LINCOM Studies in Second Language Teaching

LINCOM Studies in Second Language Teaching

full text research abstracts of all titles monthly updates

LINCOM webshop www.lincom-europa.com

Artificial Intelligence and Error Correction in Second and Foreign Language Pedagogy

Kerwin Anthony Livingstone

2012 LINCOM EUROPA

Published by LINCOM GmbH 2012.

LINCOM GmbH Gmunder Str. 35 D-81379 Muenchen [email protected] http://home.t-online.de/home/LINCOM.EUROPA www.lincom-europa.com webshop: lincom.eu All rights reserved, including the rights of translation into any foreign language. No part of this book may be reproduced in any way without the permission of the publisher. Printed in E.C. Printed on chlorine-free paper Die Deutsche Bibliothek - CIP Cataloguing-in-Publication-Data A catalogue record for this publication is available from Die Deutsche Bibliothek (http://www.ddb.de)

TABLE OF CONTENTS

Items

Page Number

Abstract

2

Introduction

2-3

Historic Framework of CALL

3-6

Computers and Language Error Correction

7

Error Correction

7-10

Computers and Error Correction

10-13

Written Errors

13-16

Speech Processing

16-20

Speech Recognition in CALL

20-23

Some examples of CALL interactive voice systems

23-25

Conclusion

25-27

References

27-33

1

Artificial Intelligence and Error Correction in Second and Foreign Language Pedagogy Kerwin A. Livingstone Applied Linguist [email protected]

Abstract This paper is an attempt to ascertain the suitability of computers for second and foreign language (SL/FL) error correction, especially those made by SL/FL learners. For this purpose, the handling of such errors proposed in Second Language Acquisition (SLA) literature will be examined. Subsequently the technologies capable of evaluating student output and identifying and correcting Non-Native Speaker (NNS) errors will be examined. Though it will be made quite clear that the computer cannot substitute a human being in total language processing, some strengths of artificial intelligence in partial language processing will be pointed out and their suitability for L2 error correction will be highlighted. The article will conclude emphasizing that the use of high quality multimedia applications and programmes stimulates and fosters language learning.

Key words: computer, language errors, error correction, second language (L2), Computer Assisted Language Learning (CALL), Intelligent CALL (ICALL).

Introduction L2 learners around the world seem to be doomed to make mistakes which clearly label them as non-native speakers (NNS) in their attempts to acquire, in addition to their mother tongue, another language. The nature of their expression is sometimes treated with patience and understanding by native speakers of the target language, while other times, patience and understanding seem to diminish. Van Lier (1996) notes that sometimes intolerance prevails towards NNS statements, and, therefore, native speakers put the onus on the NNS to improve their expressions, so that it can be produced at the level of the standard language. This can lead to frustration for both 2

sides, especially if the NNS cannot produce the correct language that is expected. This can affect the motivation of the NNS. Moreover, many language learners throughout the world expose themselves to such risks. They do this by moving temporarily or permanently to another country, often with the goal of completing their college education. In the humanities and social sciences, language is the crucial factor that influences academic success, often being disadvantageous to the very non-native speakers. Artificial Intelligence (AI), on the other hand, can deal with new problems, once the general principles and techniques for working with computer applications for language learning have been learnt. For example, to recognise the erroneous sentence of a student, a computer programme should have the same correct and incorrect forms. In this way, and so that the computer application can provide the student with feedback about why the sentence has errors, the system should also include a record of wrong rules. For an identical syntax error but with a different vocabulary, this application again should have the exact expression pre-stored in its memory. However, an intelligent program would only have to have a rule that the student uses for such an erroneous production. The programme theoretically could recognise the same type of error in any context and any vocabulary. Thus, given the power of AI techniques to correct the user at the same time that he is committing the error, such applications are considered as the best ally of the teacher in correcting L2 errors. By taking into account the above, this paper aims to highlight the importance of computer and technological applications and programmes to correct language errors, as it relates to Second Language Acquisition (SLA). Thus, this paper outlines the main criteria to consider when they intend to introduce computer applications in teaching and learning and language acquisition and automatic processing of natural language.

Historic Framework of CALL CALL (Computer-Assisted Language Learning), or Assisted Language Learning Computer is a type of educational technology designed to serve as a learning tool. In simple terms, it refers to the use of computer applications in teaching and learning languages. For many years, foreign language teaching has traditionally been limited to opportunities created by the teacher in the classroom, getting and extracting information from text books, tapes, recordings, among others. In the 90's, alternative ways of teaching languages were developed such as distance learning and self-directed learning through the 3

Internet, with two purposes: first, save resources and, secondly, reduce geographic barriers and time constraints when taking a foreign language course. Indeed, the use of computers in language teaching and learning has generated major changes in how second language are taught and learnt- better learning in less time, more lasting learning and improving communicative competence. (Chapelle, 2001 & 2003; Levy, 1997; Warschauer, 2000). The use of computational tools has become a new medium which shapes the processes and products of communication. Because multimedia technology has opened new opportunities for communication between teachers and learners, and among those who speak a second language, many language teachers have realised the enormous potential for teaching computer-mediated learning (Levy, 1998; Warschauer & Healy, 1998; Warschauer & Kern, 2000). Since the 60's, language teachers have witnessed dramatic changes corresponding to the teaching-learning of languages. This approach has been extended from the teaching of discreet grammatical structures to the development and improvement of the communicative ability. Computers had begun to be used by universities, especially in the United States. Its use became an integral part in educational training of university students in some careers. Soon, multimedia technology has started to be used experimentally in other levels of education (Garrett, 1987; Heift & Schulze, 2003; Levin & Evans, 1995). The first phase of CALL, conceived in the 50's and implemented in the sixties and seventies, was based on the then BEHAVIOURIST THEORY which stressed on repetitive language exercises (Chapelle, 2001 & 2003; Levy, 1997; Warschauer, 2000; Taylor 1980). This stage is based on the model of the computer as tutor and its arguments were that: (1) the computer was ideal for carrying out repeated exercises because the machine, unlike a teacher, it can do reps with the same linguistic material without getting tired or making mistakes, as well as providing immediate feedback. (2) the computer could present such material on an individual basis, allowing students to work at their own pace and paving the way to success. In summary, the BEHAVIOURIST THEORY focused on the development of language skills. It focused its attention primarily on grammar or focus-on-form exercises. In the seventies and eighties BEHAVIOURIST CALL was undermined by two important factors. First, because behaviourist approaches to language learning had been rejected in

