Exploiting comparable corpora


November 29, 2006


Dragos Munteanu


University of Southern California


One of the major bottlenecks in the development of Statistical Machine Translation systems for most language pairs is the lack of bilingual parallel training data. Currently available parallel corpora span relatively few language pairs and very few domains; building new ones of sufficiently large size and high quality is time-consuming and expensive. In this talk, I will present methods that enable automatic creation of parallel corpora by exploiting a rich, diverse, and readily available resource: comparable corpora. Comparable corpora are bilingual texts that, while not parallel in the strict sense, are somewhat related and convey overlapping information. Such texts exist in large quantities on the Web; a good example are the multilingual news feeds produced by news agencies such as Agence France Presse, CNN, and BBC. I will present novel methods for extracting good-quality parallel data from such comparable collections. I will show how to detect parallelism at various granularity levels, and thus find parallel documents (if there are any in the collection), parallel sentences, and parallel sub-sentential fragments. In order to demonstrate the validity of this approach, I use my method to extract data from large-scale comparable corpora for various language pairs, and show that the extracted data helps improve the end-to-end performance of a state-of-the art machine translation system.


Dragos Munteanu

Dragos Munteanu is a Ph.D. student at the University of Southern California, Information Sciences Institute, where he recently defended his Ph.D. thesis. He received his M.Sc in Computer Science from the University of Iowa, and B.S. from the university of Bucharest, Romania. His interests include machine translation, summarization, and machine learning.