Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Lexical Syntax for Dependency-based Language Models Statistical Machine Translation Incremental Dependency-based Language Model (IDLM) DTM2 Hany Hassan Dependency-based SMT Future Work DCU & IBM Conclusion and In collaboration with: Discussion Andy Way and Khalil Sima’an
Outline Hany Hassan Introduction Syntax for Phrase-based Introduction SMT Supertagged Phrase-based SMT Syntax for Phrase-based SMT From Supertagged to Dependency-based Language Models Supertagged Phrase-based SMT Incremental Dependency-based Language Model (IDLM) From Supertagged to Dependency-based Language Models DTM2 Dependency-based SMT Incremental Dependency-based Language Model (IDLM) Future Work Conclusion and Discussion DTM2 Dependency-based SMT Future Work Conclusion and Discussion
Outline Hany Hassan Introduction Syntax for Phrase-based Introduction SMT Supertagged Phrase-based SMT Syntax for Phrase-based SMT From Supertagged to Dependency-based Language Models Supertagged Phrase-based SMT Incremental Dependency-based Language Model (IDLM) From Supertagged to Dependency-based Language Models DTM2 Dependency-based SMT Incremental Dependency-based Language Model (IDLM) Future Work Conclusion and Discussion DTM2 Dependency-based SMT Future Work Conclusion and Discussion
Can linguistic syntax improve PBSMT? Hany Hassan Introduction (Koehn et al 2003) tried to impose syntactic constituents Syntax for Phrase-based SMT on phrase extraction Supertagged Phrase-based SMT From Supertagged to Dependency-based Hierarchical Phrase structure (Chiang 2005) Language Models Incremental ◮ Allows for hierarchical phrases Dependency-based Language Model (IDLM) ◮ Handles a range of reordering problems DTM2 ◮ The syntax induced is not linguistically motivated. Dependency-based SMT Future Work Conclusion and Syntactified target phrases (Marcu et. al. 2006) Discussion ◮ Induces millions of xRs rules from parallel corpus ◮ Mismatch between constituent (xRs) and phrase ◮ Subtrees for phrases: leads to spurious ambiguity in phrase table Do subtrees/constituents fit well with phrases?
Can linguistic syntax improve PBSMT? Hany Hassan Introduction (Koehn et al 2003) tried to impose syntactic constituents Syntax for Phrase-based SMT on phrase extraction Supertagged Phrase-based SMT From Supertagged to Dependency-based Hierarchical Phrase structure (Chiang 2005) Language Models Incremental ◮ Allows for hierarchical phrases Dependency-based Language Model (IDLM) ◮ Handles a range of reordering problems DTM2 ◮ The syntax induced is not linguistically motivated. Dependency-based SMT Future Work Conclusion and Syntactified target phrases (Marcu et. al. 2006) Discussion ◮ Induces millions of xRs rules from parallel corpus ◮ Mismatch between constituent (xRs) and phrase ◮ Subtrees for phrases: leads to spurious ambiguity in phrase table Do subtrees/constituents fit well with phrases?
Do subtrees/constituents fit well with phrases? Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to S Dependency-based Language Models Incremental VP Dependency-based Language Model (IDLM) DTM2 NP V NP Dependency-based SMT NP NP Future Work Conclusion and Discussion The president meets Saudi economic officials
Spurious Ambiguity: Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT S From Supertagged to Dependency-based Language Models Incremental VP Dependency-based Language Model (IDLM) DTM2 obj Dependency-based SMT NP Future Work Conclusion and NP V NP NP PP NP Discussion The president meets Saudi economical officials in Riad next week
Do subtrees/constituents fit well with phrases? Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Why subtress do not match SMT phrases? Phrase-based SMT From Supertagged to ◮ Syntactic constituents mismatch phrase concept Dependency-based Language Models ◮ Which level of tree structure should be incorporated ? Incremental Dependency-based ◮ This leads to spurious ambiguity Language Model (IDLM) DTM2 Dependency-based SMT Can linguistic syntax improve PBSMT? Future Work Conclusion and Discussion Trees/constituents do NOT fit well with phrases What syntax does fit then ?
