|
The group aims to leverage currently existing NLP tools in CALL applications for several different learner groups. The philosophy of the group is to avoid "reinventing the wheel" and to utilise existing NLP resources. A strong emphasis is placed on what the learner really wants to do with (I)CALL and how NLP applications can be successfully integrated into the learners' language learning process (rather than being "technology-driven").
One strand of research was the design, development, implementation and evaluation of an interactive plurilingual ICALL software system (ESPRIT) for contrastive learning of French, Spanish and Italian. Plurilingual teaching and learning of Romance languages exploits the similarities between these languages to teach them contrastively and to raise the language awareness of the learner. The learner is assumed to be an advanced speaker of at least one Romance language. The aim is to leverage the learner's current Romance
language knowledge in the learning process.
The ESPRIT toolset comprises dictionary tools, a multilingual concordancer based on small, specialised corpora, a plurilingual input analysis and feedback module (including a plurilingual parser), custom-made animated grammar presentations and an authoring tool for animated text. ESPRIT represents an interactive and flexible learning environment and is designed for autonomous learning. [PhD abstract]
Another area of research is to detect grammatical errors with existing
probabilistic parsers induced from syntactically annotated corpora
(treebanks). Such parsers are successfully employed in other applications,
for example machine translation, but today's grammar checkers often use
hand-crafted rule systems. Two approaches are considered that both use
machine learning methods and large corpora. The main difference is whether
ungrammatical training data is required. If grammatical training data is
sufficient, it will be much easier to adapt the grammar checker to other
languages.
The third strand of research is that of using NLP resources in CALL materials for Primary School students. In this context, there are two important constraints that can be availed of - one is the L1 language level of the students and the second is the scope of the L2 being studied (ab-initio). The idea is to avail of these constraints to use knowledge based NLP technology that in other more open contexts can be extremely difficult to scale.
Simple, small coverage DCGs can be used in the analysis of both existing
texts and student produced material. POS taggers and simple FSTs can also
be used in this context to save time in the generation of exercises and
audio materials.
Initial work on the building of an artificial co-learner resulted in
a cognate exercise that has been presented at the InSTIL/ICALL 2004
symposium.
|