A Review of Intelligent CALL Systems Johann Gamper Judith Knapp Free University of Bozen European Academy of Bozen Mustergasse 4 Drususallee 1 39100 Bozen, Italy 39100 Bozen, Italy
[email protected] [email protected] June 14, 2002 Abstract Computer-assisted language learning is a research field which explores the use of computational methods and techniques as well as new media for language learning and teaching. Recently, an increasing number of language learning systems have been developed which adopt Artificial Intelligence techniques. This paper provides an overview about intelligent computer-assisted language learning. The most advanced systems have been investigated and classified along five dimensions: supported languages, Artificial Intelligence techniques, language skills, language elements, and availability. The paper concludes with a discussion about outstanding problems which still need further research in order to exploit the full potential of intelligent technologies in modern language learning environments.
1
Introduction
The computer as an ubiquitous tool is influencing almost everything in human life. This includes also the domain of education, where new technologies dramatically change the storage and presentation of information (printed documents are substituted by multimedia digital content), the interaction between teacher and learner, the collaboration between students, etc. The use of these new technologies and media for language learning and teaching has become an own discipline, known as computer-assisted language learning (CALL). Intelligent computer-assisted language learning (ICALL) explores the use of Artificial Intelligence methods and techniques for language learning. In this paper we present a survey about ICALL systems. We do not consider commercially available system, rather we will focus on research prototypes which explore the potential of leading edge technologies. As the gap between research prototypes and commercial systems is becoming smaller and the time to market research results is becoming shorter, some of the technologies discussed here are already available in one form or another in commercial systems. While such systems are certainly very useful and improve previous systems, there is still a great potential in new technologies and media, which has to be explored. The research prototypes discussed here touch some of these new possibilities and indicate directions for future investigations.
The paper is organized as follows: Chapter 2 sketches the historical development of ICALL. In chapter 3 we introduce a framework for the classification of ICALL systems, which is based on the five dimensions language, intelligent features, language skills, language elements, and availability. In chapter 4 we present in detail the results of our analysis. Chapter 5 discusses some open research issues and outstanding problems.
2
Evolution of ICALL
ICALL started as an own research field about a decade ago, when some Artificial Intelligence (AI) technologies were mature enough to be included in language learning systems, at least in experimental settings. The beginning of the new field was characterized by Intelligent Tutoring Systems (ITS), which embedded some Natural Language Processing (NLP) features to extend the functionality of traditional language learning systems. The continuos advances in ICALL systems over the last ten years has been documented in several publications. The first seminal publication is the book Intelligent Tutoring Systems for Foreign Language Learning, edited by M. L. Swartz and M. Yazdani in 1992. This book forms a systematic basis for this multidisciplinary research discipline and tries to combine ITS and NLP to build an instructional framework for language learning. Several intelligent methods such as grammar checking, error analysis, user modeling, and tutoring are discussed and how they can be adapted and combined to be useful for language learning systems. Two years later in 1994, a special issue of the International Journal of Artificial Intelligence in Education was dedicated to the topic of language learning. Besides the technological aspects, researchers begun to include results from pedagogy and cognitive science as well as to consider teaching pragmatics and socio-linguistic competence. System development was still at a prototypical stage. At that time, none of the systems presented in the issue was used in real learning situations. In 1995 the book Intelligent Language Tutors – Theory Shaping Technology was published. While the main focus was still on ITS augmented with NLP, many systems were presented which provided a more comprehensive language teaching environment including negotiations and discourse between the learner and the system. NLP technologies were developed to train fluency, systems became available on the personal computer, and huge computational lexicons as well as large linguistic databases for different languages have been added. In the next years, research in Automated Speech Recognition (ASR) matured, and the technology became powerful enough to be included into ICALL systems to support the training of pronunciation and communication skills. In 1999 the CALICO journal devoted a special issue to speech recognition techniques for language learning. Most of the systems presented there implemented discrete speech recognition. This technology is more reliable than continuos speech recognition, but the user is limited and has to reproduce one from several predefined patterns. Continuous speech recognition allows to process freely produced sentences. Only a few systems implemented this technology. The book CALL – Media, Design & Applications, published in 1999, devotes some chapters to intelligent technologies used in CALL. The use of ASR and NLP were again the main topics. At that time, both technologies were successfully implemented in some fully working systems, but they had to be supported by alternate input modalities or limitations of the domain of discourse. The inclusion of semantics and pragmatics in the recognition process
was still an open problem. In 2001 a workshop on ICALL was held at the Conference on Artificial Intelligence in Education. Again ASR and NLP projects were presented, but also some new approaches like the inclusion of AI techniques for vocabulary acquisition, the use of machine translation, or video annotation were discussed.
