Grammatical machine translation

Abstract

We present an approach to statistical machine translation that combines ideas from phrase-based SMT and traditional grammar-based MT. Our system incorporates the concept of multi-word translation units into transfer of dependency structure snippets, and models and trains statistical components according to stateof-the-art SMT systems. Compliant with classical transfer-based MT, target dependency structure snippets are input to a grammar-based generator. An experimental evaluation shows that the incorporation of a grammar-based generator into an SMT framework provides improved grammaticality while achieving state-of-the-art quality on in-coverage examples, suggesting a possible hybrid framework.

Randomized Significance Tests in Machine Translation

“…Riezler and Maxwell (2005) provide a comparison of approximate randomization with bootstrap resampling (distinct from paired bootstrap resampling), and conclude that since approximate randomization produces higher p-values for a set of apparently equally-performing systems, it more conservatively concludes statistically significant differences, and recommend preference of approximate randomization over bootstrap resampling for MT evaluation. Conclusions drawn from experiments provided in Riezler and Maxwell (2005) are oft-cited, with experiments interpreted as evidence that bootstrap resampling is overly optimistic in reporting significant differences (Riezler and Maxwell, 2006;Koehn and Monz, 2006;Galley and Manning, 2008;Green et al, 2010;Monz, 2011;Clark et al, 2011).…”

An Open Source Rule Induction Tool for Transfer-Based SMT

“…Statistical Machine Translation (SMT) using deep linguistic representations for transfer is a relatively new and growing research area (Bojar and Hajič, 2008, Graham, 2008, Bojar et al, 2007a, Bojar and Čmejrek, 2007b, Graham et al, 2007, Ding and Palmer, 2006, Riezler and Maxwell, 2006, Ding and Palmer, 2005, Ding and Palmer, 2004a, Ding and Palmer, 2004b, Cmejrek et al, 2003, Eisner, 2003, Ding and Palmer, 2003, Hajič et al, 2002, Alshawi et. al., 2000a, Alshawi et.…”

Sulis: An Open Source Transfer Decoder for Deep Syntactic Statistical Machine Translation

“…Bojar and Hajič (2008) use the Synchronous Tree Substitution Grammar formalism, in which the probability of attaching pairs of dependency treelets into aligned pairs of frontiers given frontier state labels is used as a main feature function, as well as a bigram dependency-based language model. Riezler and Maxwell (2006), on the other hand, use a translation model computed from relative frequencies of automatically induced transfer rules and a trigram dependencybased language model. Riezler and Maxwell (2006) diverge somewhat from PB-SMT, however, by only applying the language model after decoding on the n-best decoder output.…”

“…Riezler and Maxwell (2006), on the other hand, use a translation model computed from relative frequencies of automatically induced transfer rules and a trigram dependencybased language model. Riezler and Maxwell (2006) diverge somewhat from PB-SMT, however, by only applying the language model after decoding on the n-best decoder output.…”

“…For example, both Riezler and Maxwell (2006) and Bojar and Hajič (2008) use beam search decoding and a log-linear model to combine feature scores when transferring from source language (SL) deep syntactic structure to target language (TL) deep syntactic structure. Each uses different statistical models for transfer, however.…”