Learning non-isomorphic tree mappings for machine translation

Abstract

Often one may wish to learn a tree-to-tree mapping, training it on unaligned pairs of trees, or on a mixture of trees and strings. Unlike previous statistical formalisms (limited to isomorphic trees), synchronous TSG allows local distortion of the tree topology. We reformulate it to permit dependency trees, and sketch EM/Viterbi algorithms for alignment, training, and decoding.

Bootstrapping parsers via syntactic projection across parallel texts

“…The research presented here constitutes a successful first step in applying the annotation projection approach to syntactic representations. In contrast to the focus in research on tree-to-tree or bilingual grammar models (Wu 1997;Eisner 2003), we produce not a model but a treebank, which can be used for stochastic parser training or as the starting point for manual correction (Marcus et al 1993).…”

“…Whether or not stated explicitly, the DCA is actually an underlying assumption in most formal attempts to model cross-language correspondences in syntactic relationships (Wu 1995;Alshawi et al 2000;Yamada and Knight 2001;Eisner 2003;Gildea 2003;Melamed et al 2004;Smith and Smith 2004). Table 1 illustrates the principle with the following English-Basque sentence pair as an example:…”

Intension, Attitude, and Tense Annotation in a High-Fidelity Semantic Representation

“…Similarly, we expect machine translation methods such as Synchronous Tree Substitution Grammars (Eisner, 2003;Gildea, 2003) to be successful in automating this annotation because of the close syntactic correspondence to the surface form.…”

Large aligned treebanks for syntax-based machine translation

“…1 There are other, more complex types of synchronous grammars which can also be learned from parallel treebanks and which accept trees on both the left and right hand side of the grammar rules, allowing such operations as raising and lowering of nodes (Chiang, 2006). Depending on the operations that are allowed, we distinguish between synchronous tree substitution grammars (STSGs) (Schabes, 1990;Eisner, 2003) which allow substitution, synchronous tree insertion grammars (STIGs) (Nesson et al, 2006), which allow both substitution and insertion; and synchronous tree adjoining grammars (STAG) (Shieber and Schabes, 1990) which allow substitution, insertion and adjoinment. We used the parallel treebank, with alignments at the con-1 http://www.statmt.org/moses/ stituent level to train the PaCo-MT system, which is a syntax-based MT engine using an STSG transducer to bridge the gap between source sentence parse tree and target language parse tree.…”