The TALP-UPC Ngram-Based Statistical Machine Translation System for ACL-WMT 2008

Proceedings of the Third Workshop on Statistical Machine Translation, pages 127–130, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics The TALP-UPC Ngram-based statistical machine translation system for ACL-WMT 2008 Maxim Khalilov, Adolfo Hernández H., Marta R. Costa-jussà, Josep M. Crego, Carlos A. Henríquez Q., Patrik Lambert, José A. R. Fonollosa, José B. Mariño and Rafael E. Banchs Department of Signal Theory and Communications TALP Research Center (UPC) Barcelona 08034, Spain (khalilov, adolfohh, mruiz, jmcrego, carloshq, lambert, adrian, canton, rbanchs)@gps.tsc.upc.edu Abstract This paper reports on the participation of the TALP Research Center of the UPC (Universitat Politècnica de Catalunya) to the ACL WMT 2008 evaluation campaign. This year’s system is the evolution of the one we em- ployed for the 2007 campaign. Main updates and extensions involve linguistically motivated word re- ordering based on the reordering patterns technique. In addition, this system introduces a target language model, based on linguistic classes (Part-of-Speech), morphology reduction for an inflectional language (Spanish) and an improved optimization procedure. Results obtained over the development and test sets on Spanish to English (and the other way round) translations for both the traditional Europarl and a challenging News stories tasks are analyzed and commented. 1 Introduction Over the past few years, the Statistical Machine Transla- tion (SMT) group of the TALP-UPC has been develop- ing the Ngram-based SMT system (Mariño et al., 2006). In previous evaluation campaigns the Ngram-based ap- proach has proved to be comparable with the state-of- the-art phrase-based systems, as shown in Koehn and Monz(2006), Callison-Burch et al. (2007). We present a summary of the TALP-UPC Ngram- based SMT system used for this shared task. We dis- cuss the system configuration and novel features, namely linguistically motivated reordering technique, which is applied on the decoding step. Additionally, the reorder- ing procedure is supported by an Ngram language model (LM) of reordered source Part-of-Speech tags (POS). In this year’s evaluation we submitted systems for Spanish-English and English-Spanish language pairs for the traditional (Europarl) and challenging (News) tasks. In each case, we used only the supplied data for each lan- guage pair for models training and optimization. This paper is organized as follows. Section 2 briefly outlines the 2008 system, including tuple definition and extraction, translation model and additional feature mod- els, decoding tool and optimization procedure. Section 3 describes the word reordering problem and presents the proposed technique of reordering patterns learning and application. Later on, Section 4 reports on the experi- mental setups of the WMT 2008 evaluation campaign. In Section 5 we sum up the main conclusions from the pa- per. 2 Ngram-based SMT System Our translation system implements a log-linear model in which a foreign language sentence f J is translated into another language eI1 = f1, f2, ..., fJ 1 = f1, f2, ..., eI by searching for the translation hypothesis ˆeI 1 maximizing a log-linear combination of several feature models (Brown et al., 1990): ˆeI 1 = arg max eI 1 � M � m=1 λmhm(eI 1, f J …

Content truncated to 500 words