Linguistically Inspired Language Model Augmentation for MT

Linguistically Inspired Language Model Augmentation for MT George Tambouratzis, Vassiliki Pouli, Institute for Language and Speech Processing / Athena Research Centre, 6 Artemidos & Epidavrou, Paradissos Amaroussiou, Athens, GR-15125, Greece. E-mail: giorg_t@ilsp.gr , vpouli@ilsp.gr Abstract The present article reports on efforts to improve the translation accuracy of a corpus–based Machine Translation (MT) system. In order to achieve that, an error analysis performed on past translation outputs has indicated the likelihood of improving the translation accuracy by augmenting the coverage of the Target-Language (TL) side language model. The method adopted for improving the language model is initially presented, based on the concatenation of consecutive phrases. The algorithmic steps are then described that form the process for augmenting the language model. The key idea is to only augment the language model to cover the most frequent cases of phrase sequences, as counted over a TL-side corpus, in order to maximize the cases covered by the new language model entries. Experiments presented in the article show that substantial improvements in translation accuracy are achieved via the proposed method, when integrating the grown language model to the corpus-based MT system. Keywords: corpus-based MT, language model augmentation, extraction of knowledge from corpora. 1. Introduction Currently a large proportion of MT approaches which are readily portable to new language pairs are based on the Statistical Machine Translation (SMT) paradigm. The main obstacle to the creation of an SMT system is the requirement for parallel corpora between the Source Language (SL) and the Target Language (TL), which are of a sufficient size to allow the extraction of meaningful translation models. As the amount of parallel corpora is limited for most language pairs, researchers of SMT systems are investigating the extraction of information from monolingual corpora, including creation of lexica (Koehn et al., 2002) and of topic-specific information (Su et al., 2012). The present article studies a hybrid method for developing MT systems which has been developed as an alternative to SMT systems. This hybrid method is corpus-based and is intended to support the creation of MT systems using a common set of software modules, while having minimal requirements for specialized resources. This methodology is specifically designed to address the scarcity of parallel corpora from which an MT system is trained, using instead predominantly monolingual corpora. This allows the current methodology to be usable for language pairs for which parallel corpora are very limited. Based on this hybrid translation method and having as baseline the hybrid MT system developed in (Tambouratzis et al., 2013), the aim here is to investigate how the translation quality can be improved by augmenting the coverage of the Target Language Model (TLM). 2. The Translation Process The hybrid translation methodology studied here is built on a two-stage core translation engine developed in the framework of the PRESEMT project (www.presemt.eu) which uses very small parallel corpora (ca. 200 sentences) supplemented by monolingual corpora. The final resource that is used is a bilingual lexicon between SL and TL which contains the translations of terms. PRESEMT …

Content truncated to 500 words