In this chapter, we introduced ICALL as a melting pot at the crossroad of several research disciplines. We saw that ICALL research requires expertise from areas as diverse as Second Language Acquisition, Psycholinguistics, Human-Computer In- teraction, User Modelling, Foreign Language Teaching and Learning and Natural Language Processing – the last two being central to this thesis.
We introduced the main issues involved in the development of Intelligent Lan- guage Tutoring Systems by providing a detailed description of the seminal work on this topic. We characterised and compared the main features of the three Intelligent Language Tutoring Systems that present a longer and more sustained trajectory both in theory and in practice. We characterised their contexts of use and saw the importance of having the teachers involved from the beginning as a guarantee for the usefulness of the systems. We saw there was substantial progress in the ability to check for the correctness and well-formedness of learner responses, as well as on the adaptation of the system to the learner’s performance and its level.
As we are most interested in approaches that integrated ICALL solutions in lan- guage instruction settings following communicative approaches to language teaching, we reviewed the research that in one way or another has prioritised the pedagogical goal of the resulting ICALL applications. Particularly, we described the details of ICALL systems that prioritised their technical solutions as much as they prioritised their integration in instruction settings following a Task-Based Language Teaching approach.
We reviewed the first and only study that attempted to characterise ICALL activ- ities both in terms of pedagogical features and in terms of computational complexity in order to determine the range of FL learning activities that are pedagogically meaningful and computationally feasible. This led to the introduction of the con- cept viable processing ground. Making the viable processing ground concrete in form of a pedagogically- and computationally-informed activity characterisation framework is one of the goals of this thesis, namely the one tackled in Part III.
We emphasised the importance of involving teachers in the creation of CALL materials, and we described the little research that was actually carried out in this respect. This grounded an essential part of the second goal of the thesis, namely to empower teachers in real-world instruction settings to produce ICALL activities without the need to be trained in programming. This goal supposes an effort to transfer to teachers in real-world instruction settings the knowledge necessary to understand what is NLP capable of, as well as to understand their professional context and needs. This is the focus of Part IV.
Part II
Background
In the [then] recent broad Survey of the state of the art in human language technology (Cole et al. 1996), there is not a single word about (human) language learning. Similarly, CALL contributions to the biennal international conference on computational linguistics (COLING) have been next to nonexistent. [...] Thus, while certainly not part of the core of NLP, CALL seems not to have a place even in its periphery. [...]
The power of CALL (Pennington 1996), which, according to the back cover blurb, [...] “is destined to be the standard reference on CALL and the textbook of choice for teacher training courses covering the use of technology in language learning”, contains basically nothing on the uses of NLP in CALL.
Chapelle (1997, 1999, 2001) is not optimistic about the contributions of AI/NLP to CALL, although at least in her 2001 book, the NLP work that she reviews [...] is in most cases more than a decade old, and sometimes more than two decades, in a field which has seen very rapid development in the last ten years.
“What have you done for me lately? The fickle alignment of NLP and CALL” NLP in CALL – New Light Penetrates or No Longer Pertinent? 3rd Pre-conference Workshop at EUROCALL 2002 in Jyv¨askyl¨a (Finland) Lars Borin (2002: p. 2)
Chapter 3
Natural Language Processing
This chapter presents the methodological and technical background of this thesis from the perspective of Natural Language Processing. We introduce its object of study and its applications to real life. Our presentation introduces three relevant NLP issues for the purpose of this thesis: the technical approach, domain adaptivity, and the robustness of processing strategies. These three concepts impact the strategy followed to develop and use the NLP tools for the processing of learner language. The approach determines the way linguistic information is abstracted: on the basis of human insights and in form of linguistic principles, or on the basis of mathematically- sound algorithms capable of abstracting linguistic properties from annotated data.
Since an unrestricted approach to human language understanding by means of computer-based language processing is today unfeasible (by “unrestricted approach” we mean the possibility of having any given text processed and “understood” by a computer), the notion of domain and domain-specific NLP strategies is introduced. We review how domain adaptivity plays a central role in the usefulness and the feasibility of implementing NLP-enhanced real-life applications.
Finally, robustness is necessary for the stability of system behaviour, since we cannot afford a real-life application to break down even if the analysed language presents non-standard characteristics. We review the different techniques used in NLP to process ill-formed language, the term used in Linguistics and Computational Linguistics to refer to linguistic objects that do not comply with the standard gram- mar, or the generally accepted conventions, of the language – an abstraction of the linguistic competence of a native-speaker. In particular, we present the NLP research focusing on the analysis of learner language, one of the types of ill-formed language that received attention from the NLP community.