4

theory and pedagogy. Second, because, with the advent of the microcomputer, the way was paved for a new range of possibilities and a new stage of CALL. The new phase of CALL was based on the COMMUNICATIVE THEORY of teaching which became prominent in the 70 and 80 (Warschauer, 1996). Followers of this theory argued that the exercise and practice programmes of the previous decade did not provide enough value to authentic communication. Creativity in expressing oneself was valued above memorization. The negotiation of meaning had become more powerful and important. Comprehension had become a more fundamental domain in education. One of the main representatives of this new theory was Underwood (1984), who proposed a set of principles for COMMUNICATIVE CALL : (1) it focuses on the use of the forms themselves; (2) it teaches grammar implicitly rather than explicitly; (3) it allows and gives encouragement to learners to generate original statements rather than manipulating the fabricated language; (4) it does not judge or evaluate everything that the students nor does it give them congratulatory messages, lights or bells, and (5) it prevents saying to the students that they are not right and is flexible to a variety of responses from the students. In the late 80's, it was felt that CALL still left much to be desired and that it still was not able to extract the full potential of computers. Stevens (1989) was one of the critics who said the computer was used in an ad hoc and disconnected way and therefore, it was necessary to make a greater contribution to the marginal rather than to the central elements of the process of teaching languages. The challenge for supporters of CALL was to develop models that would integrate the various aspects of the language learning process. Fortunately, advances in computer technology were providing opportunities to do that. The INTERACTIVE and INTEGRATIVE THEORY of CALL, developed by Warschauer (1996), Pennington (1989) and Garrett (1991), is based on two technological advances of the past decade: the computers with multimedia technology and the Internet. This allows learners to browse at their own pace, simply by clicking, using the mouse. This type of INTEGRATIVE THEORY

of CALL generated a large number of advantages for language learning.

CALL in the past decade stopped being a mere phenomenon in life and transformed itself into an indispensable tool for teaching modern languages. Along with other technological advances, such as video, the number of students who participate in the 5

experience of CALL continues to increase speedily. Computer Mediated Communication (CMC), which has existed primitively since the 60 (Warschauer, 2000), spread only twenty years ago and today is probably one of the computer applications with a profound impact on the teaching- learning of languages. For the first time, students of modern languages can communicate directly and conveniently with other learners or speakers of the target language all day, from their school, work or home. CMC allows students to share not only brief messages, but also long documents which has therefore given way to and promoted COLLABORATIVE WRITING1. Thanks to CMC, students can actively participate in the search for authentic material from millions of electronic files, share graphs and tables, sounds and video, access Web pages. Students, within a short time, locate, and access various materials (newspaper and magazine articles, video clips, reviews of movies, lots of books, among others), taking, in this way, control of their learning. They can also use the web to publish articles or multimedia materials, with the aim of sharing with colleagues or the public. In other words, CALL suggests and affirms that the computer has a variety of uses for language teaching: it can be a tutor, offering practical skills; it can be a stimulus or catalyst for discussion and interaction, or as a tool for writing and research. Garrett (1987:70) highlights the fact that “the use of computers is not a method but is an environment in which we can implement a variety of methods, approaches and educational philosophies”. In other words, the effectiveness of CALL is not in the medium itself, but how it is used in the process of language teaching and learning. This assertion by Garrett has much validity since the most important potential of the computer lies in its ability to provide an environment for language learning in which students are supported individually to develop, expand and refine their own language and communication skills in a new language. Computer Assisted Language Learning must be the focus of our efforts and without doubt, the development of their potential will significantly affect the way in which languages are taught and learnt in this century. Teaching in this century has been combined with the use of assistive technological resources within which the computer has played a leading role by the benefits it incorporates, 1

This is a group of persons (or communities) who, using online communication (Internet) and by means of set of software tools (Blogs, Word Press, Wiki, etc) make individual contributions to create a specific document.

6

both for explaining concepts and for its appropriateness. As technology continues to advance, methods that are effective for the educational process have been sought.

Computers and Language Error Correction Second language (L2) errors and possible ways to address them have been treated by different theories of L2 acquisition, having radically different views of the meaning of the language errors produced by L2 learners. For example, the BEHAVIOURIST THEORY sees mistakes as a way of getting the wrong item and therefore should be avoided. Instead, interlanguage (IL) researchers (Corder, 1967, James, 1998; Muñoz, 1991) see errors as idiosyncrasies in the learner's L2 system, and therefore are not errors according to the learner's interlanguage, only for the target language, which is not exactly what the learner produces. Error Correction The error correction concept was proposed by Corder (1967), who was instrumental in the short-term revival of error analysis before subscribing to 'idiosyncratic dialect'. He said that mistakes were evidence of the learner's internal syllabus and imminent difference between the input (which is being taught) and output (what is being learnt). According to Corder (1967), errors affect the learner's L2 competence. At the same time, he says that mistakes can be corrected and therefore also reflect student performance. These two concepts are based on Chomsky's theory. Error analysis is important for three reasons: (i) it informs the teacher about what should be taught, (ii) it informs the researcher on the learning course, and (iii) they are a result of hypothesis testing of the learner’s L2 (James, 1998:12). It is considered that the sources of errors are the redundancy of the code (intralanguage), several sources of interference (interlanguage) and inadequate presentation. Grammatical awareness is an appropriate response to L2 errors that will help the student to pay attention to the linguistic structures of the target language and give him time to reflect on them. This trend is evident in linguistic awareness as well as in focus on form. The latter refers to a change of focus occasional on linguistic form in a lesson focused on meaning. An example of this is a paraphrase of an erroneous sentence spoken by the student, which follows the same student's 7

