Argument structure of adverbial derivatives in Russian

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1071–1080, Dublin, Ireland, August 23-29 2014. Argument structure of adverbial derivatives in Russian♦ Igor Boguslavsky Institute for Information Transmission Problems, Russian Academy of Sciences / 19, B.Karetnyj, GSP-4, Moscow, Russia Universidad Politécnica de Madrid / Facultad de Informática, Campus de Montegancedo, 28660 Boadilla del Monte, Madrid, Spain Abstract Adverbial derivatives (AdvD) of nouns of the type v jarosti ‘in a rage’, s naslaždeniem ‘with pleasure’, pod predlogom ‘under the pretext of’ etc. often inherit the arguments (actants) of the noun they are derived from. However, as a rule, in case of AdvDs these arguments are realized in a way very different from the nouns. The main linguistic findings of the paper consist in the set of positions the arguments may take with respect to AdvD. In a general case, a actant slot of an AdvD can be either (a) blocked, or (b) filled by a dependent of the AdvD itself (e.g. pod predlogom bolezni ‘under the pretext of illness’, v dokazatel’stvo svoej nevinovnosti ‘as a proof of his innocence’), or (c) filled by the dominating verb (po privyčke prosnulsja rano ‘woke up early out of habit’, slushal pesnju s naslaždeniem ‘listened to the song with relish’), or (d) filled somewhere within the clause organized by the dominating verb; in this case the AdvD argument may be identified based on (d1) its syntactic position (po privyčke ‘by habit’), or (d2) its semantic role with respect to its mother element (v podarok ‘as a present’), or (d3) its communicative function (v bol’šinstve ‘mostly’). A notation is proposed that permits to present the argument structure of AdvDs in a compact way. 1 Introduction This paper is not about computation, it is about linguistics. It does not describe any electronic resource. It is not inspired by weaknesses of NLP applications that need to be fixed. We investigate certain heavily understudied and even largely unnoticed linguistic phenomena that deserve scientific study independently of whether their neglect causes serious errors in today’s NLP applications or not. However, on the other hand, taking these phenomena into account is definitely useful for applications, such as semantic parsing, question answering, recognizing textual entailments, information extraction (e.g. Meyers et al. 1998), machine reading, machine translation, etc. Indeed, semantic parsers should represent the content of the text by means of elementary propositions independently of the syntactic status of the main predicate in these propositions, be it a verb, a noun, or an adverbial. They should be able to understand that such expressions as I believe (that) he is wrong – My opinion is (that) he is wrong) – In my opinion <to my mind>, he is wrong are different NL realizations of the same proposition. Question answering systems should be able to obtain an answer to the question What habits does John have? from the sentence John woke up early out of habit, although the argument frame of the noun habit does not cover this type of …

Content truncated to 500 words