Discontinuous Verb Phrases in Parsing and Machine Translation of English and German

Discontinuous Verb Phrases in Parsing and Machine Translation of English and German Sharid Lo´aiciga & Kristina Gulordava University of Geneva D´epartement de Linguistique Centre Universitaire d’Informatique sharid.loaiciga@unige.ch, kristina.gulordava@unige.ch Abstract In this paper, we focus on the verb-particle (V-Prt) split construction in English and German and its difficulty for parsing and Machine Translation (MT). For German, we use an existing test suite of V-Prt split constructions, while for English, we build a new and comparable test suite from raw data. These two data sets are then used to perform an analysis of errors in dependency parsing, word-level alignment and MT, which arise from the discontinuous order in V-Prt split constructions. In the automatic alignments of parallel corpora, most of the particles align to NULL. These mis-alignments and the inability of phrase-based MT system to recover discontinuous phrases result in low quality translations of V-Prt split constructions both in English and German. However, our results show that the V-Prt split phrases are correctly parsed in 90% of cases, suggesting that syntactic-based MT should perform better on these constructions. We evaluate a syntactic-based MT system on German and compare its performance to the phrase-based system. Keywords: verb-particle split, discontinuous phrases, test suite, parsing, machine translation 1. Introduction Discontinuous phrases are syntactic constructions which have proved to be hard for natural language processing (NLP) tasks. A multi-word expression (MWE), for in- stance, is hard to identify if its elements are separated by other words in a sentence (Sag et al., 2002). Discontinu- ous phrases can give rise to long and non-projective depen- dencies which pose problems for dependency parsing (Mc- Donald and Nivre, 2011). But machine translation (MT) is probably the NLP task where discontinuity poses the most evident problems. Most of the modern MT systems are based significantly on phrase-based alignment models. While phrase-based MT manages very well the translation of frequent continuous phrases and their reordering, it per- forms poorly on discontinuous phrases. In this paper, we focus on one particular discontinuous verb phrase construction. Specifically, we look at the verb- particle (V-Prt) split in English and German, such as in the examples below: (1) takeV the shoes offP rt (2) machtV schon wieder blauP rt From a syntactic perspective, these two languages behave very differently: in German, the particle position is clause- final and a particle can be separated from its verb by an embedded clause, while in English a particle can be sep- arated from the verb only by its direct object and a very long split is impossible. However, from a MT perspective, these constructions are rather similar as they involve the same type of lexical items (verbs and particles) which must be translated in connection to each other for a correct out- put. Indeed, verb-particle split phrases are hard to translate as illustrated in (3)-(4) (translated using Google Translate), where the same type of error is produced. (3) EN We are going to putV a video outP rt about this. FR Nous allons mettre une vid´eo sur `a …

Content truncated to 500 words