intended meaning. Long and Robinson (1998) report a high effectiveness of this procedure error analysis - especially with adult learners. Allan (1999) stipulates that linguistic awareness has a strong emphasis on inductive learning with the objective of going beyond the grammar in the traditional sense. He defines the linguistic awareness and sensitivity of an individual and the nature of language and its role in human life; meanwhile, along with Ellis (1997) and James (1998), he makes the distinction between linguistic awareness and grammatical awareness. While grammatical awareness is responsible for focusing on what the student does not know, linguistic awareness

is

concerned

to

make

explicit

what

is

already

known

implicitly

(James, 1998). Therefore, correcting errors (Corder, 1967) would concur with grammar awareness, and correcting faults with linguistic/language awareness. Moreover, the Chomskyan concept of native speaker COMPETENCE has led to the INTERACTION HYPOTHESIS,

a term used in reference to non-native speaker competence in

contrast to native speaker's competence. Chomsky (1965) asserts that there may be strong or weak equivalence between two grammars. The weak equivalence allows both grammars to produce the same types of sentences. The strong equivalence, however, allows these types of sentences to have exactly the same meaning. The question of whether the grammars of native and non-native speaker may be weak or strong equivalent has not been resolved (James, 1998: 54). This, however, has opened the door to doubt that a non-native speaker can ever fully master an L2, or, in other words, to reach ultimate achievement, which has also been shared by some researchers. The eminently nativist conception of Chomsky, in fact, allowed little influence of linguistic evidence in language learning, especially in first language (L1) learning. Krashen (1987), as well as the other behaviorists, believes in the power of positive evidence in language learning. Positive evidence here means exposure to well-formed expressions. Gregg (2001) makes use of equivalent lines of distinction between positive evidence and the use of language on the one hand, and negative evidence and the words of the language; on the other hand, asserting as well that the use of meta-linguistic evidence is negative, if this confirms the hypothesis of the learner. Gregg (2001) nonetheless concedes that the negative evidence in L2 acquisition most commonly means "to be alert to someone’s linguistic errors." This can happen in many 8

different ways and in varying degrees. Some followers of the INTERACTION HYPOTHESIS, for example, believe in paraphrase as a remedy (Long & Robinson, 1998). Mitchell & Myles (1998), on the other hand, argue that despite the valuable negative evidence that it offers, paraphrasing does not require students to correct themselves. Apart from paraphrasing, research and theory on the types of correctness and effectiveness include the declaration of the relevant linguistic rules, the error rate indication without paraphrasing, the mere highlighting and counting the number of errors per line, although comparison between these methods now appear inconclusive (James, 1998). Learners' preferences for certain types of correction also seem to vary, although it is not certain that the preferred correction method is most useful. It is, however, quite clear that learners want to be corrected. Individual differences of learners also seem important in deciding how to execute the repair or correction of errors, but there is no general agreement among researchers in this case. However, there is a body of research evidence supporting the hypothesis that error correction is beneficial and necessary, and can lead to learning (Schulze, 2003; Gregg, 2001). Most of the theoretical research of L2 acquisition, however, agrees that AWARENESS is the crucial event in the correction of language errors and in learning. For James (1998), this supports grammatical awareness that is compared with the explanation of the unknown with what Krashen (1987) and Ellis (1997) designate as 'learning' (explicit or conscious) or the type of learning that is responsible for accuracy. Practice, moreover, which had been favoured both by the audio-lingual method, as well as the communicative language learning method, requires participatory attention. Therefore, it is thought that it can lead to acquisition, or unconscious learning (implicit), and fluency, which was highly valued by the audio-lingual approach. In fact, it is in high regard by the proponents of communicative language learning. Therefore, NOTICING or AWARENESS of errors invites a cognitive comparison (Ellis, 1985) between the interlanguage and the target language. As for Doughty (2001), this is a cognitive intrusion designed to allow planning between a conceptual and a new linguistic form under the influence of new pragmatic, semantic, syntactic and phonological information. James (1998) identifies this comparison as a form of error analysis, a procedure normally associated with the activity of the researcher or teacher, not the student.

9

While advocates of the INTERACTION HYPOTHESIS seem to use the terms CONSCIOUSNESS, ATTENTION,

and AWARENESS interchangeably, James (1998) and Ellis

(1997) make a distinction between them. According to James (1998), L1 awareness is another element of success in L2 learning. Linguistic awareness is caused by explanation, that is, something the student already knows implicitly, while AWARENESS can be achieved from something previously unknown to the student. James (1998) believes that a coordinated approach to L1 CONSCIOUSNESS and L2 AWARENESS may lead to a better understanding of the L2 in terms of its parallels with the L1. Understanding how language functions in general seems to be the goal of this activity. The development of the INTERACTION HYPOTHESIS is closely related to the OPERATING PRINCIPLES of Slobin

(Gregg, 2001; Doughty, 2001), which refer to how students

perceive, store and organize information about language. These processes are supposed to lead to positive transfer, an SLA theory which seems to complement the theory of contrastive analysis (James, 1998). Grammatical awareness, according to James (1998), serves its purpose when it gives the student the relevant rule in simple language. This allows the perception of the structure and understanding of the grammatical rule governing the production of linguistic structures. The review of major points of view relating to errors and correction has made us aware of a multitude of psychological, philosophical and theoretical reinforcements towards research and practice of error handling. The analysis of some key terms used in this context by both the L2 acquisition and CALL could help put several proposals discussed within a wider framework of the history of human ideas. Computers and Error Correction Two trends are made obvious from the above discussion: first, the inconsistency of automatic test evaluators to assess the writing of native speakers on the one hand, and the writing of non-native speaker on the other hand, and second, the inability evident from the parsers designed to support the writing of native speakers to deal with non-native speaker or L2 language errors. A number of CALL authors signal the latter as the main problem of probabilistic parsers, as those found in machine translation and word processing programmes (Tschichold, 1999, James, 1998). Tschichold (2003), in particular, identifies the lack of semantic, 10