3
Classification Framework
In this section we introduce a classification framework which allows us to analyze ICALL systems along the following five dimensions: the supported languages, the applied AI technologies, the language skills which can be trained, the language elements which can be learned, and the availability of the systems. The first dimension concerns the target languages which can be learned with the system: English (e), Japanese (j), French (f), German (d), Spanish (s), Italian (i), Russian (r), Greek (g), Mandarin Chinese (c), Arabic (a), Hebrew (h), Thai (t), and Malai (m), the national language of Singapore. An asterisk (*) means that the system is language-independent. The core part in our analysis concerns the use of AI systems and techniques in computerassisted language learning. Some approaches follow the line of expert systems (EX) and store domain knowledge, which allows to provide detailed feedback to the learner. Intelligent tutoring systems (ITS) guide the user individually through the learning space. Some systems store information about each individual learner in a so-called user model. In some systems the user model is aimed to be inspected by the learner or the teacher (UM). Other systems use this information to adapt to the needs of the individual user (A). Natural language processing (NLP) and natural language generation (NLG) are two other technologies which are mainly applied in systems for training writing skills. Systems for training speaking skills often include an automated speech recognition (ASR) component which allows to control the learner’s pronunciation. Finally, machine translation (MT) is used in a couple of projects in order to enhance communication and translation skills. The third dimension classifies ICALL systems according to different language skills which can be trained. A common, high-level distinction is between reading (R), writing (W), listening (L), and speaking (S) skills. In our analysis we consider translating (T) as an additional high-level language skill. Speaking skills are further divided into pronunciation (p), fluency (f), and social skills (s). In the fourth dimension we classify systems according to different language elements which can be trained with the system. The present evaluation distinguishes between grammar (G), vocabulary (V), and dialogue elements (D). Grammar may be approached in two different ways: deductively (d), where students are given a rule which they practice, or inductively (i), where students infer the rules by themselves. Vocabulary acquisition can occur intentionally (t) or incidentally (c). Finally, we indicate whether a system is Web-based (w), a demo version is available (a), or whether the product can be purchased (p).
4
Detailed Analysis
In this section we present a detailed analysis of 40 ICALL systems according to the classification framework introduced above. Table 1 summarizes the results. In many cases it is rather difficult to make a clear distinction between the different features of the systems, especially for AI technologies and language skills. In these cases we stress the main features of the systems.
4.1
Languages
The first dimension concerns the target languages which can be learned with the system. By far the most language learning programs have been developed for English, followed by Japanese, French, and German. As the mistakes in foreign language learning very often depend on the mother tongue of the student, many prototypes in fact have been implemented for a language pair in the sense that a specific source language has been considered in the development process. This allows among other things to use error grammars which model typical mistakes of students with that specific native language. Moreover, the menu and explanations can be shown in the source language.
