Artificial Intelligence

Generate-and-Test Models for Alignment and Machine Translation

When:
Friday, December 16, 2011, 3:00pm - 4:00pm PSTiCal
Where:
11th Floor Conf. Room (#1135)
Speaker:
Chris Dyer (Carnegie Mellon)
Description:

Abstract: I discuss translation as an optimization problem subject to three kinds of constraints: lexical, configurational, and constraints enforcing target-language wellformedness. Lexical constraints ensure that the lexical choices in the output are meaning-preserving; configurational constraints ensure that the relationships between source words and phrases (e.g., semantic roles and modifier-head relationships) are properly transformed in translation; and target-language wellformedness constraints ensure the grammaticality of the output. In terms of the traditional source-channel model of Brown et al. (1993), the "translation model" encodes lexical and configurational constraints and the "language model" encodes target language wellformedness constraints. On the other hand, the constraint-based framework suggests a generate-and-test (discriminative) model of translation in which features sensitive to input and output structures, and the feature weights are trained to maximize the (conditional) likelihood of a corpus of example translations. The specified features represent empirical hypotheses about what variables correlate (but not why) and thus encode domain-specific knowledge that is useful for the problem at hand; the learned weights indicate to what extent these hypotheses are confirmed or refuted.

To verify the usefulness of the feature-based approach, I discuss the performance two models: first, a lexical translation model evaluated by the word alignments it learns. Unlike previous unsupervised alignment models, the new model utilizes features that capture diverse lexical and alignment relationships, including morphological relatedness, orthographic similarity, and conventional co-occurrence statistics. Results from typologically diverse language pairs demonstrate that the generate-and-test model provides substantial performance benefits compared to state-of-the-art generative baselines. Second, I discuss the results of an end-to-end translation model in which lexical, configurational, and wellformedness constraints are modeled independently. Because of the independence assumptions, the model is substantially more compact than state-of-the-art translation models, but still performs significantly better on languages where source-target word order differences are substantial.

Bio: Chris Dyer is a postdoctoral researcher in Noah Smith's lab in the Language Technologies Institute at Carnegie Mellon University. He completed his PhD on statistical machine translation with Philip Resnik at the University of Maryland in 2010. Together with Jimmy Lin, he is author of "Data-Intensive Text Processing with MapReduce", published by Morgan & Claypool in 2010. Current research interests include machine translation, unsupervised learning, Bayesian techniques, and "big data" problems in NLP.

« Return to Events