🧑‍🔬 William A. Gale

William A. Gale works on statistical methods for processing text, with papers here from 1993 to 1996. Both deal with basic problems that must be solved before larger language systems can run: matching the sentences of a text with those of its translation, and finding word boundaries in writing that does not separate words with spaces. The methods rely on simple measurable signals and probability estimates rather than hand-built linguistic rules, and they connect to machine translation and to tokenization and segmentation.

The most cited paper here, A Program for Aligning Sentences in Bilingual Corpora (1993), written with Kenneth W. Church, describes a program called align. It pairs sentences across a text and its translation without using any knowledge of the words themselves. It depends instead on the fact that translated sentences tend to have similar lengths when measured in characters. Such alignment supports statistical translation and the building of bilingual dictionaries.

The later paper, A Stochastic Finite-State Word-Segmentation Algorithm for Chinese (1996), with Richard Sproat and other colleagues at Bell Laboratories, applies the same statistical approach to written Chinese. Segmentation is treated as a search through weighted finite-state transducers. A fixed dictionary is combined with separate models for words that no list fully covers, namely derived forms, Chinese personal names and transliterated foreign names. The Good-Turing method supplies costs for unseen cases of these kinds. Gale's papers carry the affiliation AT&T (United States) from 1985 to 2002, and a few papers from 1999 are listed under Johns Hopkins University.

William A. Gale has 2 papers in our library, in 1993–1996, most of them on Machine translation, Tokenization & segmentation; often with Kenneth W. Church, Richard Sproat.