4.2
Artificial Intelligence Systems and Techniques
Over the last decade, more and more AI techniques have been adopted in CALL systems. While the first systems mainly focused on expert systems and NLP techniques, nowadays also ASR and MT are frequently used as well as a combination of different techniques. 4.2.1
Expert Systems
Expert systems (EX) for language learning store a large body of knowledge about language learning such as typical mistakes, learning strategies, questions and answers, etc. This knowledge is then used for the analysis of the student’s interaction with the system including the submitted answers and allows to provide a more detailed feedback than in traditional systems [4, 18, 38]. For example, the system Spengels [4] is equipped with knowledge about the spelling of verb forms. The student can either learn the spelling and conjugation rules from scratch or practice them by filling in gapped sentences. Error specific feedback can be given at each stage of the learning process. 4.2.2
Intelligent Tutoring Systems
Intelligent Tutoring Systems (ITS) usually consist of the following core modules [39]: (1) an expert module which stores the domain knowledge, (2) a learner module which describes the learners’ knowledge about the domain and allows the tutor module to plan the interaction between the student and the system, (3) a tutor module which represents the tutoring strategies and learning goals, and (4) a graphical user interface. There are several classical ITS which have been developed for language learning [5, 11, 15, 19, 21, 23]. As an example we mention the RECALL system [21] in which the student is guided through communicative role-play scenarios. If the student makes a grammatical error, the system provides user
System Web.Pass.Voice [44] RECALL [21, 26] MTsystran [42] ISLE [24] FollowYou! [35] ArtCheck [34] L2tutor [32] Inst.Dict. [28] LISTEN [25] CAPIT [23] Spengels [4] ILTS [41] MT.system [36, 37] Fluent-2 [15] Jap.Tech.Texts [46] CoCoaJ [31] Subarashii [3] Nihongo-Cali [29] CompLex [18] FreeText [43] GLOSSER [30] ´ PILEFACE [22] McGill [13] Herr Kommisar [10] Lice [5] German Tutor [19] Word Manager [40] The Spanish Verb [38] Vocab.Tutor [7] Conversim T M [17] TLS/CATL [9] Targumatic [2] ELDIT [14] Tait [45] LingWorlds [11] Glossaries [6] Pronto [8] Fluency [12] MILT [20] Athena [27]
Lang. e e e e e e e e e e e e e e j j j j j f f f f d d d d s c a t h d,i d,s j,e i,e s,e f,e,g e,s,a *
AI systems and techniques NLP ITS,NLP MT A,ASR A NLP UM,NLP ASR,MT, ASR ITS EX,UM A,NLP,MT NLP,MT ITS,NLP,NLG NLP NLP ASR NLP EX,UM NLP NLP NLP UM,NLP UM,NLP,NLG ITS ITS,NLP NLP EX,NLG A ASR UM,NLP MT A A,ASR ITS,NLG NLP A,ASR ASR A,NLP,ASR NLP,NLG,ASR
R
Language skills W L S * *
T
* p * *
* *
f p
* * * * *
* *
* * * *
f,s
* * * * * *
*
s
* * *
f
* * * *
*
p,f,s
* *
*
* p * * * * *
* *
p p p,f p
Lang. elem. G V D d d * i c * t d * c * d d d c i c * i i * d t d i c * i c * i d i c d t * d i c t * * i c * * * *
Av. w w a
w
w
w
a w w
p w w
p
Table 1: Classification of ICALL systems. The letters in column “Lang.” are abbreviations for languages and are described in section 3. The meaning of the other letters is as follows: p = training pronunciation, f = training fluency, s = developing social skills; d = deductive approach, i = inductive approach; c = incidental acquisition, t = intentional acquisition, w = web based system, a = demo available, p = system purchasable; h = project homepage exists
specific feedback on different levels. Grammatical rules can be studied systematically and individually for each student. 4.2.3
User Modeling and Adaptivity
Many language learning systems include a so-called inspectable or viewable user model (UM) which records the user’s steps and mistakes. The mistakes are systematically classified by the system and can be examined by the student and the teacher after the learning session [4, 9, 10, 13, 18]. The system in [32] additionally uses NLP techniques to analyze writing errors in more detail. Traditional teachware is designed for a prototypical learner, which is not very useful in most learning situations. New media in combination with adaptive technologies (A) allow the development of teachware which based on the information in the user model adapts content and/or presentation to the individual learner. For example, remedial exercises are offered in [41, 24, 45] depending on the error frequency, adaptive sequencing is used in [7, 8, 20, 26, 35], and adaptive annotations are provided in [14]. The system FollowYou! [35] automatically generates a language lesson adapted to the specific needs of the learner. Since language input should be comprehensible and a little further beyond the learner’s current level of competence, the system checks the user model to determine the information which should be included in the generated lesson. 4.2.4
Natural Language Processing and Natural Language Generation
Natural language processing (NLP) and natural language generation (NLG) are among the earliest AI techniques which have been explored in computer-assisted language learning. Although these techniques have been applied since more than 20 years, only recently we have broadly conceived and working systems [16]. NLP and NLG are very promising for language learning. Grammar checkers in combination with a lexicon can be used to check written input from the user for spelling errors and grammatical correctness. Some grammar checkers apply constraint relaxation techniques in order to deal better with erroneous user input [10, 43]. Another method is to restrict the possible answers to a limited number of patterns and to apply simple pattern matching techniques, which considerably improves speed and reliability of the system [32]. Finally, some systems explicitly model an error grammar in order to provide meaningful feedback [21, 36, 44]. A full-fledged analysis of written text in all its complexity is a very difficult task, which exceeds current state of the art technology in NLP. In general, four different levels of complexity can be distinguished. At the first level only morphology is considered, which includes errors on the word level like gender, number, and conjugation [6, 40, 31, 30, 38]. The second level is about the syntax and includes errors on the sentence level like noun, verb, and prepositional agreement [41, 19, 21, 32, 34, 43, 44]. The third level includes semantics. It requires a large body of domain knowledge for a system to “understand” a conversation [9, 10, 13, 15, 20, 29, 27, 32, 46]. The fourth level is about pragmatics. Only very few projects deal with this delicate relationship between speaker and listener. The system in [22] contains a large body of knowledge about greeting situations and applies this knowledge to deal with pragmatic aspects of the language.
