The Conceptual Aspects of Terminographic Definitions ... · The Conceptual Aspects of...

Programme doctoral CUSO 2007 en Sciences du Langage - Leysin, 19-22 mars 2007

The Conceptual Aspects of Terminographic Definitions: Towards

Automatic Genus Tagging

Selja Seppälä

TerminologyMultilingual Information Processing Department

Modules M1-M2, Approches formelles et cognitives du langage

Selja Seppälä Programme doctoral CUSO 2007 en Sciences du Langage - Leysin, 19-22 mars 2007 2

Outline

• Definitions• Background and open questions• Methodological approach• Case study: towards automatic genus

tagging• Conclusions and perspectives


What is terminology?

• An applied activity: writing mono- or multilingual dictionaries for specialised domains (sciences, activities, practices, etc.)

• A scientific discipline: study of terminological phenomena on a linguistic or a conceptual level, or both


What is a terminographic definition?

• A linguistic representation of a concept of a specialised domain

peptidyltransferase = An enzyme located on the large ribosomal subunit that catalyzes peptide bond formation.

• A synthesis of knowledge rich contexts (Meyer, 2001)

• It reflects the structure of the concept• Is subject to some formal restrictions


What is a genus?

• Relates defined & superordinate concepts(peptidyltransferase = An enzyme…)

• Different types:– Conceptual relation to defined concept

• IS_A (Rebeyrolle 2000)• PART_OF (Iris, et al. 1988; L'Homme 2003)• SET_OF (L'Homme 2003)

– Conceptual category (An enzyme = MATERIAL ENTITY)

• Different forms:– simple unit (Protein…)– complex unit (A chemical…, The set of modifications…)

• In French, often the 1st word (65% of test corpus)


Open questions for work

• What makes a context more relevant than another to define a certain concept?

• Since the definition is rather limited in space, what are the characteristics of the concept to be included in the definition?


Hypotheses

• Property selection related to:– type of conceptual category

– type of domain– language

• Can be studied through conceptualstructure of definitions


Conceptual structure of a definition

peptidyltransferase{An enzyme} [located on the large ribosomal subunit] [that catalyzes peptide bond formation.]

[PEPTIDYLTRANSFERASE] IS_A{GENMATERIAL_ENTITY} [SPELOCATION] [SPEFUNCTION]


Methodological approach

• Corpus study of terminographic definitions– complying to generally accepted definition

writing rules– conceptually annotated

• conceptual category(A peptidyltransferase = MATERIAL_ENTITY)

• conceptual relations– between the genus and the defined concept

(peptidyltransferase IS_A enzyme)

– between the specific and the genus(that catalyzes peptide bond formation = FUNCTION)

• Requires conceptual parsing


Case study: Identification of the task

• Find the genus element in a definition (semi-)automatically

• Mark it with an XML tag including the corresponding relation to the defined conceptpeptidyltransferase<GEN relation_VE="IS_A">An enzyme</GEN>located on the large ribosomal subunit that catalyzes peptide bond formation.


Genus extraction: State of the artObjectives: information retrieval to build:• lexical resources for NLP, ontologies, knowledge bases…

(Alshawi 1987, Ide & Véronis 1993, Markowitz et al. 1986)

• terminological resources (Pozzi & Medina 2005, Rebeyrolle 2000)

Text types:• lexicographic definitions (Barnbrook 2002, Ide & Véronis 1993, Markowitz et

al. 1986)

• terminological definitions (L’Homme 2003, Pozzi & Medina 2005)

• free text (Auger 1997, Cartier 2004, Copestake 1990, Rebeyrolle 2000)

Methods:• based on (boundary) markers (Barnbrook 2002, Rebeyrolle 2000)

• statistical methods (Pozzi & Medina 2005)

• use of external resources (POS tagger…) (Vossen et al. 1989)


Case study: How to find a genus?

Search for its boundarieson the basis of:• fixed phrases (part of,

group of...)

• morphosyntacticpatterns (present & past participles, relative pronouns...)

Finding regularities to form extraction patterns• in litterature (Iris et al. 1988;

L'Homme 2003, Rebeyrolle 2000…)

• with a concordancer(GEN + end of following word; "Type of " + 1 to 3 words…)

By taking advantage of:1. the definition sublanguage (Barnbrook 2002)

– fixed phrases

– present & past participles

– characteristic morphosyntactic & lexical items


Case study: How to find a genus?

By taking advantage of:2. the hierarchical structure of the domain

molecule = A structural unit of matter consisting of one or more atoms.

nucleic acid = A molecule formed by nucleotides located on one strand or two.

