Learner Modelling
Abstract:
Focussing on learner-modelling and what can be gained for the user
This chapter contains information about user modelling in CALL systems as opposed to the previous more general chapter.
One aspect in this area is the need for collecting information about the user, i.e. the language learner.
Usually this can be done by "watching" the user interact with the system selecting pieces of information.
In the language learning scenario there is the additional chance to collect information from the learner about his performance while working on exercises.
As mentioned in the previous chapter a learner profile can then be generated.
Therefore there are two main areas for using a learner profile in CALL systems.
One is the adaptation of the system itself towards the user and the other one is the generation of "fitting" exercises.
One project, which aims at doing learner-modelling in a dictionary-setting is
ELDIT.
In the system 5 features can be adapted according to the learners needs.
|
Feature |
Setttings |
| model |
monolingual/semibilingual
|
| domain | general/medical/technical |
| difficulty | beginner/advanced |
| help | novice/familiar |
| pronunciation | local/standard |
From table
» Eldit one can see, that there are various possibilities to adapt the program to the user's needs.
Two main areas can be identified.
1. The general aspects of HCI.
A system can be adapted to needs and to the preferences of a learner in the simple tasks of interaction.
This can be e.g. the language used for buttons, hints and help-texts.
Some learners might prefer their L1 whereas more advanced learners might prefer the foreign language.
2. The presentation of learning materials.
An ICALL system should ideally adapt to the assumed knowledge of the learner about the foreign language he is learning.
This can only be done of course, if either the learner places himself on a proficiency-scale or the system analyses some exercises solved by the learner.
Another project is the
ICICLE-project.
This project aims at deaf native speakers of American Sign Language (ASL) to help them learn writing.
In the program user-modelling is done according to possible language acquisition models.
Comparing the learner performance of the learner with the acquisition model the system can e.g. establish hypothesis about the knowledge of the leaner.
A second aspect of learner modelling in this project is the more precise evaluation of language input.
Using the system the learner enters a sentence, which is parsed, i.e. morpho-syntactically analyzed.
But because more than one parse may result, a learner model helps to select the most viable one.
In the ICICLE-project this is done according to the so called overlay-model.
Every grammar rule in the grammar the parser uses is marked according to a certain proficiency level the learner can reach.
With the help of these markings and a seperate expert model the system can provide some information about the proficiency of the learner.
This also helps in determining a parse result to choose.
Furthermore the system updates the information about a certain learner.