Grupos de Investigación » ARTIFICIAL INTELLIGENCE TECHNIQUES FOR L

Proyectos


What follows is a chronological list of the projects undertaken by this group and its members (accessible in the menu on the left). When available, there is a demonstration of the software generated in each project:
 

1. Virtual Verb Trainer (VVT)

2. Virtual Authoring Tool (VAT)

3. I-Peter I (Intelligent Personalised Tutoring Enviornment for Business English)

4. I-Peter II

5. COPPER (Collaborative Oral and Written English Production Environment)
 
6. I-AGENT (Intelligent Adaptive Generic English Tutor)
 
 
1. Virtual Verb Trainer
 

The Virtual Verb Trainer' is a tool that enables English language students to practise basic verb use within sentences. 


The most important feature of the system is its generative nature, which is to say that using implicit linguistic knowledge the system can produce nearly an infinite number of different examples of the 50 most common verbs in the English language. It is in essence a conjugation exercise. Incomplete sentences are presented to the user, and they are expected to enter the completed sentence below, conjugating the verb in the process.

Programs that automate the exercise of practising grammatical structures is by no means new (just look at the many CD based commercial examples around). The obvious problem with this type of system is that after the CD has been used for while it becomes obsolete. Has the student learnt the verb forms (for example) or merely learnt the examples that the system contains! The essential feature of The Virtual Verb Trainer, is its generative nature, something we believe to be a big step forward.

Interestingly enough the system does not contain a single hard coded example. Instead it contains the same linguistic rules and knowledge that the student will need to learn and dominate in order to be able to both understand and use the grammatical structures contained here. Example sentences are produced randomly, guided by syntactic rules and semantic restrictions. Since, as Chomsky argued, language by its very nature is generative, then the system can use these linguistic rules to generate a nearly infinite set of examples, combining together time clauses, verbs, subjects, direct and indirect objects, etc.

Whilst the extent to which language users are rule based animals can be debated, in this instance the question is essentially irrelevant. The two sides of the debate can be considered in terms of an adult second language learner (with presumably limited aditional neural plasticity to help!). In which case the language learning task can be conceptualised to be a question of:

  1. The explicit learning of the grammatical rules that make up the language, which subsequently, are built into the subject's mental lexicon, and used in language understanding and production.
  2. The implicit assimilation of language structures in a contextual situation, without the artificial use of rules.

Regardless of the conceptual and theoretical stance of the user regarding how language is learnt, the student can be seen to benefit from using it. The learning process can be conceptualised either in terms of a mechanical practise of pseudo-real sentences and grammatical structures or as an implicit exposure to (and the consequent repetition of the grammatically correct form) common linguistic structures.

The current version of The Virtual Verb Trainer contains a fairly basic sentence structure, that of: Subject + Verb + (Direct Object) + (Indirect Object) + Location + Company.

The interface to the system contains two text windows, a status message, two selector buttons, a help menu, and a series of buttons. Pressing the 'Give me a new sentence' button generates a random incomplete sentence which is placed in the top most text window. A sentence will typically contain three points of interest:

  1. An initial time clause which is important to locate the action 'in time'.
  2. A temporal point of reference, placed in square brackets, necessary to enable verb conjugation. In examples in which the reference point is evident, for example "Tomorrow morning...", no temporal reference point is provided.
  3. Finally, the verb of the sentence in infinitive form in brackets.

The user needs (depending on the answer entry option specified) to enter the complete correct sentence in the second text area below. Two important points need to be noted:

  1. The temporal information regarding the point of reference, presented in square brackets (if present) should NOT be included in the answer.
  2. The verb should be conjugated, and included in the sentence WITHOUT the brackets.

While this may appear a little complicated it is in fact very simple. Let's consider an example. The system could generate a sentence like:

'This month [1st, 06.00] he (to give) Bill's jacket to the opera singer in the city centre.'

In which case, the student can see that the time clause is 'this month' and that the observer is located on the first day of this month, at six o'clock in the morning. Therefore it is evident that we are considering a future time clause. Hence the student needs to consider one of the legal conjugations for this time clause, e.g., 'will give', and enter the correct sentence in the second text box, without the information in either the square or round brackets:

'This month he will give Bill's jacket to the opera singer in the city centre.'

To see if the student has answered the question correctly, he/she needs simply to press the 'Check my answer' button, whereupon the status of the answer will appear above, in this case:

Answer status: CORRECT, WELL DONE!

If the student is unable to conjugate the sentence correctly, he/she will be able to ask The Virtual Verb Trainer for the correct answer by pressing the 'Tell me the correct answer' button. At which point a dialogue box will appear with the possible correct answers.

If the student becomes bored of entering the entire sentence, selecting the 'Enter the verb only' option will make the system expect only the conjugated verb and not the complete sentence. In the previous example, a correct answer (remember that there is more than one) would be:

'will give'

The final feature that the system has is the ability to give help to the student regarding the conjugation required. As can be seen in the menu on the right hand side of the window the 'Current level of help' is set to 'None'. Clicking on top of the menu will reveal two levels of help, which are available: 'Verb tense' and 'Verb tense description'.

