Graph-Based Translation Via Graph Segmentation

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 97–107, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Graph-Based Translation Via Graph Segmentation Liangyou Li and Andy Way and Qun Liu ADAPT Centre, School of Computing Dublin City University, Ireland {liangyouli,away,qliu}@computing.dcu.ie Abstract One major drawback of phrase-based translation is that it segments an input sen- tence into continuous phrases. To sup- port linguistically informed source discon- tinuity, in this paper we construct graphs which combine bigram and dependency relations and propose a graph-based trans- lation model. The model segments an input graph into connected subgraphs, each of which may cover a discontinuous phrase. We use beam search to combine translations of each subgraph left-to-right to produce a complete translation. Experi- ments on Chinese–English and German– English tasks show that our system is significantly better than the phrase-based model by up to +1.5/+0.5 BLEU scores. By explicitly modeling the graph segmen- tation, our system obtains further improve- ment, especially on German–English. 1 Introduction Statistical machine translation (SMT) starts from sequence-based models. The well-known phrase- based (PB) translation model (Koehn et al., 2003) has significantly advanced the progress of SMT by extending translation units from single words to phrases. By using phrases, PB models can cap- ture local phenomena, such as word order, word deletion, and word insertion. However, one of the significant weaknesses in conventional PB models is that only continuous phrases are used, so gen- eralizations such as French ne . . . pas to English not cannot be learned. To solve this, syntax-based models (Galley et al., 2004; Chiang, 2005; Liu et al., 2006; Marcu et al., 2006) take tree struc- tures into consideration to learn translation pat- terns by using non-terminals for generalization. Model C D S (Koehn et al., 2003) • sequence (Galley and Manning, 2010) • • sequence (Quirk et al., 2005) and • tree (Menezes and Quirk, 2005) This work • • graph Table 1: Comparison between our work and pre- vious work in terms of three aspects: keeping continuous phrases (C), allowing discontinuous phrases (D), and input structures (S). However, the expressiveness of these models is confined by hierarchical constraints of the gram- mars used (Galley and Manning, 2010) since these patterns still cover continuous spans of an input sentence. By contrast, Quirk et al. (2005), Menezes and Quirk (2005) and Xiong et al. (2007) take treelets from dependency trees as the basic translation units. These treelets are connected and may cover discontinuous phrases. However, their mod- els lack the ability to handle continuous phrases which are not connected in trees but could in fact be extremely important to system performance (Koehn et al., 2003). Galley and Manning (2010) directly extract discontinuous phrases from input sequences. However, without imposing additional restrictions on discontinuity, the amount of ex- tracted rules can be very large and unreliable. Different from previous work (as shown in Ta- ble 1), in this paper we use graphs as input struc- tures and propose a …

Content truncated to 500 words