Sentence Alignment

Abstract

A sentence is probably the most well-studied linguistic unit in the history of linguistics and natural language processing. A sentence is usually characterized by a meaningful grammatical structure expressing some kind of statement, question, exclamation, request or command. Sentences are especially interesting in computational linguistics because they are, in many respects, self-contained and informative enough to make it possible to perform extensive syntactic and semantic analyses. On the other hand, they are restricted enough to allow efficient processing techniques (still complex enough to leave many open research questions). Therefore, many NLP applications focus on processing sentences (in isolation) instead of larger textual units; machine translation is one of them. Bitexts had been discovered for translation studies and soon thereafter also for data-driven techniques in automatic and semi-automatic translation. This created a lot of attention in the field of computational linguistics and brought a wide range of alignment techniques to appear in the late 1980’s and early 1990’s. Today, sentence-aligned bitexts are the most important resources for statistical machine translation but also for computer-aided translation (translation memories) and other machine translation approaches.