pragmatic and contrastive linguistic knowledge in such parsers as the root of its failure in helping L2, knowledge on which a foreign language teacher can rely. The reason why the latest version of grammar check programmes of word processors has not found, for example, that the word "score" should be a noun instead of a verb when used more than once in a sentence, is because it does not look at the total sentence. Most likely, it concentrates on two or three adjacent words at a time and calculates the statistical probability for simultaneous co-occurrence in a text. This kind of parser is called a probabilistic parser (Smith, 1991). Liou (1991) highlights that feedback can be misleading because, as shown in some evidence, students tend to rely too much on computers (Holland, Maisano, Alderks & Martin, 1993). Therefore, Intelligent CALL (ICALL) searches for other ways of dealing with errors of non-native speakers (Tomlin, 1995; Ferreira, 2006 & 2007). No doubt that one of the most researched themes in the area of Intelligent Tutoring Systems (ITS) has been the identification and implementation of feedback strategies that facilitate student learning (Ferreira, 2006 & 2007). Ferreira (2006) conducted an empirical study based on effective feedback strategies for the teaching of languages in e-learning contexts. Much of this research had been directed to dealing with procedural skills’ teaching systems in areas such as algebra, physics or computer programming, etc. However, there has been little emphasis on studies and research on such strategies in language teaching (ITS for foreign languages). This paper reported on the design of effective strategies for corrective feedback ITS in foreign languages. Empirical evidence was explored concerning the effectiveness of feedback strategies in a study based on the experimental design - pre-test/post-test and control group - in which students interact with an e-learning application. The objective was to provide effective guidelines for researchers who develop feedback strategies for ITS for foreign language learning. Two groups of corrective feedback strategies were investigated: Group 1, which included the repetition of error and explicit correction, and Group 2 considered metalinguistic keys and elicitations from the response of the student (without giving the response) (Ferreira, 2006). Ferreira (2006) reveals that, in general, the results showed that the strategies of Group 2 (metalinguistic clues and elicitations) supported the teaching-learning process of the 11

subjunctive in Spanish more effectively than the strategies of Group 1 (repetition and explicit error correction). After three weeks of the treatment process, the strategies, attempting to look for, extract or elicit responses about the sequence of tenses and subjunctive clauses, were statistically more effective in producing the correct forms in contexts that required the use the subjunctive mood. Ferreira (2006:123) states, “Now, as the treatment period was relatively short (3 weeks) and also small number of subjects (24 subjects), we will have to conduct further studies to confirm the trends have been observed in this work”. However, despite these limitations, the study suggests that students of intermediate and advanced levels were supported in their learning more significantly by Group 2 strategies. It is proposed therefore that ITS for a foreign language should implement corrective feedback strategies that encourage students to correct themselves and their mistakes. In another research done, Morales & Ferreira (2008) conducted an empirical study based on BLENDED LEARNING (face to face and e-learning classes) in which they provided provide effective guidelines for researchers who develop computer platforms for foreign language learning. The main objective was to visualize how the methodological principles from the language teaching approaches - Task-Based Language Teaching (TBLT) and Cooperative Language Learning (CLL) - could be applied effectively in the design of activities to develop language skills in e-learning and blended environments. To this end, empirical evidence was explored about the effectiveness of learning English as a foreign language, in the face to face vs. blended modalities, in a study based on an experimental design - pre-test and post-test with control group. The results showed that the increase in learning English as L2 was higher in the experimental group that used a blended format than the control group who worked with the face to face modality. We propose, then, that models of blended learning methodology be included and implemented in the design of platforms for language teaching. The use of feedback included in the platform JClic 2 for focus on form exercises strengthened the statement about the importance of using different strategies defined in CALL and ICALL applications investigated in studies which suggest that its use increases

2

JClic is an authoring tool that allows teachers to easily create digital educational resources. The large user base with which his predecessor had, Click, will certainly be expanded as JClic to create greater variety of activities, with new features and to create resources whose display is not restricted to any particular operating system.

12

second language acquisition (Ferreira, 2006 & 2007). By incorporating these strategies in this model, the student was able to reflect and analyze in depth the linguistic elements of the target language. In this case, the feedback was an aid for learning, by using the application in the non face to face moments (e-learning) periods. Also, as JClic platform provides different resources for presenting materials, students benefited from a richer input in relation to the grammatical form than what generally would be provided in traditional instruction. Ferreira & Kotz (2010), in another study about the importance of feedback in CALL, designed and implemented a computational parser for the processing of grammatical errors in Spanish as a Foreign Language. The particularity of this input parser is that it must process erroneous entries and for this to happen, it is necessary to predict the mistakes that the user might make at a particular time of learning and specific grammar topic. The particular objective is

to

contrast

the

taxonomy

literature with empirical samples resulting

from from an

a

theoretical

observational

basis

on specialized

study in

traditional

classes in order to obtain more detailed information about the mistakes that ELE students could potentially make, especially for those whose native language is English. This

research

focus has been enriched by

research from

different

disciplines,

including second language acquisition, intelligent tutoring systems in procedural contexts, and intelligent tutoring systems for foreign languages. Thus, this research not only favours a specific area of study but it also nurtures both the face-to-face, non face-to-face and semi face-to-face modalities for language teaching. According to the above, it should be noted that modern computer technology allows students to practise and get feedback on both their written and spoken output (Krashen, 1987). The spoken output requires the kind of evaluative technology that might not be necessary for the assessment of written output; often, it includes the analytical elements that characterize some of the computer support applications for writing. Therefore, the analysis begins with identifying the main trends that lead to errors and processing errors written in the context of CALL. Written Errors There are basically three ways in which a computer can identify and treat a written linguistic error, one produced by a L2 student in what is supposed to be the target language. You can perform a PAIRING OF FORMS operation, use a parser, or use a hybrid system, in which the analysis is combined with the enunciation in an efficient manner. The 13

