Why Generative Phrase Models Underperform Surface Heuristics
نویسندگان
چکیده
We investigate why weights from generative models underperform heuristic estimates in phrasebased machine translation. We first propose a simple generative, phrase-based model and verify that its estimates are inferior to those given by surface statistics. The performance gap stems primarily from the addition of a hidden segmentation variable, which increases the capacity for overfitting during maximum likelihood training with EM. In particular, while word level models benefit greatly from re-estimation, phrase-level models do not: the crucial difference is that distinct word alignments cannot all be correct, while distinct segmentations can. Alternate segmentations rather than alternate alignments compete, resulting in increased determinization of the phrase table, decreased generalization, and decreased final BLEU score. We also show that interpolation of the two methods can result in a modest increase in BLEU score.
منابع مشابه
Generative Models of Monolingual and Bilingual Gappy Patterns
A growing body of machine translation research aims to exploit lexical patterns (e.g., ngrams and phrase pairs) with gaps (Simard et al., 2005; Chiang, 2005; Xiong et al., 2011). Typically, these “gappy patterns” are discovered using heuristics based on word alignments or local statistics such as mutual information. In this paper, we develop generative models of monolingual and parallel text th...
متن کاملA Detailed Analysis of Phrase-based and Syntax-based Machine Translation: The Search for Systematic Differences
This paper describes a range of automatic and manual comparisons of phrase-based and syntax-based statistical machine translation methods applied to English-German and English-French translation of user-generated content. The syntax-based methods underperform the phrase-based models and the relaxation of syntactic constraints to broaden translation rule coverage means that these models do not n...
متن کاملA generative grammar approach to diatonic harmonic structure
This paper aims to give a hierarchical, generative account of diatonic harmony progressions and proposes a generative phrase-structure grammar. The formalism accounts for structural properties of key, functional, scale and surface level. Being related to linguistic approaches in generative syntax and to the hierarchical account of tonality in the generative theory of tonal music (GTTM) [1], cad...
متن کاملWhy Steiner-tree type algorithms work for community detection
We consider the problem of reconstructing a specific connected community S ⊂ V in a graph G = (V,E), where each node v is associated with a signal whose strength grows with the likelihood that v belongs to S. This problem appears in social or protein interaction network, the latter also referred to as the signaling pathway reconstruction problem (Bailly-Bechet et al., 2011). We study this commu...
متن کاملRevisiting Recurrent Networks for Paraphrastic Sentence Embeddings
We consider the problem of learning general-purpose, paraphrastic sentence embeddings, revisiting the setting of Wieting et al. (2016b). While they found LSTM recurrent networks to underperform word averaging, we present several developments that together produce the opposite conclusion. These include training on sentence pairs rather than phrase pairs, averaging states to represent sequences, ...
متن کامل