Benjamin Han 1st choice: The utility of Mutual Information for Assessing MT quality Advisor: Alon Lavie 2nd choice: Optimizing a MT system using Evolutionary Strategy Advisor: Stephan Vogel --------------------------------------------------------------------- Shyamsundar Jayaraman 1st choice: Comparison of ROVER and MEMT Advisor: Bob Frederking 2nd choice: Advanced experimentation on MT Evaluation Metrics Advisor: Alon Lavie ---------------------------------------------------------------------- kerry hannan 1st choice: Optimizing a MT system using Evolutionary Strategy Advisor: Stephan Vogel 2nd choice: Rapid MT Prototyping using the AVENUE Transfer Framework Advisor: Alon Lavie ---------------------------------------------------------------------- "Simon Fung" 1st choice: Rapid MT Prototyping using the AVENUE Transfer Framework Advisor: Alon Lavie 2nd choice: Building a Chinese to English MT system with KANTOO Advisor: Teruko Mitamura ---------------------------------------------------------------------- Frank Lin 1st choice: Rapid MT Prototyping using the AVENUE Transfer Framework Advisor: Alon Lavie 2nd choice: Building a Chinese to English MT system with KANTOO Advisor: Teruko Mitamura My self-proposed project: "Selecting Correct Translations of High-Density Language Pairs Using the World Wide Web" This project focuses on selecting the correct translation of phrases or words from multiple translation outputs, mainly for MT applications in Cross-Lingual IR and Question-Answering, but can be extended to translation of text if source-target alignment is known. Problem: For many translation tasks, especially Cross-Lingual IR and QA, it is important to select the "most" correct translation of a given term. Sometimes a term can be transliterated correctly (proper names) or translated literally (technical terms, new terms), but is the translation in the form that is commonly used? This would depend on the coverage and the up-to-date-ness of the dictionary of a MT system, but how do we determine which MT system uses a better dictionary of a given term? Assumption: We can find the occurrence of the translated term and the co-occurrence of the term in both languages in web pages using a search engine to determine how likely the translation is a good translation. Tasks: -Build training and testing data -Program code to extract statistics from the web search engines -Tune parameters -Run tests -Analyze results