ALLES: Integrating NLP in ICALL Applications

This paper describes how mature NLP that has been successfully applied in the area of controlled language checking can be used to deliver intelligent CALL applications1. It describe how an autonomous, long-distance second-language learning system for advanced learners can be created. The architecture of the system consists of a multimodal user interface, a set of skill-specific learning tools, and a set of NLP-based evaluation tools. All modules are integrated in a flexible and scalable software architecture allowing for the use of NLP and ensuring a user-friendly environment based on advanced concepts in language didactics. The multimodal user interface is web-based and incorporate off-the-shelf ASR. The set of skill-specific learning tools consists of a reading, listening, speaking and a writing tool. The thrust of the project is to show the potential of NLP in evaluation of students’ productions. 1 The paper is based on work done in the ALLES project (Advanced Long-distance Language Education System) (IST-2001-34246). 2 The system is provided for four languages: English, Spanish, German, Catalan and confined to the domain of business and economy. Introduction This paper describes how mature linguistic resources (successfully applied in controlled language checking) are used for CALL applications to deliver an autonomous, long-distance 2-language learning course for advanced learners. The architecture of the system consists of a multi modal user interface, a set of skill-specific learning tools, and a set of NLP-based evaluation tools. All the modules are integrated in a flexible and scalable architecture providing a user-friendly environment. The set of learning tools includes reading, listening, speaking and writing tool. The NLP tools incorporate orthography, grammar and style checking facilities, a tool for evaluating linguistic richness and a domain-specific and exercisecentred content checking. The system will be implemented as a client-server architecture that allows for long distance remote access through Internet. The user interface will be based on web services and will be developed in a way to assure a high degree of usability. With regard to the learning module, there is an elaborate didactic design that is the basis for the specification of the learning sequence and the learning units which are composed of sets of tasks. Syllabuses have been designed in accordance with European and national standards for the teaching of second languages. The compilation of teaching material follows the principle of felicity and authenticity. NLP tools are used to evaluate learners’ output in a self-learning environment. They have been applied already successfully in for example controlled language checking. There is spell checking, grammar and style checking for non constrained language, for constrained language content and semantic checking. ALLES NLP tools provide the means with which to evaluate students’ production. The evaluation can be done from at least two viewpoints: 1. Linguistic correctness 2. Linguistic richness The most important aspect of this distinction is the fact that assessing linguistic correctness implies that the input text may contain ill-formed sentences, whereas assessing linguistic richness implies that the input text is correct. ALLES will produce tools and strategies that enable the automatic evaluation of written and oral texts both ways. Assessment of linguistic correctness As suggested above the task of evaluating linguistic correctness of a text is determined by the kind of errors the system expects from the learner. Traditionally, linguistic checking means correction of orthography, use of grammatical structure, adequate semantic use of words, and discourse structure. However, during the relatively recent history of NLP, only spell-checking and, partially, grammar checking have been tackled (Kukich 1992). This is a consequence of the fact that automatic evaluation of semantics and discourse seems to depend on full natural language understanding (which is not available for unrestricted text). ALLES will, however, explore how NL checking could be used beyond spell and grammar checking to limited contexts and based on pattern matching. An important consequence of this is that ALLES tools will have to be developed differently for global checking and exercise-specific checking: Global checking is about the detection and diagnosis of errors that may occur in any communicative context. Exercise specific checking refers to the detection and diagnosis of errors that are exclusively applicable in certain communicative contexts.