Igor Boguslavsky

Parsing the SynTagRus Treebank of Russian

We present the first results on parsing the SynTagRus treebank of Russian with a data-driven dependency parser, achieving a labeled attachment score of over 82% and an unlabeled attachment score of 89%. A feature analysis shows that high parsing accuracy is crucially dependent on the use of both lexical and morphological features. We conjecture that the latter result can be generalized to richly inflected languages in general, provided that sufficient amounts of training data are available.

The UNL initiative: an overview

We present here a description of the UNL initiative based on the Universal Networking Language (UNL). This language was conceived to support multilingual communication on the Internet across linguistic barriers. This initiative was launched by the Institute of Advanced Studies of the United Nations University in 1996. The initial consortium was formed to support 15 languages. Eight years later, this initial consortium changed, many components and resources were developed, and the UNL language itself evolved to support different types of applications, from multilingual generation to “knowledge repositories” or cross- lingual information retrieval applications. We describe the main features of this UNL Language, making a comparison with some similar approaches, such as interlinguas. We also describe some organizational and managerial aspects of the UNL according to criteria of quality and maturity, placing emphasis on the fact that the initiative is open to any interested group or researcher.

Multilinguality in ETAP-3: reuse of lexical resources

The paper presents the work done at the Institute for Information Transmission Problems (Russian Academy of Sciences, Moscow) on the multifunctional linguistic processor ETAP-3. Its two multilingual options are discussed -- machine translation in a variety of language pairs and translation to and from UNL, a meaning representation language. For each working language, ETAP has one integral dictionary, which is used in all applications both for the analysis and synthesis (generation) of the given language. In difficult cases, interactive dialogue with the user is used for disambiguation. Emphasis is laid on multiple use of lexical resources in the multilingual environment.

Semantic Paraphrasing for Information Retrieval and Extraction

The paper is devoted to the development of a system of synonymous and quasi-synonymous paraphrasing and its practical applications, first of all in the domain of search engine optimization and information extraction. This system is part of the ETAP-3 multifunctional NLP environment created by the Laboratory of Computational Linguistics of the Kharkevich Institute for Information Transmission Problems. Combinatorial dictionaries of Russian, English and some other languages and a rule-driven parser constitute the core of ETAP-3 while a variety of generating modules are used in a number of applications. The paraphrase generator, based on the apparatus of lexical functions, is one such module. We describe the general layout of the paraphrase generator and discuss an experiment that demonstrates its potential as a tool for search optimization.

Enlarging the Diversity of Valency Instantiation Patterns and Its Implications

In the prototypical case, arguments (=actants) are directly subordinated to their predicates and occupy positions of the subject and direct or indirect object. Valency slots filled in this way are called active. In non-prototypical cases, arguments can syntactically subordinate their predicate (passive valency slots) and even have no immediate syntactic link with it (discontinuous valency slots). These types of valency slots are mostly characteristic of adjectives, adverbs and nouns. A number of linguistic concepts are related, directly or indirectly, to the notion of actant. However, usually only prototypical --- active --- valency instantiation is taken into account. If one includes into consideration passive and discontinuous valency slot filling, the area of actant-related phenomena expands greatly. We discuss some of these phenomena and show that the notions of diathesis and conversion require broader generalization.