Lexical Syntax (Supertags) Matches Phrases Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Dependency-based Language Models Incremental Dependency-based Language Model (IDLM) DTM2 Dependency-based SMT Future Work Conclusion and Discussion
Lexical Syntax (Supertags) Hany Hassan Introduction Syntax for Phrase-based SMT Linguistics offers lexical-syntax (Supertags): Supertagged Phrase-based SMT ◮ Lexicalized Tree Adjoining Grammar (LTAG) : (Joshi & From Supertagged to Dependency-based Schabes, 1992) & (Srinivas & Joshi, 1999) Language Models Incremental ◮ Combinatory Categorical Grammar (CCG) (Steedman,2000) Dependency-based Language Model (IDLM) DTM2 Rich lexical categories Dependency-based SMT Future Work ◮ Localizing syntactic dependencies Conclusion and Discussion ◮ Representing predicate argument constraints on the word level ◮ Markovian language model on the sequence produce almost parsing ◮ Handful of Combination Operators are used to construct dependency tree
Lexical Syntax (Supertags) Hany Hassan Introduction Syntax for Phrase-based SMT Linguistics offers lexical-syntax (Supertags): Supertagged Phrase-based SMT ◮ Lexicalized Tree Adjoining Grammar (LTAG) : (Joshi & From Supertagged to Dependency-based Schabes, 1992) & (Srinivas & Joshi, 1999) Language Models Incremental ◮ Combinatory Categorical Grammar (CCG) (Steedman,2000) Dependency-based Language Model (IDLM) DTM2 Rich lexical categories Dependency-based SMT Future Work ◮ Localizing syntactic dependencies Conclusion and Discussion ◮ Representing predicate argument constraints on the word level ◮ Markovian language model on the sequence produce almost parsing ◮ Handful of Combination Operators are used to construct dependency tree
Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Dependency-based Language Models Incremental The purchase price includes taxes Dependency-based Language Model (IDLM) DTM2 Dependency-based SMT Future Work Conclusion and Discussion
Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Dependency-based Language Models Incremental The purchase price includes taxes Dependency-based Language Model (IDLM) NP / NP (NP) NP (S \ NP) / NP NP DTM2 Dependency-based SMT Future Work Conclusion and Discussion
Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Dependency-based Language Models The purchase price includes taxes Incremental Dependency-based NP / NP NP (S \ NP) / NP NP (NP) Language Model (IDLM) > FA > FA DTM2 S \ NP NP Dependency-based SMT Future Work Conclusion and Discussion
Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Dependency-based The purchase price includes taxes Language Models Incremental NP / NP (NP) NP (S \ NP) / NP NP Dependency-based Language Model (IDLM) > FA > FA NP S \ NP DTM2 > FA Dependency-based SMT NP Future Work Conclusion and Discussion
Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to The purchase price includes taxes Dependency-based Language Models NP / NP (NP) NP (S \ NP) / NP NP Incremental Dependency-based > FA > FA Language Model (IDLM) NP S \ NP DTM2 > FA NP Dependency-based SMT < BA S Future Work Conclusion and Discussion
Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to The purchase price includes taxes Dependency-based Language Models NP / NP (NP) NP (S \ NP) / NP NP Incremental Dependency-based > FA > FA Language Model (IDLM) NP S \ NP DTM2 > FA NP Dependency-based SMT < BA S Future Work Conclusion and Discussion
Lexical Syntax for SMT Hany Hassan Introduction Syntax for Phrase-based SMT Supertagged Phrase-based SMT From Supertagged to Dependency-based Language Models Incremental Dependency-based Two levels of support: Language Model (IDLM) DTM2 ◮ Supertagged TM & LM Dependency-based SMT ◮ Fully incremental parsing Future Work Conclusion and Discussion
Outline Hany Hassan Introduction Syntax for Phrase-based Introduction SMT Supertagged Phrase-based SMT Syntax for Phrase-based SMT From Supertagged to Dependency-based Language Models Supertagged Phrase-based SMT Incremental Dependency-based Language Model (IDLM) From Supertagged to Dependency-based Language Models DTM2 Dependency-based SMT Incremental Dependency-based Language Model (IDLM) Future Work Conclusion and Discussion DTM2 Dependency-based SMT Future Work Conclusion and Discussion
Recommend
More recommend
