Regional survey: M(A)T in North America

MT Summit VII Sept. 1999 Regional Survey: M(A)T in North America1 Elliott Macklovitch RALI - DIRO Université de Montréal Canada Abstract We examine two North American case stud- ies, each of which illustrates a different strat-egy for coming to terms with high-volume, high- quality translation. The first eschews MT in favour of translation memory technology; the second employs a controlled language to simplify the input to an MT system. Both strategies betray a certain dissatisfaction with the current state of machine translation, al- though neither alternative, it turns out, fully lives up to its expectations. 1 Caveat Emptor Allow me to begin with a few disclaimers. The first are not my own, but are taken from the Web sites of two well-kown commercial MT developers whose identities need not be revealed here. “A word about language translation software: Because human language is complex, results will vary depending on the source text for each translation. Therefore consider your translations as drafts which enable you to quickly and inex- pensively handle day-to-day business communi- cations in foreign languages.” I think we can all agree that human language is complex and that this is indeed one of the principal reasons why machine translation is so very difficult. This being said, we surely do want our MT results to vary with the different source texts we submit. Just imagine if the contrary were the case! Ah, but perhaps I am being too literal here? Perhaps this caveat is ac- tually an illustration of another of the difficulties that plagues machine translation: the fact that people often don’t say or write what they really mean. Here, for instance, what the writer probably meant was that the quality of the MT results will vary with the quality of the source text. In fact, a linked page on this same site states that “the best way to help the quality of your 1 The abbreviation M(A)T stands for both machine translation and machine-aided translation, taken together. The latter is used to refer to various types of translator support tools, including so-called translation memories. translations is to make sure your source (original) text is well written.” But as our citation illustrates, this is easier said than done; and. as we shall see below, a text that is well-written for a machine is not necessar- ily the same as a well-written text for humans. Human translators understand that it is the author’s intended meaning which they must translate, and when they complain about a text being poorly written, it is often because they have difficulty reconstituting that mean- ing from the source text. MT systems, on the other hand, never complain; but that is only because they have as yet no way of realizing that discrepancies may exist between the intended meaning and the literal source text. The latter is all they have to go on. The following caveat is found among the Frequently Asked Questions on the Web site of another major MT vendor: “MT cannot replace a …

Content truncated to 500 words