same grammatical analysis can be performed by a variety of parsers which will be subsequently identified. It can also vary depending on how it recognises and responds to errors. In addition, the system can have a component that allows you to address the linguistic levels of student output separately and therefore, perhaps in a more efficient way (Yoshii & Milne, 1995). These and other related issues will be discussed below. The PAIRING OF FORMS (Yoshii & Milne, 1995) is based on matching patterns of enunciations of the student’s output with a list of expected and pre-recorded responses. The statements are “contiguous sequences of characters that the application designer might want to find in the student’s input” (Yoshii & Milne, 1995: 64). The question is, however, whether this could be considered intelligent CALL, since there is no parser or other device usually associated with intelligent behaviour. Given that the system is nevertheless capable of great flexibility and is free from discrepancy in word choice, the part of speech or inflection, it can give the impression to users that it really “understands” their input, in which case it passes the Turing Test 3 and can be classified as intelligent. The primary objective of a parser is to decipher whether a sentence is grammatical, i.e. whether it conforms to the rules in the grammar of the parser. The uses of parsers are varied, including CALL applications, grammar and articles of other writers, translation software, dialogue systems, document retrieval and automatic extracting of main ideas. Even though the most sophisticated systems use the semantic pragmatic and topical analysis, as well as parsing, Holland et al. (1993: 30) believe that “they are the abstract linguistic rules which give the natural language processing power to handle a huge range and variety of text input”. The parsers in CALL can allow for language production, rather than a mere reception, and analyse a variety of sentences that do not have to be pre-programmed into the system. However, as should be the case in the pairing of statements that is quite remarkable, they also have a number of limitations. First, parsers rarely go beyond syntax - focus on form rather than meaning (Holland et al., 1993) - and this seems to break down the objectives of communicative 3

language

learning

is

presently

the

prevailing

theory

for

L2

Turing test called the procedure developed by Alan Turing for the existence of intelligence in a machine. The test consists of a challenge. The machine has to impersonate human in a conversation with a man through a chat-style text communication. The subject did not warn you if you are talking to a machine or a person. If the subject is unable to determine if the other party of communication is human or machine, then it is considered that the machine has reached a certain level of maturity: it's smart.

14

acquisition. Second, parsers are not infallible and may well fail to catch mistakes or recognize a completely correct sentence as such. Finally, the development of parsers and systems capable of using them is very expensive. This may be the main reason why we see relatively few of them in actual use with regard to CALL. “One way to improve the efficiency of parsers is to use techniques that encode uncertainty, so the parser does not need to select an arbitrary choice and later on retreat” (Allen, 1995). RETREAT is a procedure by which a parser can return to a previous state in the analysis if the chosen path does not seem to lead to successful parsing. With reduction parsers change, uncertainty or ambiguity is passed along to the point where all possibilities, except one, can be eliminated. This technique was developed to complement grammars designed for artificial languages so that they do not have any ambiguity and therefore only one interpretation is therefore possible (Allen, 1995; Tomlin, 1995; Ferreira, 2006). Because of the typical nature of ambiguity of the human language, such techniques help to avoid incorrect analysis or the need to delay a number of steps and thus slow down the process. The reduction parsers instead have the ability to look forward to the information that can resolve ambiguities and are, therefore, fast and efficient. The complexity is the nature of the systems that want to deal successfully with human language. The parsers alone often can handle a lot of language in terms of structures, but they may not be able to test, for example, if a response provided by a student has the right content included. While some systems build artificial constraints as to what input to allow, other systems use domain knowledge for verification of content. Two examples of this system are described below. The favourite example of Desmedt (1995) of an ICALL parser system that works on multiple levels and surely captures a series of errors of learners is the murder mystery game set in the Amber Productions "Herr Kommissar," designed for intermediate level German learners (Desmedt, 1995). The role of the learner in this game is to interrogate suspects in a murder case. According to Desmedt (1995), this task is not only a communicative immersion, but also allows the teaching and learning of languages through tasks, where language is used meaningfully to accomplish an extra-linguistic mission. The fact that the focus is on meaning, as suggested by Doughty & Williams (1998b) and Long & Robinson (1998), may cause 15

automatism in the use of linguistic form deemed necessary by some researchers (Ellis, 2001; MacWhinney , 2001). Therefore, the intelligent identification of errors and diagnostic systems, with regard to written language, can be quite sophisticated and use a number of subsystems combined with

a

variety

of

processing

levels,

making

them

almost

human

and

truly

intelligent. Immediately below, reference will be made to the oral production language skill and how it can be articulated and evaluated, using speech processing technology. Speech Processing Speaking is a language skill which, in particular, has had and continues to have a lot of importance within the framework of communicative language learning (Egan, 1999). According to Ehsani & Knodt (1998: 46), “foreign language syllabuses focus on productive skills with special emphasis on communicative competence”. Therefore, Eskenazi (1999: 62) makes the following statement, “Below a certain level, although the grammar and vocabulary are completely correct, effective communication cannot occur without the correct pronunciation because poor phonetics and prosody can distract the listener and prevent him from understanding the message”. However, to achieve a similar pronunciation, to that of a native speaker, as an adult L2 learner, is not an easy task (Ehsani & Knodt, 1998). By paying attention to the physiology of speech and hearing, researchers claim that the auditory nerves are specialized for auditory tasks in the early years of the life of a person, and thus restrict the range of sounds heard and interpreted (Ellis, 2001). This makes the task of recognizing and then repeating the phonemes of another language correctly more difficult. Therefore, an adult L2 student must follow a series of steps, which take a long time, to improve his pronunciation. These steps include producing a large number of sentences, receive pertinent corrective feedback, hear many different native models, emphasise prosody (amplitude, duration, and tone), and feel relaxed in the language learning situation (Eskenazi, 1999: 62). Therefore, the adult L2 learner has to perform a difficult task with limited learning abilities without feeling uncomfortable or lacking confidence, which is a kind of contradiction. Ideally, the amount of output required for this effort would be achieved in interactive language situations in pairs (Eskenazi, 1999: 62), which are often impractical and

