Writing Support for Controlled Natural Languages

Writing Support for Controlled Natural Languages Tobias Kuhn Department of Informatics University of Z¨urich Switzerland tkuhn@ifi.uzh.ch Rolf Schwitter Centre for Language Technology Macquarie University Sydney NSW 2109, Australia rolfs@ics.mq.edu.au Abstract In this paper we present interface tech- niques that support the writing pro- cess of machine-oriented controlled natu- ral languages which are well-defined and tractable fragments of English that can be translated unambiguously into a formal tar- get language. Since these languages have grammatical and lexical restrictions, it is important to provide a text editor that as- sists the writing process by using looka- head information derived from the gram- mar. We will discuss the requirements to such a lookahead text editor and introduce the semantic wiki AceWiki as an applica- tion where this technology plays an impor- tant role. We investigate two different ap- proaches how lookahead information can be generated dynamically while a text is written and compare the runtimes and prac- ticality of these approaches in detail. 1 Introduction Natural language interfaces have been a popular area of research in the 70’s and 80’s, in partic- ular natural language interfaces to databases re- ceived a lot of attention and some of these inter- faces found their way into commercial applica- tions (Copestake and Jones, 1990; Androutsopou- los et al., 1995). However, most of the early re- search prototypes focused on the exploration of specific NLP techniques and formalisms and were not very robust and therefore not very successful – and research drifted away from this topic in the 90’s although there is still no agreed consensus theory on how to build these text-based interfaces. The need for text-based natural language inter- faces has recently gained again momentum since more and more non-specialists are required to ac- cess databases and maintain knowledge systems in Semantic Web applications through their web browsers and portable devices (Cimiano et al., 2008). One advantage of natural language in- terfaces is that the user is not required to learn a formal language to communicate with the sys- tem. But the use of natural language as an inter- face style is not without problems since full nat- ural language is highly ambiguous and context- dependent. In practice, natural language inter- faces can only understand a restricted subset of written (or spoken) input, and they have often only limited capabilities to make their coverage transparent to the user (Chintaphally et al., 2007). If the system is unable to understand a sentence and does not provide accurate feedback, then there is the risk that the user makes potentially incorrect assumptions about other sentences that the system actually can understand. An obvious solution to this problem is to use a set of training sessions that teach the user what the system can and cannot understand. But this solution has at least two drawbacks: (a) untrained users cannot immediately use the system; and (b) infrequent users might not remember what they learned about the system’s capabilities (Tennant et al., 1983a). A better solution is to build natural language …

Content truncated to 500 words