The NLP/NLG system of the Athena language learning project [27] is a very sophisticated system which includes morphological, syntactic, semantic, and pragmatic components. Several applications including the Fluent-2 program [15] incorporate this NLP system. In the meantime, commercially available courses for different languages have been developed using the Athena system. 4.2.5
Automated Speech Recognition
During the last couple of years automated speech recognition (ASR) technologies became mature enough to deliver reliable results. Since only a limited number of answers has to be expected from the learner, speech recognition in a CALL context is quite fast and reliable. The major drawback is the big difference between native and non-native accents, which makes it difficult for the software to provide a fine grained analysis. Two ASR technologies have to be distinguished: discrete and continuous speech recognition. Discrete speech recognition allows to analyze single patterns which are known to the system. This technique is often used to train pronunciation [8, 12, 27, 28]. It is also used for training fluency, where the user can choose from a predefined set of patterns [17, 20]. Continuous speech recognition aims at analyzing free and fluently spoken input [3, 45]. While accurate recognition of spontaneous speech is still beyond the state of the art [1], it works to some extent if the system can expect a certain input [24, 25]. The system Conversim T M [17] for example allows users to have extensive ”face to face” dialogues in real time with virtual characters. The system prompts continuously three relevant questions which the user can ask. A video instructor helps to pronounce correctly the questions and phrases. The user must rely on language skills, experience, and intuition to evaluate whether the character lies or not. 4.2.6
Machine Translation
While translating is a commonly used method in traditional language learning, machine translation (MT) technologies entered only a few years ago into the field of computer-assisted language learning. Translations provided by a system can be used as a preliminary version of a text in the target language, on which the student should continue to work [2, 42]. Instead of using other MT systems, some researchers developed ICALL systems with MT like capabilities which are easier to integrate into the learning process [41, 36, 37]. The ILTS [41] system shows to the learner texts with typical key English patterns, which have to be translated by the learner. The system compares these translations with correct and erroneous model sentences and provides error comments and adaptive remediation. A quite unusual approach has been used in a study which aims at investigating whether a handheld translation machine used in daily live could be helpful for language learning [28]. An evaluation showed improved pronunciation and increased vocabulary of the students.
4.3
Language Skills
The third dimension classifies ICALL systems according to different language skills which can be trained. We will distinguish between reading (R), writing (W), listening (L), speaking (S), and translating (T) skills.
