Summary (Автоматизация лексико-типологических исследований методы и инструменты)

Manuscript copyrightDaria RyzhovaAUTOMATING RESEARCH IN LEXICAL TYPOLOGY: METHODSAND TOOLSSummaryof the thesis for the degreeof PhD in Philology and LinguisticsMoscow 2018GENERAL OVERVIEW OF THE THESISLexical typology is a field of linguistics that applies comparative analysis to wordmeanings across languages.

Lexical typology received a powerful impetus with publication ofthe seminal work by Berlin, Kay 1969 on typology of colour terms; the authors introduced anelaborate methodology for comparing lexical meanings, which is still widely used today. Theirpaper had deep resonance with the linguistic community and spurred rapid developments inlexical typology. Presently, the interest to typological analysis of lexical meanings enjoyssteady growth.In the five decades of its development, lexical typology has accomplished remarkableachievements: a wide array of methods for data collection and analysis have been designed (e.g.see the recent overview in Koptjevskaja-Tamm et al.

2016), and a large number of lexical fieldshave been described (Newman 1998, 2009, Majid, Bowerman 2007, Майсак, Рахилина 2007,Kopecka, Narasimhan 2012б and others). However, there remains a number of methodologicalcomplications that have not been yet resolved. They are primarily associated with the fact thatlexical analysis requires an ample and representative body of lexical data which, in mostinstances, cannot be retrieved from lexicographic sources alone. Therefore, researchers arefaced with the necessity to compile task-specific questionnaires and to invite experts in multiplelanguages to interview the informants and to analyze the obtained materials.

The high cost ofsuch projects inhibits the possibility of detailed analyses of extensive semantic fields across alarger number of languages. As a consequence, researchers are compelled to either dramaticallycurtail the number of languages in the sample, or to reduce the granularity of the analysis.Meanwhile, the results of thorough and large-scale comparative analysis of lexical meaningscould have high practical utility, along with their indisputable theoretical value: they could beutilised in manual and machine translation, and in improving the efficiency of second languageteaching.The present work is based on the premises of the frame-based approach to lexicaltypology developed by the Moscow Lexical Typological Group, MLexT (Rakhilina, Reznikova2016), which, in its turn, stems from the traditions of the Moscow semantic school, seeАпресян 1974. The key notion of this research paradigm is the concept of frame; it stands toindicate a minimal situation which is denoted by a specialised lexeme in a given language.Typological description of a semantic field consists in identifying the frames that constitute it(i.e.

the types of situations that are covered by the lexemes of the field), and in describing themodels of their lexicalization (i.e. how the situations converge within lexemes).The set of frames is established by analyzing the combinability of words as attested indictionaries and corpora, and is further elaborated and finalised via interviews with speakers ofthe language. The convergence of frames under individual lexemes is visualised by means ofsemantic maps which are similar to the maps used in grammatical typology (see Haspelmath2003).The present thesis proposes new methods for a u t o m a t e d collection and analysis oflexical typological data; it is motivated by the methodology and theory of the frame-basedapproach. The suggested methods aim to:2(1) promote faster and easier collection of initial data;(2) uncover new patterns of convergence and divergence of meanings within lexemes,which could not be captured by the previously used methods;(3) minimise the dependence of lexical typological analysis on the researcher’s initialtheoretical premises and his / her native language, as well as on the languages fromwhich the data for the questionnaire is drawn.The primary goal of the thesis is accomplished through the following series of tasks:(1) formalise the procedure of frame-based lexical typological research, i.e.

subdivide itinto stages and formulate the problems to be addressed in each stage;(2) suggest automated methods for implementation of each stage and test them;(3) analyze the performance of the suggested methods and outline the directions forfurther development of the most efficient algorithms.The algorithms for automated collection and analysis of lexical data proposed in thepresent thesis are based on the following methods: distributional semantic models (Baroni et al.2013), cluster analysis (Everitt 2011), and formal concept analysis (Ganter, Wille 1999).The relevance of the thesis is determined by the proven demand for automated lexicalanalysis that exists in both theoretical and applied linguistics.The scientific novelty of the present research is warranted by the fact that lexicaltypology remains a largely understudied area, and by the relatively small number of projectsdeveloping computational methods of lexical analysis.