Selecting either of these help levels will cause the system to select (randomly) one of the legal conjugations for the sentence and present help information about that selection. Subsequent verb conjugation by the student, will of course, be restricted to the one selected by the system. Let us consider the previous sentence as an example. If the student selects the 'Verb tense' help level, the 'Answer status:...' message will change to:

Verb tense: modal future 'Verb tense: modal future Answer status:

That is to say that the system has selected the modal future verb tense for the example (from the three options possible: immediate future, modal future, and modal future continuous):

At this point the student can use this message to help them conjugate the verb correctly if they know, or can remember, what form the modal future takes. If they are not familiar with this term of reference, or simple require more help, the help level can be changed to 'Verb tense description', in which case the 'Answer status:...' message is changed to:

Verb tense: modal future (will + base verb form) Answer status:


At which point, hopefully, the student will realise that the modal future form of the verb is made up of the auxiliary 'will' together with the base form of the verb, i.e. the verb without the 'to'. For the previous example:

The verb in its infinitive form is: (to give)
Hence the correct conjugation is: will + give

 

2. Virtual Authoring Tool

 

 

This is a didactic tool that is being designed to assist with the difficult task of composing a written text. The text must not only be correctly expressed from a grammatical point of view, but also follow the standard stylistic patterns of a text of this type. The tool is designed to operate interactively with the user, with the double goal of increasing your didactic and operational efficiency. In order to do that, it provides you with all the necessary knowledge to undertake the correction of your own text. It presents the aspects of the language that are known to be particularly difficult for the average Spanish speaker, so that you can check your correction in the text. Furthermore, The Virtual Authoring Tool highlights all the words and punctuation marks in the essay related to the consulted rule, so that, after having carefully read it, you can revise the highlighted entries without wasting time and effort searching through the test.

The grammar included in the tool is divided into three parts which correspond to three major linguistic units: the text, the sentence and the word. Each part deals with the linguistic points that are best detected and studied within the corresponding linguistic unit. For example, the text level includes pragmatic issues about the rhetorical structure and the suprasentential connectors. The sentence level includes issues like the syntactic structures that focus on the new information and punctuation signs. The word level deals with all the syntactic categories except the conjunction. Although conjunctions often consist of a single word, their function is syntactic at a higher level than that of the phrase, namely the clause, so it is more correct to study them at the level of the sentence. Due to the history and current use of the English language around the world,

The Virtual Authoring Tool does not pretend to offer rules that attempt to cover all possible language occurrences, but provide some of the main correct and current uses of each linguistic aspect. For this reason, if you are absolutely certain of the existence of a particular use in the sub-genre of academic essays, you should not dismiss it necessarily if it is not currently included in the system

3. I-Peter I

The use of interactive on-line environments in distance learning appears to be promising because they solve some of the difficulties teachers have maintaining control of individual student progress due to the large number of students typically present in such courses. However, in the case of language learning, work is still required to design a system that is as flexible and effective as a good experienced teacher can be. I-PETER I (Intelligent Personalised English Tutoring EnviRonment) is presented as an advance in this area. Its student linguistic knowledge model is richer than that typically used in language teaching: a student's command of English is evaluated by interpreting his/her performance on specific linguistic areas in terms of related criteria, rather than by a vague linguistic competence ranking. This model enables error diagnosis to be undertaken using a Bayesian network, to reflect how teachers actually undertake this type of process in the classroom. The results of this diagnosis process enable a finer-grained control of material selection than is normally possible, giving rise to a course structure that is continuously adapted to individual student needs.

I-PETER I was developed as a prototype that would enable the system and underlying ideas to be tested before moving on to develop a full scale system that integrates all the networks and domain knowledge necessary to enable students to study the entire English language. Hence, certain aspects of the overall functionality of this prototype are greatly simplified with respect to the design of the more complete system, I-PETER II.

 

4. I-Peter II

 

I-Peter II (MEC2004-05758) takes off where I-Peter I finishes.

In order to be able to compare and contrast these two systems, and the relevant design features and objectives of the research group, the following table summarises the differences between the two:
 