16

too costly. The situation lends itself perfectly to CALL. For this particular objective, speech processing technology offers a promise (Egan, 1999). However, Ehsani & Knodt (1998) are concerned with the limited acceptance of new technologies within the language teaching community. The reasons they list for this situation are unknown to us: the lack of a unified theoretical framework, the absence of conclusive evidence, and the current limitations of technologies are among the most frequently cited (Nerbonne, Jager & van Essen, 1998; Chapelle, 1997; Salaberry, 1996; Holland, 1995). Ehsani & Knodt (1998) and Nerbonne et al. (1998) identify a major flaw in the reasoning of some of the fiercest opponents of CALL. While Nerbonne et al. (1998) use the terminology of formal logic to identify the error of division, Ehsani & Knodt (1998) use everyday language to coin the phrase “all or nothing reasoning”. The point postulated here is that what is true of the technology related to human language as a whole is little true of the technology related to restricted domains of language, such as those often found in the teaching and learning of languages. In order to make our own diagnosis of the value of speech processing technology in CALL, we must examine the current theory and practice that enhance effective teaching and learning of L2 pronunciation. Eskenazi (1999) reports that a “technique of listening and imitation” is often used to draw student attention to minimal pairs such as pin and bin of the English language. His research suggests that the awareness of L2 sounds would be the most effective way for students to learn to pronounce them. If a sound does not belong to the repertoire of speech sounds of a student, it may be associated with the nearest equivalent in the mother tongue of the learner. For example, if a native speaker of Arabic, which is just beginning to learn English, hears the sound / p / in pin of the English language, most likely he will associate it with the sound / b / of the word bin because the student does not have any awareness of / p / as a separate phoneme. As mentioned earlier, the linguistic discipline involved in the design of speech processing technology is the sub-discipline known as computational linguistics computational phonetics and phonology. While phonetics is concerned with the overall speech sounds, phonology, or phonemics, encompasses the phonemes or ideal sounds of a natural language. Computational phonetics and phonology are applied in two different approaches to speech: speech synthesis and speech analysis. Of the two, the former has a much longer 17

tradition and was started long before the advent of the computer. Thus, in 1939, Bell Laboratories piloted a device called a vocoder, or voice encoder, whose goal was to rebuild the human voice. It used a source of sounds and a set of filters whose values were derived from the analysis of human speech (O'Grady, Dobrovolsky & Arnoff, 1997). In theory, speech synthesis should be a simple procedure whereby the extracted speech sounds would be chained in words and phrases or statements. Unfortunately, this is not the case because speech sounds are not fixed, but vary according to the sounds around them. Adjacent sounds can be modified to each other in the segmental level. In addition to these local changes, supra-segmental features such as pitch, stress and intonation can have an effect on individual sounds. Therefore, there are many steps involved in speech synthesis. The text to be synthesized must be analyzed syntactically, orthographically and semantically. Subsequently, the pronunciation of words exceptional has to be found and contrastive sounds must be assigned, based on available information. After the correct sound is chosen, the system must be set in the context to select the most appropriate allophone. A parser then identifies words should that go together and must be assigned the appropriate prosodic features (O'Grady et al., 1997). The speech recognition task, on the other hand, is that of taking waves from the speech sounds and decoding them (O'Grady et al., 1997). This task is much easier for humans than for computers. Once it is simplified in steps, a human being has no trouble coping with rapid, informal or whispered speech, including defective expression in a continuous set of sounds, even under exasperating as background noise. To a computer, all these things create problems (Wachowicz & Scott, 1999), which is why some systems require low input with pauses between words, a limited vocabulary, speaker dependency and exclusion of external noise through use of special microphones. When it comes to the speaking (oral production) skill within the L2 paradigm, especially in terms of CALL, particularly if one considers the diagnosis and correction of errors, speech recognition and its quality become critical issues. Recognising and understanding human speech require a considerable amount of linguistic knowledge on a phonological, lexical, semantic, grammatical and pragmatic level. While the linguistic competence of adult native speakers covers a wide range of reconnaissance and 18

communicative activities, computer programmes work best when they are designed to operate in clearly delineated sub-fields of linguistics. Automatic speech recognition starts with the analysis of incoming speech signs. When a person speaks into a microphone, the computer takes a sample of input and creates an accurate description of the sign of speech. Afterwards, a number of acoustic parameters such as information on energy, spectral features, and tone are derived from the speech sign. This information is used differently, depending on whether the system is in the TRAINING PHASE

or RECOGNITION PHASE. In the training phase, it is used for the purposes of

modeling the speech sign. In the recognition phase, it is collated with the already existing model of the sign. Training a computer programme to recognise the spoken language means “modeling the basic speech sounds”. These are subsequently used in the recognition phase to identify words. The training requires a large amount of data representative of the type that the system is expected to recognise. In general, an automatic speech identifier is not able to process speech that differs considerably from the speech in which it has been trained (Ehsani & Knodt, 1998). Therefore, dictation systems, continuous and independent of the speaker with extensive vocabularies, are usually trained with thousands of expressions read by several speakers, including dialects and different age categories. The last piece of the puzzle is the decoder or an algorithm that maximizes the probability of a pairing between the speech sounds and the corresponding utterance. This can be described as a search problem, because, in large vocabulary systems, questions about the effectiveness and optimisation must be carefully considered. The crucial question is whether to comply with the most probable hypothesis or working with multiple solutions in parallel (Ehsani & Knodt, 1998). Although the second option might be more accurate, the first would be faster. Just like the natural language processing systems which have been discussed before, commitment would be necessary to achieve the best possible result within an acceptable time frame. It has also been warned that “the recognisers work faster and more accurately when the incoming speech is clearly articulated in an environment without noise, when task complexity is low, and when the dictionary is small” (Ehsani & Knodt, 1998: 50). Thus, the precise task definition, the ideal acoustic modeling and input quality are the essential 19