Search for terms of the domain (genus proximus)

Creating a list of terms


Case study: Implementation

• Training corpus: 500 definitions fromdifferent domains

• Test corpus: 92 definitions from the terminology of protein biosynthesis in eukaryotic cells (Bourjault, 2005)

• Perl program• Regular expressions


Case study: The processing method

Four ordered steps:

1. Insert opening tag ⇒ Search for probable specific elements preceding the genus

<SPE>The first</SPE><GEN>phase of translation </GEN>that…

2. Find GEN including termsfrom the domain

<GEN>An enzyme</GEN> located on…

…exoneukaryotic cellenzymeelongation…


Case study: The processing method

3. Search for closing boundary markers:• specific rules: fixed phrases

<GEN>The set of rules</GEN>used for…

• general rules: morphosyntactic & lexical markers

<GEN>An enzyme</GEN> located for…<GEN>A protein</GEN> that acts…

4. Tag 1st word of unmarked definitions


Case study: Performance evaluation

• Baseline : tag the 1st word (65%)

• Performance of the method: 78/92 (85%)• Raised mainly by:

– term search– fixed phrases

• Errors due to:– absence of fixed phrases indicating: 10/14

• PART relation: 9/10 (Branch of… [a science])

• WHOLE relation: 1/10 (An assembly of…)

– inexact extraction patterns: 4/14 (too greedy)


Case study: Main challenges

• Adequately refining the extraction patterns (avoid greedyness, find multiple genus…)

• Finding the best ordering of the rules• Discriminating lexicalized compound

words from free word sequences(A polypeptide chain… vs. A linear chain…)

• Collecting all forms of fixed phrases for each relation (can be domain specific)


Conclusion

• Manually created and classified rulesperform well

• Advantage: does not require external NLP resources

• Method can be used in other languages• Basis for automating:

– rule definition

– rule classification


Perspectives

• Adapt GEN processing method to SPEs• Apply method to other languages• Study conceptual regularities in definitions

to find patterns according to:– concept type– domain

– language


Thank you for your attention!


References• Alshawi, H. (1987), "Processing dictionary definitions with phrasal pattern

hierarchies", in Computational Linguistics, vol. 13, n°3-4, pp. 195-202.• Auger, A. (1997), Repérage des énoncés d'intérêt définitoire dans les bases

de données textuelles, Université de Neuchâtel, Faculté des Lettres.• Barnbrook, G. (2002), Defining Language : A local grammar of definition

sentences, John Benjamins, Amsterdam, Philadelphia.• Bourjault, A. (2005), Terminologie de la biosynthèse des protéines chez les

cellules eucaryotes : anglais-français, Mémoire de DESS en terminologie, Université de Genève.

• Cartier, E. (2004), Repérage automatique des expressions définitoires : modélisation de l'information définitoire, méthode de repérage automatique, méthodologie de développement des ressources linguistiques, description des expressions du français contemporain, mise en oeuvre informatique, Thèse de Doctorat, Paris-IV Sorbonne (décembre 2004).

• Copestake, A. (1990), An approach to building the hierarchical element of a lexical knowledge base from a machine readable dictionary, Paper presentedat the First International Workshop of Inheritance in NLP, Tilburg.

• Ide, N. et Véronis, J. (1993), "Extracting knowledge bases from machine-readable dictionaries : Have we wasted our time?" in KB&KS'93 Workshop, Tokyo


References 2• Iris, M. A. et al. (1988), "Problems of the part-whole relation", in Evens M. W., Relational Models of the Lexicon: Representing Knowledge in Semantic Networks, Cambridge University Press, Cambridge.

• L'Homme, M.-C. (2003), "Indices de relations conceptuelles dans les définitions terminologiques. Application au domaine de l’informatique", in, Bach C. et Martí J.I Jornada Internacional sobre la Investigación en Terminología y Conocimiento Especializado, Barcelona, IULA.

• Markowitz, J., et al. (1986), "Semantically significant patterns in dictionary definitions", in Proceedings of the 24th conference on Association for Computational Linguistics, pp. 112-119.

• Pozzi, M. and Medina, A. (2005), "Towards the Establishment of Criteria for Automatic Genus Extraction in Specialised Dictionaries", in TKE Workshop: "Working with Terminology and KMS", TKE2005 - 7th International conference on Terminology and Knowledge Engineering, August 19, 2005, Copenhagen.

• Rebeyrolle, J. (2000b), "Utilisation de contextes définitoires pour l’acquisition de connaissances à partir de textes", in, Tchounikine P.IC'2000 - ACTES DE LA CONFERENCE Journées francophones d'Ingénierie des Connaissances, Toulouse, IRIT, 10-12 mai 2000.

• Vossen, P., et al. (1989), "Meaning and structure in dictionary definitions", in Boguraev B. et Briscoe T., Computational Lexicography for Natural Language Processing, Longman, London, New York

The Conceptual Aspects of Terminographic Definitions ... · The Conceptual Aspects of...

Documents

Transcript of The Conceptual Aspects of Terminographic Definitions ... · The Conceptual Aspects of...