Towards Minimal Recursion Semantics over Bulgarian Dependency Parsing

Proceedings of Recent Advances in Natural Language Processing, pages 471–478, Hissar, Bulgaria, 12-14 September 2011. Towards Minimal Recursion Semantics over Bulgarian Dependency Parsing Kiril Simov LMD, IICT-BAS kivs@bultreebank.org Petya Osenova LMD, IICT-BAS petya@bultreebank.org Abstract The paper discusses the transferring rules of the output from a dependency parser for Bul- garian into RMRS analyses. This task is re- quired by the machine translation compatibil- ity between Bulgarian and English resources. Since the Bulgarian HPSG grammar is still be- ing developed, a repairing mechanism has been envisaged by parsing the Bulgarian data with the Malt Dependency Parser, and then re- trieving RMRS analyses by exploring the lin- guistic knowledge within BulTreeBank-DP. 1 Introduction Recently a number of machine translation efforts have focused on grammatical formalisms in per- forming source language analysis, transfer rule application and target language generation. It is worth mentioning several works, such as (Bond, 2005) exploiting DELPH-IN infrastructure for developing of HPSG grammars, (Riezler and Maxwell III, 2006) using LFG grammar, (Oepen et al, 2007) working on a hybrid architecture consisting of an LFG grammar, an HPSG gram- mar, partial parsing, and (Bojar and Hajic, 2008) using the Functional Generative Description framework to language analysis on analytical and tectogrammatical level. All the approaches rely on the advances in the development of deep grammar natural language parsing. The ap- proaches share similar architecture and tech- niques to overcome the drawbacks of the deep processing in comparison to statistical shallow methods. Within the above mentioned context, we are constructing a Bulgarian-to-English translation system, based on HPSG. The transfer rules are implemented on the level of MRS (Minimal Re- cursion Semantic) structures (Copestake et al, 2005). The HPSG deep grammar for Bulgarian still has a limited coverage. Thus, for many input sentences it will fail to produce MRS analyses. In such cases, we rely on a dependency parser (Malt parser trained on the BulTreeBank data) to produce a dependency parse for the sentence. Then, we construct an RMRS (Robust Minimal Recursion Semantic) analysis over the depend- ency parse. Thus our input processing architec- ture consists of two grammars – HPSG grammar which produces MRS structures, and Depend- ency grammar which produces RMRS structures. The resulting semantic analysis is the input for the transfer module of the machine translation system. The paper focuses on the dependency tagset and the rules for the construction of RMRS analyses over the dependency parses. It is structured as follows: First, the context of our work is presented. Then our dependency tagset is discussed. In the following section the HPSG- based Bulgarian grammar BURGER is briefly outlined. Finally, the basic rules for the construc- tion of the RMRS analyses from dependency parses are described. 2 Background Our approach is inspired by the work on MRS and RMRS (see (Copestake, 2003; 2007)) and the previous work on transfer of dependency analyses into RMRS structures described in (Spreyer and Frank, 2005) and (Jakob et al, 2010). MRS is introduced as an underspecified se- mantic formalism (Copestake et al, 2005). It …

Content truncated to 500 words