characteristics of successful speech recognition. Acoustic modeling must take into account a sufficient number of representative speakers, and features can be increased by a single speaker if necessary. The quality of the input requires a good sound amplifier and microphone to exclude background noise, and is mounted at a constant distance from the speaker's mouth. This is a basic description of a speech recognition system. How such systems are being used in CALL will be described immediately below. Speech Recognition in CALL Lately, these have begun to be used in interactive voice systems CALL (Ehsani & Knodt, 1998, Egan, 1999) to teach pronunciation, reading aloud and some limited conversation skills (Holland, 1999; Wachowicz & Scott, 1999; Egan, 1999). Of the three, the teaching of pronunciation has the most demands on the system in terms of feedback. It would seem that even in the area of pronunciation, computers can capture deviations from what is expected and provide feedback that does not depend on the very perception of the student ( Ehsani & Nkodt, 1998). This is possible both on the segmental (speech sounds) and the supra-segmental (prosody of the sentence) levels. The traditional linguistic theory in the past focused its attention on speech sounds, i.e. the segmental level. For this reason the teaching of pronunciation used to focus on segments or articulatory phonetics or individual sounds (Chun, 1998). A number of ancient CALL programmes also paid attention to the suprasegmental level (Wachowicz & Scott, 1999). Ehsani & Knodt (1998) provide us with a description of a task for feedback on the supra-segmental level. A conventional speech recogniser is designed to process the speaker's expression, whatever the pronunciation of it. The acoustic models are generalised to accept and correctly recognise a wide range of different accents and pronunciations. A pronunciation tutor, however, must be trained both to recognise and correct subtle deviations from standard native pronunciations. Technically, the design of an interactive voice pronunciation tutor goes beyond the state of the art required by commercial dictation systems. While grammar and vocabulary of a pronunciation tutor are relatively simple, the underlying technology of speech processing tends to be complex and must be customised to recognise and evaluate non-fluent speech of language learners (Ehsani & Knodt, 1998). 20

For the reasons stated above, the speech data obtained from non-native speakers is a very important when it comes to training independent, continuous, and large vocabulary voice recognisers, (Egan, 1999). Objective data can be simplified into three main categories: read data, planned or careful data, or spontaneous data (Tamokiyo & Burger, 1999). Two examples of spontaneous speech are conversation and question. The distinctive feature of spontaneous speech is its subconscious quality (Tamokiyo & Burger, 1999), which means that the speaker does not really pay attention to the speech act. There are doubts as to whether this could be said of non-native semi-fluent speakers (Tamokiyo & Burger, 1999). Thus, it is unclear whether, and to what extent, their speech is really spontaneous. Such reasoning seems to be like the comprehensible output hypothesis (Swain, 1998) that represents the process of generating non-native speaker ouput, as a series of conscious decisions and projects for testing hypotheses. There are, however, some observable differences between the speech of native and nonnative speakers. Tamokiyo & Burger (1999) provide evidence that while native speakers’ spontaneous speech contains disfluencies, pauses, crutches and a choice of syntactic structures that are often characteristic of a particular speech style, non-native speakers can demonstrate an extreme measure of disfluencies, pauses between words and errors without these constituting signs of a developed conversational style. Read speech, on the other hand, contains reading errors and hesitation which do not occur in spontaneous speech. Non-native speakers may have a large number of reading errors, especially when dealing with unfamiliar vocabulary. For all these reasons, Tamokiyo & Burger (1999) recommend collecting sampling of read and spontaneous speech of non-native speakers. The collection of samples of read speech is important because when an announcer reads a sentence or a paragraph written correctly, he reads it continuously, without hesitations, corrections, etc. As for spontaneous speech, obtaining models for it provides valuable information because it is characterised by the incorporation of extra-linguistic elements such as filled pauses and corrections to grammar, among which false starts of sentences and phrases, repetitions, self-corrections, etc., are included. Reading aloud is a pronunciation exercise and literacy skill (Ehsani & Knodt, 1998) which helps students establish a link between the speech sounds and writing. Since this ability is important, the architecture of the speech recognition reading tutor is simple since 21

there is only one correct answer possible. However, this is a more challenging task of recognising and responding adequately to disfluencies such as hesitation, mispronunciations, false starts and self-correction. The aim of such systems is to measure reading fluency represented by the variable such as reading speed, silence between words and disfluency measurement, i.e. false starts, self-corrections and omissions (Ehsani & Knodt, 1998). Conversation, on the other hand, is a more divergent task than reading aloud. However, systems designed to recognise speech in a conversational environment can be designed in two different ways, allowing closed-ended or convergent, or open-ended or divergent responses (Ehsani & Knodt, 1998). The first allows the student to choose one of the answers provided (LaRocca, Morgan & Bellinger, 1999), while the latter does not reveal the answer given by the student. However, the differences between the two architectures are not large, since both should have all possible correct answers. The student, however, faces a major challenge because he has to solve the answers for himself (Wachowicz & Scott, 1999) without any help from the system. This is a response to the concerns of Eskenazi (1999) about the passive role of the student with closed-ended response systems, where the student reads aloud a written option or repeats a sentence learned in response to a question. Ehsani & Knodt (1998) highlight that the majority of CALL interactive voice systems are designed to teach and evaluate language forms, including pronunciation, vocabulary and grammatical structures. One reason for this is that formal features can be clearly identified that contribute to the robust performance of the system. Another reason is that cognitive theories of L2 acquisition and linguistic approaches such as awareness (Allen, 1999, James, 1998) on the one hand, and focus on form (Long & Robinson, 1998, Doughty & Williams, 1998) on the other hand, see the advantages of making focus on form and integral part of the language learning process. To not deviate from the fundamental statement raised in this article, the ways in which CALL systems based on speech recognition and how they detect and respond to errors must be examined. To ascertain whether they can identify errors, one can say that these systems themselves are able to both diagnose and correct either segmental or supra-segmental errors (Eskenazi, 1999), albeit with restrictions (Wachowicz & Scott, 1999; Egan, 1999). There is, however, a huge difference between the segmental and supra-segmental levels. While the number and nature of phoneme and acceptable pronunciation space for each 22

