LTG vs. ITG Coverage of Cross-Lingual Verb Frame Alternations

LTG vs. ITG Coverage of Cross-Lingual Verb Frame Alternations Karteek Addanki, Chi-kiu Lo, Markus Saers and Dekai Wu Human Language Technology Center Dept. of Computer Science and Engineering Hong Kong University of Science and Technology {vskaddanki|jackielo|masaers|dekai}@cs.ust.hk Abstract We show in an empirical study that not only did all cross-lingual alternations of verb frames across Chinese–English trans- lations fall within the reordering capacity of Inversion Transduction Grammars, but more surprisingly, about 97% of the alter- nations were expressible by the far more restrictive Linear Transduction Grammars. Also, about 71% of the cross-lingual verb frame alternations turn out to be mono- tonic even for diverse language pairs such as Chinese–English. We also observe that a source verb frame alternation pattern translates into a small subset of the possi- ble target verb frame alternation patterns, based on the construction of the source sen- tence and the frame set definitions. As a part of our evaluation, we also present a novel linear time algorithm to determine whether a particular syntactic alignment falls within the expressiveness of Linear Transduction Grammars. To our knowl- edge, this is the first study that attempts to analyze the cross-lingual alternation be- havior of semantic frames and the extent of their coverage under syntax-based machine translation formalisms. 1 Introduction In this paper we present a first empirical study on the cross-lingual verb frame alternations by align- ing semantic role fillers in parallel sentences. We evaluate how many of these alignments fall within the expressiveness of two well known syntax based machine translation formalisms: Inversion Trans- duction Grammars (Wu, 1997) and Linear Trans- duction Grammars (Saers, 2011). As a part of our evaluation, we discuss the reordering of semantic roles within a frame and across frames within a sen- tence. We also present a novel algorithm to deter- mine whether there exists a canonical parse for an alignment under Linear Transduction Grammars. © 2012 European Association for Machine Translation. While recent years have seen continued improve- ments in the accuracy of SMT using tree-structured and syntactic models (Wu, 1997; Wu and Chi- ang, 2009; Wu, 2010; Wu and Fung, 2009b,a), only a few attempts (Wu and Fung, 2009b) have been made towards using semantic roles to guide SMT. Recent studies (Wu and Fung, 2009a) show that most of the glaring errors made by statistical machine translation systems are a result of con- fused semantic roles which result in serious mis- understanding of the essential meaning. Seman- tic roles have also been successfully used in eval- uating translation utility (Giménez and Màrquez, 2007, 2008; Callison-Burch et al., 2007, 2008; Lo and Wu, 2011a,b). However, no effort has been made to identify the reordering of semantic role fillers across languages. Such an analysis is inter- esting for two reasons: (1) to determine how much reordering we really need in order to preserve mean- ing while translating, and (2) to determine which existing syntactic SMT models have an inherent bias towards such a reordering. The first reason helps us determine an upper bound on …

Content truncated to 500 words