Readings in Machine Translation (review)

As the editors make clear in the introduction, this collection of reprinted papers about machine translation (MT) is not a survey of the state of the art but instead provides important papers in the field, many of which were previously difficult to find. As a consequence of the book’s intent, there is not a single paper in it that postdates the turn of the millennium. All of the papers are reprints and have undergone some editing (mostly minor things, correcting typos, etc.) and have been re-typeset (even to the extent of redrawing some figures and diagrams). Not all of the papers are reprinted in full; hence the occasional bracketed ellipses.

The book is divided into three sections. Each section has its own introduction, written by a different editor. Since the papers in each section are too numerous to list individually, each section is reviewed as a whole and not paper by paper.

Section 1 is historical. It contains papers dating up to the 1960s. Many of the papers are written by movers and shakers in the early days of MT, and quite a few names are recognizable even to outsiders to the field, such as Yehoshua Bar-Hillel. The papers run an interesting gamut in terms of their view of MT’s viability, ranging from the sanguine (e.g. the paper by Warren Weaver, who was instrumental in prioritizing MT research funding at the NSF) to the downright pessimistic (e.g. the paper by Yehoshua Bar-Hillel, which asserts that fully automated high quality translation (FAHQT) is ‘a dream’ which will not come true in the foreseeable future).

Section 2 covers theoretical and methodological issues in MT, and its contents are slightly less dated, with some papers as recent as 1990. The major issues addressed are the nine identified by Yorick Wilks: the role of knowledge in MT representations, the argument about representations, the question of language-specificity, the role of linguistic theory, analysis or degeneration-drivenness, the role of sublanguage, the scope of language and the issue of genre, MT versus machine assisted translation (MAT), and extensible or metaphoric lexical meaning.

Section 3 concentrates on system design. It focuses in part on particular systems but also extends the second section’s emphasis on methodological issues by looking at some of the issues involved in implementing real working systems. Its chronological scope is wider than the previous sections’ and includes not only some older classics but also some reports on systems from the 1990s. The papers cover a number of different kinds of systems, ranging from MAT systems to MT systems proper to speech translation systems.

One minor quibble with the book is that information about the source of the papers is listed at the end of the book, making retrieval of the original publication information for each paper somewhat inconvenient. It would be preferable to have that information provided at the beginning of each paper. The book will obviously appeal to anyone interested in MT and its history but would probably also be of interest to anyone interested in computational linguistics in general.