|
Greg Hanneman
General Contact: Personal Contact: |
|
||
|
Current Work I came to Carnegie Mellon in the fall of 2005, got my master's degree in the spring of 2007, and am currently a third-year graduate student (first-year Ph.D. student) at the LTI. My advisor is Dr. Alon Lavie. Multi-Engine Machine Translation (MEMT) Given translations of a common input produced by independent machine translation systems, the MEMT engine produces a new synthetic translation by combining elements of the individual outputs. The original translations are first aligned to each other in a word-matching phase; after that, synthetic combinations of words are produced by a decoder and evaluated against a language model. The MEMT algorithm is explained in more detail in Jayaraman and Lavie, "Multi-Engine Machine Translation Guided by Explicit Word Matching," Proceedings of the 10th Annual Conference of the European Association for Machine Translation, 2005. I gave a talk called "When Does Multi-Engine Machine Translation Work and Why?" at the LTI's Student Research Symposium in October 2006. French–English MT I'm also working on a French-to-English translation system that grew out of a semester project for the LTI's machine translation class and is now one of many MT systems developed under the AVENUE project's statistical transfer MT framework. The system makes use of a large parallel parsed corpus and statistical word alignment to automatically build a large syntax-driven translation lexicon. A synchronous transfer grammar is also automatically learned from the parse trees. Translation is carried out using the lexicon and grammar by the "Stat-XFER" framework, which constructs a lattice of possible translations for a given input and selects a final output hypothesis after a decoding process. Our large-scale French and German systems, built this way, are described in Hanneman, Huber, Agarwal, Ambati, Parlikar, Peterson, and Lavie, "Statistical Transfer Systems for French–English and German–English Machine Translation," Proceedings of the Third Workshop on Statistical Machine Translation, 2008. More details on an original system that used a manual grammar are available in my final report for the class project. Teaching I was the Fall 2007 TA for Algorithms for Natural Language Processing (11-711). You can visit the class website or the detailed schedule. Courses
Past Life In May 2005 I received a B.S. degree in computer science with a minor in French from Case Western Reserve University in Cleveland, Ohio. I spent all eight semesters at Case as a staff member of The Observer student newspaper, where I was a reporter, layout editor, copy editor, and news editor. I'm originally from the Cleveland suburbs. Outside Interests I work on the copy staff of The Tartan, CMU's student newspaper, where I also previously wrote occasional articles for the news section. Aside from a certain neurotic obsession with writing and editing that I picked up from my newspaper work, I first got interested in language by studying French in high school and as an undergraduate. I also enjoy picking out speech variations and "accents" from different parts of the U.S. When I'm not thinking about things academic, I try to make time for photography (and developing my own black-and-white prints), reading, camping, cooking with friends, and indoor rock climbing. I've also recently started learning Thai. My nascent digital photography site is here. |
Page created 26 August 2005; last updated 28 June 2008 in glorious Technicolor CSS.
