The intersection of tree transducer-based translation models with n-gram language models results in huge dynamic programs for machine translation decoding. We propose a multipass, coarse-to-fine approach in which the language model complexity is incrementally introduced. In contrast to previous orderbased bigram-to-trigram approaches, we focus on encoding-based methods, which use a clustered encoding of the target language. Across various encoding schemes, and for multiple language pairs, we show speed-ups of up to 50 times over single-pass decoding while improving BLEU score. Moreover, our entire decoding cascade for trigram language models is faster than the corresponding bigram pass alone of a bigram-to-trigram decoder. 1

String-to-string transduction is a central problem in computational linguistics and natural language processing. It occurs in tasks as diverse as name transliteration, spelling correction, pronunciation modeling and inflectional morphology. We present a conditional loglinear model for string-to-string transduction, which employs overlapping features over latent alignment sequences, and which learns latent classes and latent string pair regions from incomplete training data. We evaluate our approach on morphological tasks and demonstrate that latent variables can dramatically improve results, even when trained on small data sets. On the task of generating morphological forms, we outperform a baseline method reducing the error rate by up to 48%. On a lemmatization task, we reduce the error rates in Wicentowski (2002) by 38–92%. 1

Current tree-to-tree models suffer from parsing errors as they usually use only 1-best parses for rule extraction and decoding. We instead propose a forest-based tree-to-tree model that uses packed forests. The model is based on a probabilistic synchronous tree substitution grammar (STSG), which can be learned from aligned forest pairs automatically. The decoder finds ways of decomposing trees in the source forest into elementary trees using the source projection of STSG while building target forest in parallel. Comparable to the state-of-the-art phrase-based system Moses, using packed forests in tree-to-tree translation results in a significant absolute improvement of 3.6 BLEU points over using 1-best trees. 1

Combinatorial Categorial Grammar (CCG) supertags present phrase-based machine translation with an opportunity to access rich syntactic information at a word level. The challenge is incorporating this information into the translation process. Factored translation models allow the inclusion of supertags as a factor in the source or target language. We show that this results in an improvement in the quality of translation and that the value of syntactic supertags in flat structured phrase-based models is largely due to better local reorderings. 1

In this paper, we apply a method of unsupervised morphology learning to a state-of-the-art phrase-based statistical machine translation (SMT) system. In SMT, words are traditionally used as the smallest units of translation. Such a system generalizes poorly to word forms that do not occur in the training data. In particular, this is problematic for languages that are highly compounding, highly inflecting, or both. An alternative way is to use sub-word units, such as morphemes. We use the Morfessor algorithm to find statistical morphemelike units (called morphs) that can be used to reduce the size of the lexicon and improve the ability to generalize. Translation and language models are trained directly on morphs instead of words. The approach is tested on three Nordic languages (Danish, Finnish, and Swedish) that are included in the Europarl corpus consisting of the Proceedings of the European Parliament. However, in our experiments we did not obtain higher BLEU scores for the morph model than for the standard word-based approach. Nonetheless, the proposed morph-based solution has clear benefits, as morphologically well motivated structures (phrases) are learned, and the proportion of words left untranslated is clearly reduced.

The alignment problem—establishing links between corresponding phrases in two related sentences—is as important in natural language inference (NLI) as it is in machine translation (MT). But the tools and techniques of MT alignment do not readily transfer to NLI, where one cannot assume semantic equivalence, and for which large volumes of bitext are lacking. We present a new NLI aligner, the MANLI system, designed to address these challenges. It uses a phrase-based alignment representation, exploits external lexical resources, and capitalizes on a new set of supervised training data. We compare the performance of MANLI to existing NLI and MT aligners on an NLI alignment task over the well-known Recognizing Textual Entailment data. We show that MANLI significantly outperforms existing aligners, achieving gains of 6.2 % in F1 over a representative NLI aligner and 10.5 % over GIZA++. 1

We present a unified view of many translation algorithms that synthesizes work on deductive parsing, semiring parsing, and efficient approximate search algorithms. This gives rise to clean analyses and compact descriptions that can serve as the basis for modular implementations. We illustrate this with several examples, showing how to build search spaces for several disparate phrase-based search strategies, integrate non-local features, and devise novel models. Although the framework is drawn from parsing and applied to translation, it is applicable to many dynamic programming problems arising in natural language processing and other areas. 1

We offer a simple, effective, and scalable method for statistical machine translation parameter tuning based on the pairwise approach to ranking (Herbrich et al., 1999). Unlike the popular MERT algorithm (Och, 2003), our pairwise ranking optimization (PRO) method is not limited to a handful of parameters and can easily handle systems with thousands of features. Moreover, unlike recent approaches built upon the MIRA algorithm of Crammer and Singer (2003) (Watanabe et al., 2007; Chiang et al., 2008b), PRO is easy to implement. It uses off-the-shelf linear binary classifier software and can be built on top of an existing MERT framework in a matter of hours. We establish PRO’s scalability and effectiveness by comparing it to MERT and MIRA and demonstrate parity on both phrase-based and syntax-based systems in a variety of language pairs, using large scale data scenarios. 1

Domain adaptation has recently gained interest in statistical machine translation to cope with the performance drop observed when testing conditions deviate from training conditions. The basic idea is that in-domain training data can be exploited to adapt all components of an already developed system. Previous work showed small performance gains by adapting from limited in-domain bilingual data. Here, we aim instead at significant performance gains by exploiting large but cheap monolingual in-domain data, either in the source or in the target language. We propose to synthesize a bilingual corpus by translating the monolingual adaptation data into the counterpart language. Investigations were conducted on a stateof-the-art phrase-based system trained on the Spanish–English part of the UN corpus, and adapted on the corresponding Europarl data. Translation, re-ordering, and language models were estimated after translating in-domain texts with the baseline. By optimizing the interpolation of these models on a development set the BLEU score was improved from 22.60% to 28.10 % on a test set. 1

Current statistical machine translation systems usually extract rules from bilingual corpora annotated with 1-best alignments. They are prone to learn noisy rules due to alignment mistakes. We propose a new structure called weighted alignment matrix to encode all possible alignments for a parallel text compactly. The key idea is to assign a probability to each word pair to indicate how well they are aligned. We design new algorithms for extracting phrase pairs from weighted alignment matrices and estimating their probabilities. Our experiments on multiple language pairs show that using weighted matrices achieves consistent improvements over using n-best lists in significant less extraction time. 1