementation
• type - obligatory / optional
• possible morphemic form(s)
• mluvit to speak
• ... ACT(1obl) ADDR(s7obl)
  PAT(o6opt)
• mluvila s ním o detech she spoke with him about
  their children
• additional syntactic information

CIL XVII, Prague, July 26, 2003 13
14
Valency lexicon of verbs VALLEX II.

• additional syntactic information for particular
  valency frames
• reflexivity (in progress)
• reciprocity
• control
• aspect and aspectual counterparts
• possible diatheses, passivization (future plans)
• primary / secondary / idiomatic usage
• syntactic/semantic class (in progress)
• pointers to Czech EuroWordNet (in progress)
• frequency of a particular frame in samples of CNK
  (60 occurrences of each verb lexeme)

CIL XVII, Prague, July 26, 2003 14
15
Valency lexicon of verbs VALLEX III.

• current state
• 1 400 verbs with 3 860 frames (i.e. 2,7 frames
  per verb)
• verbs chosen according to their frequency in
  Czech National Corpus and PDT
• about 85 on running text in PDT
• open questions
• enriched valency frame
• syntactic-semantic classes
• alternative frames
• frozen collocations

CIL XVII, Prague, July 26, 2003 15
16
Valency lexicon of verbs

• Why two branches?
• PDT-VALLEX extensive
• necessary for annotation
• recall improves relatively quickly
• VALLEX intensive
• the whole lexeme is analyzed en bloc ? adequate
  and consistent description
• precision improves
• the two branches are supposed to be merged
• PDT-VALLEX valuable source for VALLEX

CIL XVII, Prague, July 26, 2003 16
17
(No Transcript)
18
Enriched valency frames I.

• inner participants
• each inner participant can occur only once
• (with single occurrence of a verb)
• combination of inner participants must be listed
  for a particular verb
• morphemic form is predicted by the governing verb
• concept of shifting is applied
• free modifications
• each free modification can be repeated
• syntactically, they can modify any verb
• (only semantic restrictions are often present)
• they have typical semantics
• they do not undergo the shifting

CIL XVII, Prague, July 26, 2003 18
19
Enriched valency frames II.

• quasi-valency complementations (also Panevová,
  2003)
• each quasi-valency complementation can occur only
  once
• (with any occurrence of a verb)
• each quasi-valency complementation is
  characteristic for a limited list of verbs
• morphemic form is predicted by the governing verb
• they have typical semantics
• they do not undergo the shifting
• Obstacle uhodit hlavou o vetev.OBST to bump
  one's head against a bough
• zavadit o stul.OBST to brush against a
  table
• Difference prodloužit o hodinu.DIFF to prolong
  by one hour
• Mediator vzít nekoho za ruku.MDT to take sb
  by his/her hand

CIL XVII, Prague, July 26, 2003 19
20
Enriched valency frames III.

• typical modifications
• optional free modifications commonly used with
  a verb
• usually modify group of verbs with similar
  meaning
• morphemic form
• prototypical for some modifications
• e.g. Dative case or prep. group pro forAcc for
  Benefactor
• determined by the typical semantics of the
  modifying members
• e.g. prep. groups na onLoc and v inLoc
  typically specify Location
• verbs of motion typically modified by
  Direction modification (provided that
  Direction is not obligatory)
• jít do kina / pres les / jít z domova
• to go to cinema / through the wood / from home
• verbs of exchange typically modified by
  modification of Recompense
• dát / dostat / získat / kupovat / brát neco.PAT
  za neco.RCMP
• to give / get / obtain / buy / take something
  for something

CIL XVII, Prague, July 26, 2003 20
21
Exploitation of the valency lexicon

• reaching the consistency of assigning the valency
  structure (PDT-VALLEX)
• automatic syntactic analysis (shallow parsing)
• tectogrammatical parser
• automatic system for creating an underlying
  representation of Czech sentences
• source data for building the valency lexicon of
  nouns

CIL XVII, Prague, July 26, 2003 21
22
Resources

• theoretical articles on valency (Panevová)
• The Manual for Tectogrammatical Tagging of the
  Prague Dependency Treebank (Hajicová et al.,
  2000)
• lists of particular valency frames created by
  annotators
• electronic valency dictionary of surface
  realizations of verbal modifiers
• (FI MU Brno, Pala, Ševecek, 1997)
• printed dictionaries
• Slovesa pro praxi (SPP, 1997), valency
  specification of 767 most frequent
• verbs
• Slovník spisovného jazyka ceského (SSJC, 1964)
• Slovník spisovné ceštiny pro školu a verejnost
  (SSC, 1978)
• Slovník ceských synonym (SCS, 1994)
• Slovník ceské frazeologie a idiomatiky (SCFI,
  1983)
• Czech National Corpus (CNK)
• EuroWordNet, Czech WordNet

CIL XVII, Prague, July 26, 2003 22
23
References I.

• Dorr, B.J. (2001) LCS Verb Database, Online
  Software Database of Conceptual Structures and
  Documentations, UCMP .
• Fillmore, Ch. (2002) FrameNet and the Linking
  between Semantic and Syntactic Relations.
• In COLING 2002, Proceedings, pp. xxviii-xxxvi.
• Hajicová, E. et al. (2000) A Manual for
  Tectogrammatical Tagging of the Prague Dependency
  Treebank. UFAL/CKL Technical Report TR-2000-09.
• Hajicová, E., Kucerová, I. (2002)
  Argument/Valency Structure in PropBank, LCS
  Database and Prague Dependency Treebank. In LREC
  2002, Proceedings, pp. 846-851.
• Levin, B. (1993) English Verb Classes and
  Alternations A Preliminary Investigation.
  Chicago University of Chicago.
• Lopatková, M. et al. (2002) Tektogramaticky
  anotovaný valencní slovník ceských sloves.
  UFAL/CKL Technical Report TR-2002-15.
• Lopatková, M., Žabokrtský, Z. (2002) Valency
  Dictionary of Czech Verbs. In LREC 2002,
  Proceedings, pp. 949-956.
• Lopatková, M. (2003) Valency in the Prague
  Dependency Treebank Building the Valency
  Lexicon. PBML 79. (in press)

CIL XVII, Prague, July 26, 2003 23
24
References II.

• Pala, K., Ševecek, P. (1997) Valence ceských
  sloves. In Sborník prací FFUB, Brno.
• Palmer, M. et al. (2001) Automatic Predicate
  Argument Analysis of the Penn TreeBank. In HLT
  2001, Proceedings, San Francisco Morgan Kaufamm.
• Panevová, J. (1974-75) On Verbal Frames in
  Functional Generative Description. Part I, PBML
  22, pp. 3-40, Part II, PBML 23, pp. 17-52.
• Panevová, J. (1994) Valency Frames and the
  Meaning of the Sentence. In Luelsdorff (ed.) The
  Prague School of Structural and Functional
  Linguistics, John Benjamins, pp. 223-243.
• Panevová, J. (1998) Ješte k teorii valence. Slovo
  a slovesnost 59, pp. 1-14.
• Panevová, J. (2000) Poznámky k valenci
  podstatných jmen. Ceština - univerzália a
  specifika 2, Masarykova Univerzita, Brno, pp.
  173-180.
• Panevová, J. (2003) Some Issues of Syntax and
  Semantics of Verbal Modifications. In
  Proceedings of MTT 2003, Paris. (in press)
• Sgall, P. et al. (1986) The Meaning of the
  Sentence in Its Semantic and Pragmatic Aspects.
  Dordrecht Reidel, Prague Academia.

CIL XVII, Prague, July 26, 2003 24