IBM Scientific Award 2011 Distinguishes Doctorate of the CMU Portugal
|André Martins, doctorate of the dual degree doctoral program in Language Technologies (LTI) by the Instituto Superior Técnico of the Universidade Técnica de Lisboa (IST/UTL) and CMU, won the IBM Scientific Award 2011 for his paper “Turbo Parsers: Dependency Parsing by Approximate Variational Inference.” This work was written by André Martins, with his four advisors Pedro M.Q. Aguiar and Mário Figueiredo, from Instituto Superior Técnico of the Universidade Técnica de Lisboa (Portugal), Noah A. Smith and Eric Xing from Carnegie Mellon University.
In this paper, the authors address the problem of parsing natural language text, using methods of statistical inference. André Martins explains that “this is a difficult problem, since the natural languages are highly ambiguous and enable a wide variety of constructions.” The authors believe that the statistical methods are well suited to this problem because they are able to capture some of these linguistic phenomena automatically from corpora, but in their opinion this statistical methods are generally based on simplified models. Therefore, André Martins work “aims to fill this gap by building richer statistical models, without sacrificing the efficiency of parsing algorithms.”
Why is this work important? Well, the parsing of text is very relevant for applications such as text search, machine translation and information extraction. Many of these tools are used in a day-to-day basis, for example when looking for information on the Internet through a search engine. André Martins explains that “the automatic interpretation of a text also allows to organize and to retrieve information efficiently, with a response speed that surpasses humans.” In his opinion, “there is a huge technological potential in this area,” and the “advent of social networks introduces new problems for which these technologies may be relevant, such as media monitoring and analysis of opinion pieces.”
Through the Carnegie Mellon Portugal program, André Martins have had the opportunity to spend two years at Carnegie Mellon. “During this period I interacted with other students and faculty in the statistical learning and natural language processing natural fields, and I have learned a lot with them,” Martins says. About this experience, Martins misses the network spirit at CMU.
André Martins completed his Ph.D. in this academic year, 2011/2012, and is now part of the research team in Machine Learning of the company Priberam, which is a spin-off of the Instituto Superior Técnico of the Universidade Técnica de Lisboa, in Portugal.