A large number of systems have been developed for training reading skills. For example, the system in [25] uses continuous speech recognition to listen to children while they are reading aloud in their native language. The effects of ASR errors are minimized by never saying that the student is right or wrong. Instead, the system responds to a possibly incorrect word by communicating the correct word or by indicating ”mmm?”. Effort is praised instead of performance. Glossary systems provide help for unknown words in text units. Single words or collocations are linked to additional pieces of information in an knowledge base, lexicon, or text corpus [6, 40, 30]. Some systems combine reading with vocabulary and/or grammar acquisition [7, 13, 14, 18, 42, 35]. In other systems the training of reading skills is embedded in conversations [10, 15, 17, 20, 27, 32]. Most of the analyzed systems are focusing on training writing skills. Many systems are limited to single grammatical aspects such as spelling, punctuation, or writing sentences in passive form [4, 23, 34, 38, 44]. Other systems are more general and include a sophisticated parser. The user can freely type in sentences and gets detailed feedback, e.g. about grammatical aspects or collocations [2, 41, 31, 19, 21, 36, 37]. A third class of systems concentrate on communication skills and the content of the learner’s answers [9, 22, 27, 29, 43], in some cases embedded in stimulating conversations [10, 15, 20, 32]. More sophisticated systems provide help for the overall text composition process including the structuring of texts according to different schemas [5, 46]. The system LICE [5] is an example, which provides help during all stages of the writing process, from the initial elaboration of ideas to the generation and revision of the text. Only a few systems have been developed for training listening skills including comprehension abilities. Understanding spoken language is a crucial aspect for successfully using a second language. One way to implement language comprehension is by applying automated speech recognition techniques [8]. Another approach has been implemented in the system LingWorlds [11]. A microworld is presented and the user is asked to move objects in this world (e.g. provisioning a lifeboat before an ocean liner sinks). The reactions of the user are analyzed by the system. If the user misunderstands a command, the relevant part of the sentence is repeated. Pragmatic aspects are as well very crucial for language comprehension [3, 17]. In the analysis of speaking skills we distinguish between pronunciation, fluency, and social abilities. Most of the systems are made for training pronunciation. The systems differ in the way they give feedback to the learner: transforming the utterances into visual cues [12, 17, 27], commenting orally on the user’s performance [12, 45], highlighting the actual position in the orthographic representation of the word [24], and rejecting answers if they are not understood by the system [8, 20, 28]. Training fluency requires very efficient speech analyzers. While the approach with a set of prefabricated answers is faster and more reliable [17, 20], the recognition of free answers is also possible [3]. The systems in [10, 32] approach training fluency via written tasks. Training social skills is realized by teaching cultural competence and controlling the vocabulary used in specific situations [3, 17, 22]. The Subarashii system [3] offers to beginners of Japanese the opportunity to solve simple problems through spoken dialogues with Japanese native speakers. The vocabulary needed can be learned by hearing and repeating in stand-alone mode. In addition to pronunciation and fluency also cultural competence like tactfulness is trained, e.g. the student politely has to refuse an invitation. Training translation skills can naturally be integrated into CALL systems by including a
machine translation component. The translation system is either used as supportive tool [42] or as a starting point. The student can analyze the output of the system and try to understand and eliminate the translation errors and to refine the text [2]. Instead of using a translation system, the approaches in [41, 36, 37] developed components which are specifically designed to support the training of translation skills. The system in [37] is addressed to help student translators who are simultaneously learning English as a foreign language and translating from and into that language. A generator for lexicalized sentence stems (i.e. typical phrases a native speaker is using) and a collocation matcher should be included into a writing system to help students to generate more easily a native like text in the target language.
4.4
Language Elements
For the purpose of this analysis we distinguish the following three language elements: grammar (G), vocabulary (V), and dialogue elements (D). The grammar has been considered as a core part in most language learning curricula. Inductive grammar acquisition, where the students work out the grammar rules by themselves, has been approached in several systems in different ways focusing on reading [6, 13, 40, 30], writing [5, 31, 46], and translation tasks [2, 36, 37, 42]. In deductive grammar acquisition the students first learn the grammar rules and then practice these rules on examples. Systems which support this approach have to perform a grammatical analysis of the student’s input. As the development of an efficient and reliable grammar checker for correct and wrong input is very difficult, many systems are limited to single aspects, e.g. usage of articles [34], conjugation [38], punctuation and capitalization [23], gender, number and person [43], passive sentences [44], and morphological aspects [4]. Other systems include a sentence parser combined with an error grammar and are able to detect an certain class of grammatical errors [41, 9, 19, 21, 29]. For example, in the German Tutor [19] the students can build sentences with some indicated words. Several grammar and parser modules analyze the input (spelling, word usage, grammar, punctuation). If a module detects an error, further processing is blocked until the student corrects the mistake. This approach has also been implemented for the Greek and English languages. There are several systems which apply intelligent techniques for vocabulary acquisition. The systems in [7, 14, 18, 35] for intentional vocabulary acquisition apply an adaptive approach and consider the previously learned vocabulary. For example, in [14] a tutor guides the learner through a contextualized vocabulary acquisition process, proposing to the learner individual groups of new words to be learned next as well as text units to train the new words. Most systems pursue incidental vocabulary acquisition in combination with translating [2, 41, 28, 36, 37, 42] or reading [6, 13, 40, 30]. Knowledge bases or NLP techniques are applied in order to provide extensive information for unknown words. Many systems do not explicitly address grammar or vocabulary acquisition, but allow to train dialogue elements like pragmatics [22], phonology [8, 12, 24, 45], or an entire conversation [3, 17]. Several systems focus on the training of comprehension and fluency using written dialogues [10, 11, 21, 15, 20, 27, 32]. Dialogue systems usually provide learning environments like a microworld or a virtual interlocutor. The user typically has to interview the characters or to negotiate a goal. The systems include a grammar and robust parsing abilities to analyze errors up to the semantic level and to give meaningful feedback.
