We present Minimum Bayes-Risk (MBR) decoding for statistical machine translation. This statistical approach aims to minimize expected loss of translation errors under loss functions that measure translation performance. We describe a hierarchy of loss functions that incorporate different levels of linguistic information from word strings, word-to-word alignments from an MT system, and syntactic structure from parse-trees of source and target language sentences. We report the performance of the MBR decoders on a Chinese-to-English translation task. Our results show that MBR decoding can be used to tune statistical MT performance for specific loss functions. 1

In this paper, we present several confidence measures for large vocabulary continuous speech recognition. We propose to estimate the confidence of a hypothesized word directly as its posterior probability, given all acoustic observations of the utterance. These probabilities are computed on word graphs using a forward-backward algorithm. We also study the estimation of posterior probabilities on N-best lists instead of word graphs and compare both algorithms in detail. In addition, we compare the posterior probabilities with two alternative confidence measures, i.e., the acoustic stability and the hypothesis density. We present experimental results on five different corpora: the Dutch ARISE lk evaluation corpus, the German Verbmobil '98 7k evaluation corpus, the English North American Business '94 20k and 64k development corpora, and the English Broadcast News '96 65k evaluation corpus. We show that the posterior probabilities computed on word graphs outperform all other confidence measures. The relative reduction in confidence error rate ranges between 19% and 35% compared to the baseline confidence error rate.

. Initially introduced in the late 1960s and early 1970s, dynamic programming algorithms have become increasingly popular in automatic speech recognition. There are two reasons why this has occurred: First, the dynamic programming strategy can be combined with avery e#cient and practical pruning strategy so that very large search spaces can be handled. Second, the dynamic programming strategy has turned out to be extremely #exible in adapting to new requirements. Examples of such requirements are the lexical tree organization of the pronunciation lexicon and the generation of a word graph instead of the single best sentence. In this paper, we attempt to systematically review the use of dynamic programming search strategies for small#vocabulary and large#vocabulary continuous speech recognition. The following methods are described in detail: search using a linear lexicon, search using a lexical tree, language-model look-ahead and word graph generation. 1 Introduction Search strategie...

Minimum Error Rate Training (MERT) and Minimum Bayes-Risk (MBR) decoding are used in most current state-of-theart Statistical Machine Translation (SMT) systems. The algorithms were originally developed to work with N-best lists of translations, and recently extended to lattices that encode many more hypotheses than typical N-best lists. We here extend lattice-based MERT and MBR algorithms to work with hypergraphs that encode a vast number of translations produced by MT systems based on Synchronous Context Free Grammars. These algorithms are more efficient than the lattice-based versions presented earlier. We show how MERT can be employed to optimize parameters for MBR decoding. Our experiments show speedups from MERT and MBR as well as performance improvements from MBR decoding on several language pairs. 1

Abstract. In this article we describe methods for improving the RWTH German speech recognizer used within the VERBMOBIL project. In particular, we present acceleration methods for the search based on both within-word and across-word phoneme models. We also study incremental methods to reduce the response time of the online speech recognizer. Finally, we present experimental off-line results for the three VERBMOBIL scenarios. We report on word error rates and real-time factors for both speaker independent and speaker dependent recognition. 1

The goal of interactive machine translation is to improve the productivity of human translators. An interactive machine translation system operates as follows: the automatic system proposes a translation.

This paper describes the techniques we explored to improve the translation of news text in the German-English and Hungarian-English tracks of the WMT09 shared translation task. Beginning with a convention hierarchical phrase-based system, we found benefits for using word segmentation lattices as input, explicit generation of beginning and end of sentence markers, minimum Bayes risk decoding, and incorporation of a feature scoring the alignment of function words in the hypothesized translation. We also explored the use of monolingual paraphrases to improve coverage, as well as co-training to improve the quality of the segmentation lattices used, but these did not lead to improvements. 1

Recent work has shown that translating segmentation lattices (lattices that encode alternative ways of breaking the input to an MT system into words), rather than text in any particular segmentation, improves translation quality of languages whose orthography does not mark morpheme boundaries. However, much of this work has relied on multiple segmenters that perform differently on the same input to generate sufficiently diverse source segmentation lattices. In this work, we describe a maximum entropy model of compound word splitting that relies on a few general features that can be used to generate segmentation lattices for most languages with productive compounding. Using a model optimized for German translation, we present results showing significant improvements in translation quality in German-English, Hungarian-English, and Turkish-English translation over state-ofthe-art baselines. 1

This paper presents details of the RWTH large vocabulary continuous speech recognition system used in the VERBMOBIL spontaneous speech translation system. In particular, we report on methods for accelerating the search and algorithms for fast vocal tract normalization (VTN). We focus both on the improvements in word error rate and how to speed up the recognizer with only minimal loss in recognition accuracy. Implementation details and experimental results are given for the VERBMOBIL German development corpus dev99. The 24.6% word error rate of the baseline system is reduced to 22.8% using VTN. Decreasing the real-time factor by a factor of 5 resulted in only a small degradation in recognition performance of 2% relative on average. Furthermore, we study incremental methods for reducing the response time of the online speech recognizer and an efficient method to reduce the density of word graphs. 1. Introduction This paper describes the RWTH large vocabulary continuous speech recogniti...

In this paper we describe methods for improving the RWTH German speech recognizer used within the VERBMOBIL project. In particular, we present acceleration methods for the search based on both within-word and across-word phoneme models. The recognizer in the VERBMOBIL project is used in an online environment. We will discuss some incremental methods to reduce the response time of an on-line speech recognizer. We present experimental off-line results for the VERBMOBIL task, a German spontaneous speech corpus, and report on word error rates and real time performance of the search for both within-word and across-word phoneme models. 1. INTRODUCTION The goal of the VERBMOBIL project is to develop a speaker-independent speech-to-speech translation system that performs close to real-time. In this system, speech recognition is followed by subsequent VERBMOBIL modules (like syntactic analysis and translation) which depend on the recognition result. Therefore, in this application it is partic...