The methods used in this thesis have notbeen applied to lexical typological problems before; we suggest exclusive algorithms for theirimplementation in lexical typological analysis.The following propositions are submitted for the defense:(1) The frame structure of a field is represented by overlapping clusters with distinctcentres (or “foci’);(2) A draft version of a lexical typological questionnaire can be compiled from amonolingual corpus of texts by means of distributional semantic modeling and by thesubsequent cluster analysis of the obtained distributional space;(3) The questionnaire can be translated automatically with the aid of parallel corpora,machine-readable dictionaries, and online machine translation engines; it can be filled(at least partially) with data from parallel and monolingual corpora;(4) Formal concept lattices can be used as semantic maps to visualise the relativedistances between the frames of direct meanings of different lexemes as well as thesystemic relations between the direct and the figurative meanings within a semanticdomain.The theoretical significance of the present work is defined by its contribution to thedevelopment of lexical typology in general and to the frame-based approach in particular.

Theresults of the research provide new insights into organization of the semantic space of lexicalmeanings and propose new hypotheses about the degree of their comparability.The practical significance of the thesis consists in the development of algorithms tooptimise the process of lexical typological research and thus facilitate preparation of data for the3purposes of conventional and computational lexicography, second language teaching, andmanual and machine translation.All the experiments presented in this thesis are based on the material of the earlier studiesof several semantic fields of qualitative features and verbs that were manually conducted bymembers of the MLexT group, such as the fields of ‘sharp’ (Кюсева 2012), ‘smooth’ (Кашкин2013), ‘straight’ (Лучина 2014), ‘thick’ and ‘thin’ (Козлов et al.

2016), ‘oscillation’ (Шапиро2015), ‘falling’ (Reznikova, Vyrenkova 2015), and several others.Public demonstrations of the results. The major results of the research were presentedat the following academic events: the 9th Conference on Typology and Grammar for YoungScholars (Saint-Petersburg, Russia, 2014), the Workshop on Lexical Typology at the Universityof Helsinki (Helsinki, Finland, 2014), the 16th April International Academic Conference onEconomic and Social Development at the National Research University Higher School ofEconomics / NRU HSE (Moscow, Russia, 2015), the academic seminar with the visiting scholarAsifa Majid at the NRU HSE (Moscow, 2015), the International Conference on ComputationalLinguistics “Dialogue 2015” (Moscow, 2015), the conference “Problems of ComputationalLinguistics” (Voronezh, Russia, 2015), the 1st International Conference on Foreign Languagesin Research and Education: Issues and Perspectives (Moscow, 2015), the international academicseminar “Computational Linguistics and Language Science” (Moscow, 2016), the postersession of the Typological School held by the School of Linguistics at the NRU HSE (Moscow,2016), the 10th International Conference on Language Resources and Evaluation / LREC’16(Portorož, Slovenia, 2016), the International Seminar on Perceptual Metaphor (Nijmegen, theNetherlands, 2016), the International Seminar on Motion Verbs (Paris, France, 2017), and theInternational Cognitive Linguistics Conference (Tartu, Estonia, 2017).The contents of the thesis.

The thesis consists of the Introduction, five Chapters, theConclusion, the List of References containing 158 titles, and 5 Appendices. The Introductionprovides a general overview of the thesis. The first section of Chapter 1 reviews the existingmethods of comparative lexical analysis, including the state-of-the-art computational methods.The second section of Chapter 1 discusses the frame-based approach to lexical typology thatforms the basis of this research. Chapter 2 presents the results of a series of experimentsassessing the validity of the concept of frame that stands at the core of the frame-basedparadigm.