5
Discussion
Artificial Intelligence offers many possibilities to improve computer-assisted language learning systems. However, the application of these technologies is not mature yet and still requires more research. Many interesting and promising systems remained in a prototypical stage. They were abandoned after project termination and were never used in a real language learning environment — often because of lack of funding. Most of the CALL systems using NLP techniques just concentrate on syntax, few of them include semantic components, even less try to address the problem of pragmatics. The systematic training of cultural knowledge and social abilities is considered only rarely in the training process. Sometimes technology is used quite experimentally. There are several systems which apply NLP techniques in order to enhance conversational fluency. But can oral capabilities be learned by written tasks? Machine translation has been used only rarely for language learning. The MT approach is basically to learn from mistakes, which could be valuable if the mistakes are false friends and commonly made mistakes by students. It is, however, questionable whether a student profits from artificial mistakes made by a machine? Surprisingly few systems have been developed for training listening skills, even if understanding native speakers is very important for successfully applying a foreign language. Intelligent features such as the recognition of eye movements or screen touches are relatively unexplored in CALL. Such features could be interesting for teaching silent reading or for comprehension abilities. While some systems are rather promising, additional research efforts are required in order to tackle the above mentioned problems and to develop authentic learning systems. The development of integrated systems and the wise integration of technologies is certainly very important. Another issue is to analyze and evaluate which technology best supports which language skills. More research should also be done in improving the performance of existing systems and technologies. Intelligent technologies are proposed to be very useful to support authentic language learning environments and to simulate real communication situations. On the other hand, a certain neglect of ICALL is lamented, even within the research community. The reason may be that, because of the difficult implementation process, many systems just focus on single aspects of language learning and are heavily technology driven. The development of integrated systems and their wise incorporation into a CALL environment is certainly very important for the acceptance and applicability of ICALL systems in the real language learning lab.
6
Conclusion
In this paper we provide a survey about the state of the art in intelligent computer-assisted language learning. We analyzed 40 ICALL systems and classified them along five dimensions: languages, AI techniques, language skills, language elements, and availability. The great potential of new media and technologies for language learning has been recognized by several researcher and has been explored in a number of systems. Nevertheless, there are still a lot of open problems. Many systems focus on single aspects of language learning. What is needed is a more integrated and comprehensive approach which supports also semantics, pragmatics,
cultural knowledge and social abilities, using different technologies which are tailored for the training of specific skills.
References [1] Gregory Aist. Speech recognition in computer-assisted language learning. In CALL Media, Design, & Applications, pages 165–181. Swets & Zeitlinger, 1999. [2] Don Anderson. Machine translation as a tool in second language learning. CALICO Journal, 13(1), 1995. [3] Jared Bernstein, Amir Najmi, and Farzad Ehsani. Subarashii: Encounters in Japanese spoken language education. CALICO Journal, 16(3), 1999. [4] Edwin Bos and Joke van de Plassche. A knowledge-based, English verb-form tutor. IJAIED, 5(1), 1994. [5] Chris Bowerman. An intelligent language learning system for writing in German. In Proceedings of EW-ED’94, 1994. Demoversion available at http://www.cet.sunderland.ac.uk/cbowww/SOFTWARE/LICE/. [6] Luigi Colazzo and Marco Costantino. IJAIED, 9(1–2), 1998.
