Grammatical metaphor, controlled language and machine translation

Grammatical Metaphor, Controlled Language and Machine Translation Margrethe H. Møller Institute of Business Communication and Information Science University of Southern Denmark Engstien 1 DK-6000 Kolding mhm@sitkom.sdu.dk Abstract It is a general assumption that 1) the readability and clarity of LSP texts written in a controlled language are better than uncontrolled texts and 2) that controlled languages produce bet- ter results with machine translation than uncontrolled languages. Con- trolled languages impose lexical, syn- tactic and pragmatic restrictions on the writing style of the author. We will focus on syntactic restrictions and in- vestigate whether a writing style in which various forms of grammatical metaphors have been dissolved in or- der to make the text more accessible to human readers, will in fact make the texts more suitable for MT. The basis of our investigation is a small corpus of English LSP texts that have been evaluated regarding their accessibility and acceptability by human users. The MT-system we will introduce as an additional ”user” of the texts is the English-Danish prototype of the Com- prendium MT-system. 1 Introduction It is a general assumption that LSP1 texts written in a controlled language are clearer and more read- able than uncontrolled texts2. It is also a general assumption that controlled languages produce bet- ter results with MT than uncontrolled languages3. As regards the first assumption, to our knowl- edge, only very few reports4 on what you may call “usability tests” of controlled languages have been published, although controlled languages are now widely used (Bernth and Gdaniec 2001). So it is interesting to study whether controlled language texts are actually considered more accessible and also acceptable within the genre in question by users or readers5. A few years ago, a survey was conducted at the University of Aalborg, Denmark (Lassen 1999, 2002, and 2003), in which the attitudes of different audiences to the accessibility and acceptability of a number of LSP texts, including text versions which may be regarded as controlled-language texts, were investigated. In our paper, we will investigate whether the controlled-language versions of these test texts are 1 Language for Special Purposes (LSP), as opposed to Lan- guage for General Purposes (LGP). 2 See e.g. Arnold et al. 1994; Wojcik and Hoard 1996 3 See e.g. Arnold et al. 1994. 4 See e.g. Holmback et al. 1996 5 See also Huijsen 1998 more suitable for machine translation than the original texts, and whether the text versions pre- ferred by the machine are also the ones preferred by the human users from the survey. 2 The Aalborg survey In a survey which formed the empirical basis of her ph.d. thesis, Inger Lassen distributed a ques- tionnaire to a variety of professional groups, in- cluding technical writers, translators, engineers, technical language instructors and a non-expert group of respondents with mixed occupations, who were unfamiliar with technical writing style. Each respondent was encouraged to state his or her atti- tude to the accessibility and acceptability6 of one out of six short texts which appeared in three …

Content truncated to 500 words