Non-uniform Language Detection in Technical Writing

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1892–1900, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Non-uniform Language Detection in Technical Writing Weibo Wang1, Abidalrahman Moh’d1, Aminul Islam2 Axel J. Soto3, and Evangelos E. Milios1 1Faculty of Computer Science, Dalhousie University, Canada {weibo,amohd,eem}@cs.dal.ca 2School of Computing and Informatics, University of Louisiana at Lafayette, USA aminul@louisiana.edu 3School of Computer Science, University of Manchester, UK axel.soto@manchester.ac.uk Abstract Technical writing in professional environ- ments, such as user manual authoring, re- quires the use of uniform language. Non- uniform language detection is a novel task, which aims to guarantee the consistency for technical writing by detecting sentences in a document that are intended to have the same meaning within a similar context but use dif- ferent words or writing style. This paper pro- poses an approach that utilizes text similarity algorithms at lexical, syntactic, semantic and pragmatic levels. Different features are ex- tracted and integrated by applying a machine learning classification method. We tested our method using smart phone user manuals, and compared its performance against the state-of- the-art methods in a related area. The experi- ments demonstrate that our approach achieves the upper bound performance for this task. 1 Introduction Technical writing, such as creating device operation manuals and user guide handbooks, is a special writ- ing task that requires accurate text to describe a cer- tain product or operation. To avoid ambiguity and bring accurate and straightforward understanding to readers, technical writing requires consistency in the use of terminology and uniform language (Farkas, 1985). There are always demands from modern in- dustries to improve the quality of technical docu- ments in cost-efficient ways. Non-uniform Language Detection (NLD) aims to avoid inner-inconsistency and ambiguity of techni- cal content by identifying non-uniform sentences. Such sentences are intended to have the same mean- ing or usage within a similar context but use differ- ent words or writing style. However, even though non-uniform sentences tend to have similar word- ing, similar sentence pairs do not necessarily indi- cate a non-uniform language instance. For example, here are four similar sentence pairs cited from the iPhone user manual (Apple Inc., 2015), where only two pairs are true non-uniform language instances: (1) tap the screen to show the controls. tap the screen to display the controls. (2) tap the screen to show the controls. tap the screen to display the controls. (3) if the photo hasn’t been downloaded yet, tap the download notice first. if the video hasn’t been downloaded yet, tap the download notice first. (4) you can also turn blue tooth on or off in con- trol center. you can also turn wi-fi and blue tooth on or off in control center. As we can see above, the pattern of difference within each sentence pair could be between one word and one word, or one word and multiple words, or one sentence having extra words or phrases that the other sentence does not have. Each pattern could be a …

Content truncated to 500 words