Multi-user hypertextual didactic glossaries.
[7] Roger E. Cooley. Vocabulary acquisition software: User preferences and tutorial guidance. In Proceedings of the CALL-Workshop held in conjunction with AIED’01, 2001. [8] Jonathan Dalby and Diane Kewley-Port. Explicit pronunciation training using automatic speech recognition technology. CALICO Journal, 16(3), 1999. [9] Suyada Dansuwan, Kikuko Nishina, Kanji Akahori, and Yasutaka Shimizu. Development and evaluation of a Thai learning system on the web using natural language processing. CALICO Journal, 19(1), 2001. Web based version CATL is not yet online. [10] William H. DeSmedt. Herr Kommissar: An ICALL conversation simulator for intermediate German. In Intelligent Language Tutors - Theory Shaping Technology. Lawrence Erlbaum Associates, 1995. [11] Sarah A. Douglas. LingWorlds: An intelligent object-oriented environment for second language tutoring. In Intelligent Language Tutors - Theory Shaping Technology. Lawrence Erlbaum Associates, 1995. [12] Maxime Eskenazi. Using automatic speech processing for foreign language pronunciation tutoring: Some issues and a prototype. LL&T, 2(2), 1999. Project NativeAccent at company Web-site http://www.carnegiespeech.com. [13] Carl H. Frederiksen, Janet Donin, and Michel D´ ecary. A discourse processing approach to computer-assisted language learning. In Intelligent Language Tutors - Theory Shaping Technology. Lawrence Erlbaum Associates, 1995.
[14] Johann Gamper and Judith Knapp. Adaptation in a Language Learning System. In Proceedings of ABIS-Workshop held in conjunction with LLWA’01, 2001. Preliminary version at http://www.eurac.edu/Eldit. [15] Henry Hamburger. Tutorial tools for language learning by two-medium dialogue. In Intelligent Language Tutors - Theory Shaping Technology. Lawrence Erlbaum Associates, 1995. [16] Henry Hamburger, Michael Schoelles, and Florence Reeder. More intelligent CALL. In CALL - Media, Design & Applications. Swets & Zeitlinger, 1999. [17] William G. Harless, Marcia A. Zier, and Robert C. Duncan. Virtual dialogues with native speakers: The evaluation of an interactive multimedia method. CALICO Journal, 16(3), 1999. System purchasable at http://idrama.com/LanguagePrograms.htm. [18] Michael Harrington. Complex: A tool for the development of L2 vocabulary knowledge. IJAIED, 5(4), 1994. [19] Trude Heift and Devlan Nicholson. Theoretical and practical considerations for Webbased intelligent language tutoring systems. In Proceedings of ITS’00, 2000. System(s) accessible at http://www.sfu.ca/language-lab/trude/software.html. [20] Melissa Holland, Jonathan Kaplan, and Mark Sabol. Preliminary tests of language learning in a speech-interactive graphics microworld. CALICO Journal, 16(3), 1999. Project homepage at http://www-ari.army.mil/miltfl95.htm. [21] Anja Kr¨ uger and Simon Hamilton. RECALL: individual language tutoring through intelligent error diagnosis. ReCALL, 9(2), 1997. Project homepage at http://www.infj.ulst.ac.uk/˜recall. [22] Ruddy Lelouche. Dealing with pragmatic and implicit information in an ICALL system: ´ The PILEFACE example. IJAIED, 5(4), 1994. [23] Michael Mayo and Antonia Mitrovic. Optimising ITS behaviour with bayesian networks and decision theory. IJAIED, 12(3), 2001. Project homepage at http://www.cosc.canterbury.ac.nz/˜tanja/capit.html. [24] Wolfgang Menzel, Daniel Herron, Rachel Morton, Dario Pezzotta, Patrizia Bonaventura, and Peter Howarth. Interactive pronunciation training. ReCALL, 13(1), 2001. Project homepage with demo at http://nats-www.informatik.uni-hamburg.de/˜isle. [25] Jack Mostow and Gregory Aist. Giving help and praise in a reading tutor with imperfect listening - because automated speech recognition means never being able to say you’re certain. CALICO Journal, 16(3), 1999. Project homepage at http://www.cs.cmu.edu/˜listen. [26] Maureen Murphy and Michael McTear. Learner modeling for intelligent CALL. In Proceedings of UM97, 1997.
