Syntactic Models of Alignment for Machine Translation - 2005

Explore a comprehensive lecture on syntactic models for aligning parallel text in machine translation, delivered by Dan Gildea from the University of Rochester in 2005 at the Center for Language & Speech Processing, Johns Hopkins University. Delve into tree-based alignment models that utilize syntactic information for one or both languages, as well as models that infer structure directly from bilingual text. Examine theoretical aspects of Synchronous Context Free Grammars as a translation model, including a method to factor grammars for reducing synchronous parsing complexity. Learn about various concepts such as inversion transduction grammar, string and tree alignment, grouping, complexity, cloning, and lexicalized alignment. Gain insights into hybrid machine translation approaches, alignment errors, dependencies, and headword choice. This 1-hour 15-minute talk provides a deep dive into advanced topics in computational linguistics and machine translation.

Introduction
Presentation
Hybrid Machine Translation
Outline
Inversion transduction grammar
Grid vs tree version
String alignment
Tree alignment
Grouping
Complexity
Cloning
Example
Data
Alignment Errors
Dependencies
Lexicalized Alignment
Headword Choice
Braindead models