ProphetMT: A Tree-based SMT-driven Controlled Language Authoring/Post-Editing Tool

ProphetMT: A Tree-based SMT-driven Controlled Language Authoring/Post-Editing Tools Xiaofeng Wu, Jinhua Du, Qun Liu, Andy Way ADAPT Centre, School of Computing, Dublin City University, Ireland {xiaofengwu,jdu,qliu,away}@computing.dcu.ie Abstract This paper presents ProphetMT, a tree-based SMT-driven Controlled Language (CL) authoring and post-editing tool. ProphetMT employs the source-side rules in a translation model and provides them as auto-suggestions to users. Accordingly, one might say that users are writing in a ‘Controlled Language’ that is ‘understood’ by the computer. ProphetMT also allows users to easily attach structural information as they compose content. When a specific rule is selected, a partial translation is promptly generated on-the-fly with the help of the structural information. Our experiments conducted on English-to-Chinese show that our proposed ProphetMT system can not only better regularise an author’s writing behaviour, but also significantly improve translation fluency which is vital to reduce the post-editing time. Additionally, when the writing and translation process is over, ProphetMT can provide an effective colour scheme to further improve the productivity of post-editors by explicitly featuring the relations between the source and target rules. Keywords: Controlled Language, Authoring Tool, Post-Editing, Statistical Machine Translation 1. Introduction A Controlled Language (CL) can be defined as “an ex- plicitly defined restriction of a natural language that spec- ifies constraints on lexicon, grammar, and style” (Huijsen, 1998). CL is widely used in professional writing where the aim is to write for a certain standard and style demanded by a particular profession, such as law, medicine, patent, tech- nique etc (Gough and Way, 2003; Gough and Way, 2004). For multilingual documents, CL has been shown to improve the quality of the translation output, whether the translation is done by humans or machines (Nyberg et al., 2003). The advantages of applying CL are self-evident: clear and consistent composition guidelines as well as less ambiguity in translation. However, the problems are also obvious: de- signing the rules usually requires human linguists, and rules may be difficult for end-users to grasp. In addition, the sen- tences that can be generated are often limited in length and complexity (O’Brien, 2003). To the best of our knowledge, most of the existing computer-aided authoring methods (Acrolinx, for example) employ a kind of interactive paradigm with a CL together with a grammar checker which provides user feedback. Users have to follow the ‘compose, check, revise’ loop un- til the sentence is consistent according to a given parser. Exceptions usually follow an approach called conceptual authoring (Hallett et al., 2007; Hart et al., 2008) where texts are created by short cycles of language generation and user-triggered modification actions. Obviously, these meth- ods severely restrict a user’s expressiveness in the authoring process. Even with the help of CL, current state-of-the-art ma- chine translation (MT) methods still fail to produce reliable outputs. For example, for English-to-Chinese translation, given the sentence “allows the client computers that con- nect through a token ring adapter to access the network”, it is very hard for computers to figure out: (i) the skeleton of this sentence is ‘allow …

Content truncated to 500 words