[27] Janet H. Murray. Lessons learned from the athena language learning project. In Intelligent Language Tutors - Theory Shaping Technology. Lawrence Erlbaum Associates, 1995. Implemented systems purchasable at http://web.mit.edu/jhmurray/www/Project Sampler.html. [28] Marie J. Myers. Voice recognition software and a hand-held translation machine for second-language learning. CALL, 13(1), 2000. Experiment was a part of the project described in http://www.socsci.mcmaster.ca/srnet/exsum.htm. [29] Noriko Nagata. An effective application of natural language processing in second language instruction. CALICO Journal, 13(1), 1993. [30] John Nerbonne, Duco Dokter, and Petra Smit. Morphological processing and computer-assisted language learning. CALL, 11(5), 1998. Web based version at http://odur.let.rug.nl/˜glosser/Glosser. [31] Hiroaki Ogata, Yoshiaki Hada, and Yoneo Yano. Computer supported online correction for collaborative writing. In Proceedings of IS’00, 2000. [32] Charlotte Price, Gordon McCalla, and Andrea Bunt. L2tutor: A mixed-initiative dialogue system for improving fluency. CALL, 12(2), 1999. Project homepage at http://www.cs.usask.ca/research/research groups/aries/tutoring-projects.html. [33] Marikka Rypa and Ken Feuerman. Calle: An exploratory environment for foreign language learning. In Intelligent Language Tutors - Theory Shaping Technology. Lawrence Erlbaum Associates, 1995. [34] Sue Sentance. A rule network for English article usage within an intelligent language tutoring system. CALL, 10(2), 1997. [35] Chi-Chiang Shei. FollowYou! an automatic language lesson generation system. CALL, 14(2), 2001. [36] Chi-Chiang Shei and Helen Pain. An ESL writer’s collocational aid. CALL, 13(2), 2000. [37] Chi-Chiang Shei and Helen Pain. Learning a foreign language through machine translation: Focusing on sentence stems and collocations. In Proceedings of the CALLWorkshop held in conjunction with AIED’01, 2001. [38] Jes´ us Soria. Expert CALL: data-based versus knowledge-based interaction and feedback. ReCALL, 9(2), 1997. [39] Merryanna L. Swartz and Masoud Yazdani, editors. Intelligent Tutoring systems for Foreign Language Learning, chapter Introduction. Springer-Verlag, 1992. [40] Pius ten Hacken and Cornelia Tschichold. Word manager and CALL: Structured access to the lexicon as a tool for enriching learners’ vocabulary. ReCALL, 13(1), 2001. Some components accessible at http://www.canoo.com/wmtrans/home/demos.html. [41] Naoyuki Tokuda and Liang Chen. An online tutoring system for language translation. IEEE Multimedia, 8(3), 2001. Preliminary version at http://azalea.sunflare.co.jp.
[42] Mar´ıa Dolores La Torre. A web-based resource to improve translation skills. ReCALL, 11(3), 1999. Resource accessible at http://www.hum.port.ac.uk/slas/babel/INDEX˜1.HTM. [43] Anne Vandeventer. Creating a grammar checker for CALL by constraint relaxation: A feasibility study. ReCALL, 13(1), 2001. Project homepage at http://www.latl.unige.ch/freetext/index.html. [44] Maria Virvou and Victoria Tsiriga. Web passive voice tutor: An intelligent computer assisted language learning system over the WWW. In Proceedings of ICALT’01. IEEE Computer Society Press, 2001. System is not yet online. [45] Richard C. Waters. The audio interactive tutor. Technical report, Mitsubishi Electric Research Laboratories Cambridge Research Center, 1994. Technical report available from http://www.merl.com/reports/TR94-04/. [46] Jie Chi Yang and Kanji Akahori. A discourse structure analysis of technical japanese texts and its implementation on the WWW. CALL, 13(2), 2000.