Feature I-PETER I I-PETER II
State of development Version 1.0 fully operational In development. First prototype being tested and used in COPPER.
Development history and funding UNED Vicerrectorado de Investigación (2001-03) Spanish Ministerio de Educación y Ciencia (HUM2004-05758/FILO )(2004-07)
Universities involved 2: UNED + Universidad Antonio de Nebrija 3: UNED + Universidad Antonio de Nebrija + Universidad de Castilla la Mancha
Number of researchers 4
8 + 2 student research assistants
Coverage
Pilot system for verb conjugation Fully operational system (entire English language)
Subject of course General standard language Professional English
Linguistic variant American
International
Linguistic skills Written
Written and oral
Linguistic approach Traditional didactic grammar Functional and communicative
Highest linguistic unit of study Text (exercises at sentential level) Text
Linguistic levels Lexical, morpho-syntactic and semantic Lexical, morpho-syntactic, semantic and pragmatic and discourse
Didactic approach Miscellaneous unconnected texts: mainly exercise-based with some supportive theory and examples Real task-based (Authentic Activities) approach using a real business scenario: balanced combination of theory and examples, practical activities and leisure activities
Structure of course Traditional nomenclature in four
levels (Beginner, Lower Intermediate, Upper Intermediate, Advanced)
Based upon the Common European Framework of Reference for Languages (2001)
Materials used Text
Text, images, audio and video
Type of exercise Closed, multiple choice Closed and semi-closed exercises
Language of didactic materials
Spanish
English
Learning structure Spiral approach (mechanical reproduction + non-attentive application) Spiral approach (mechanical reproduction + non-attentive application)
Underlying data model
Relational data structures XML
Bayesian network model used
1 network 55 + 1 networks at two different knowledge levels

 5. COPPER

         The COPPER educational network intends to develop and implement the collaborative part of the ATLAS’ combined individual cognitive constructivist - collaborative social constructivist framework (Read et al., 2006). This framework includes a functional activity-based syllabus based upon the Common European Framework of Reference for Languages: Learning, Teaching, Evaluation (Council of Europe, 2001; henceforth, CEFR) and Intelligent CALL techniques that enable a system to perform as a skilled second language teacher would. The functionality of this framework is intended to enable our students to improve their overall communicative competence in tourist English in an effective way. In COPPER, the collaborative aspect as per our framewotk is to be fully developed, implemented, and tested with groups of students from Lengua Inglesa I of the Diplomatura de Turismo at the UNED.

 

1. Overall functionality

           In COPPER students will be required to perform an activity (which involves communication and interaction) in a specified time by working collaboratively, which can be subsequently analysed by all members of the group involved. While the students are obviously not explicitly obliged to use language from a given language proficiency level, the tasks are designed so that they enable specific language functions, constructions and vocabulary to be practised and hence consolidated. The general student task interface for an activity can be seen in figure 1. The screen contains a generic menu on the left hand side of the screen (used to access personal information, files and resources; use shared group resources and undertake the active tasks; monitor the work of lower level students; and finally, use the available tools), and an activity specific part, on the right. The activity task structure appears toward the top of the screen, with the active task shown in detail. Specifically, the task description is presented, together with a set of methodological and linguistic recommendations, and links to the resources available for it.

            T he results that the students generate can be accessed in one of two ways: either by activating the tool that they used to create them within the task, or by clicking on the Results generated link, which leads them to a summary screen, where evaluation information also appears. The interface is “stateful”, in the sense that a student’s current location within an activity, together with the selected user preferences, are stored when the session is finished. Hence, when the student logs in again, he is returned to the same position in which he was last working. Furthermore, any changes that have been produced in the group activity in which the student is currently working are also presented when he enters again. All four tools can be used either individually or collaboratively (synchronously or asynchronously), depending upon the availability of the students to work together and practical network bandwidth restrictions. Once the students in a group complete a task, it is labelled as such, so that the monitor assigned to assess the work undertaken can do so. Meanwhile, the students are free to continue with othertasks. Once the monitor’s evaluation of a task is available to the group members, they can make the modifications they feel appropriate. The monitor has to check such changes. Once the students agree that an activity is complete, the monitor enters one final time to evaluate the overall result.

 

 

6. I-AGENT

I-AGENT (FFI2008-06030) is the name of our ongoing project. It started in 2008 and is expected to finish at the end of 2011.

           The aim of this project is to investigate the way in which face-to-face English language classes can be optimally complemented by innovative ICALL (Intelligent Computer Assisted Language Learning) software. The software uses a linguistic formalism based upon the Systemic-Functional approach integrated with the notional-functional syllabus model from The Common European Framework of Reference for Languages: Learning, Teaching, Assessment (Council of Europe, 2001), and a cognitive and social constructivist pedagogic framework to enable individual and collaborative learning. After prototypical communicative language competence has been acquired in the classroom, the ICALL software is used to enable a student to apply and extend it. Each software activity has its counterpart in the classroom, and hence, the results of the computer-based work are subsequently consolidated in face-to-face sessions, following the constructivist paradigm. Persistent limitations in natural language processing which hinder the use of ICALL software for the analysis of unrestricted student linguistic production are sidestepped here by the system’s integration with relevant learning activities in the classroom. This software needs to be generic in the sense that it is reusable for any given language course simply by changing the contents (and their causal connections), thereby overcoming the stumbling block of Artificial Intelligence system building. This research will be materialized by the design and development of a computer system called I-AGENT (Intelligent Adaptive Generic ENglish Tutor).