phoneme vary in all languages and present a major problem to the learner, the prosodic features involving the duration of the tone and intensity are the same in all language, with the variation that occurs in the relative importance, significance and variability of these traits in different languages (Eskenazi, 1999). Although interactive commercial CALL systems - of which there is a satisfactory number on sale - often misjudge the student's input, preventing learning, several strategies have emerged to minimise the weaknesses of these systems (Wachowicz & Scott, 1999). These strategies include the input verification of entry and task-based language teaching and learning, among others (Wachowicz & Scott, 1999). Input verification means that, if in doubt, the system should provide a possible representation of the expression on the screen and ask the learner if that's what he meant or wanted to say. Thus, the student may be more willing to cooperate, repeating expressions instead of feeling frustrated by the shortcomings of the system. The improvement of authenticity means that the development of the conversation may depend on the student's response and not have to be integrated. Finally, the integration of task-based language teaching and learning means that the student would be asked to complete a real life significant task through the productive skill, that of speaking. According to Wachowicz & Scott (1999), the best systems available have some of the afore-mentioned features besides giving implicit corrective feedback in multi-modal formats. They also centre their attention on relatively predictable conversations and give learners a chance to correct their own mistakes. Some examples of CALL interactive voice systems A previous example of a CALL system based on speech recognition, with a focus on both the segmental and supra-segmental level of speech is the SPEEECH OBSERVER developed by IBM (Stenson, Downing, Smith & Karen, 1992). Originally designed to work with patients suffering from various communication disorders, the SPEEECH OBSERVER provides a variety of visual demonstrations designed to focus attention on various aspects of pronunciation and develop better pronunciation skills. The three main modules of the SPEEECH OBSERVER

are AWARENESS, SKILL DEVELOPMENT and MODELING that are

subsequently simplified into the following components: The AWARENESS module monitors the tone, voice onset, intensity and the expression itself. The visual feedback for the tone is shaped like a thermometer and the mercury rises 23

with the increase in the tone. The onset of voice is represented as a train which advances at every start-point. Intensity and expression are represented by a clown whose nose increases in amplitude and whose tie changes colour when the voice changes. Similar graphic devices are used in the SKILL DEVELOPMENT module to represent tone, expression, accuracy and vowel contrast. These two modules, therefore, have a design similar to a game, while the third module, MODELING, is based on spectral analysis. It uses colour for tone and intensity, it provides the waveform model for expressions and presents spectra for vowel formants. One of the strengths of the programme is that it allows students to compare their expressions instantly produced or pre-recorded native speaker utterances. Used in an experiment with two groups of adult learners to teach L2 sentence intonation and pronunciation of technical terms and phrases related to the area, it contributed to the highest post-test grades in the experimental group were, however, insignificant (Stenson et al., 1992).The affective attitude towards the software was positive, but the hardware limitations at that time seemed quite restricted. Another example of a CALL interactive voice system is SUBARASHI, which is a system designed to practise conversation in Japanese. It is based on continuous speech recognition in a framework of divergent dialogue (Bernstein, Najmi & Ehsani, 1999). Students are involved in the interaction with the system in a number of situations in which they have to talk to various characters. To make meaningful interaction, students are expected to address problems that can only be solved by means of oral communication. Even if the student has the freedom to select the random encounters, the system is designed as an adventure game where every game is built on the previous meeting. Depending on the purpose of the meeting, the student or the system can start the conversation. If the student attempts to negotiate an outcome contrary to the objective of meeting, he is reminded of his mission. For example, they could say to a student that it is very likely that one of the characters might invite him to a tour, but that his task is to correctly and firmly reject the invitation. This system uses HIDDEN MARKOV MODELS for speech recognition tasks, and was trained with a large corpus of spoken Japanese. The latter makes it different from a number of CALL programmes that use speech recognition, which are usually trained with a limited 24

number of speech samples of the target language. Because of its extensive training with a large number of speakers and its sophisticated technology, the system is able to function better in the non-native speech recognition as compared to similar CALL applications. After the initial pilot study, learners found that some situations - negotiation of meaning, pronunciation exercises, among others - were difficult for them, prompting the authors to add several additional features: pronunciation training, closed-ended questions and obligatory grammar exercises. These modules had the function of preparing students for the meetings. The use of high-quality multimedia gives the system a realistic feel.

Conclusion The development of new technologies in recent decades has led us to reflect on the possibilities that multimedia technology can bring to foreign language teaching with its advantages and limitations. In recent years there has been a breakthrough in the use of computers applied to CALL. Until only a decade ago, the use of computers in the language class was something that was relegated to a few specialists in the field. However, with the development of multimedia technology and the increasingly widespread use of the Internet, the role of computers in the foreign language classroom has become a major issue in which an increasing number of teachers around the world are becoming involved. Education is required to address this new reality, accepting the possibilities that new technologies are offering and knowing how to guide their implementation positively to avoid imbalances generated by a purely mechanical and comfortable use. In this society, more important to remember is to teach students the strategies to select and access information according to their needs. Thus, we have to find a way to reconcile the force the world of images is occupying in our society, with the need to train people who are fully autonomous and critical thinkers. We can say that technological change has paralleled the evolution of the different foreign languages teaching methodologies in a positive and important effort to adapt to the possibilities that the society offers at all times. If the mainframe was the technological base of BEHAVIOURIST

CALL, the personal computer would be the technological base of

COMMUNICATIVE

CALL. Presently, multimedia computers are the technological foundation 25

with which INTEGRATIVE CALL functions. Currently, multimedia technology offers a variety of information, production and communication tools; in fact, it also provides the possibility for a much more integrated use of technology. This is good news for both teachers and students. Indeed, one of the fundamental characteristics of the modern world in which we live is that there is always an inescapable need to learn to read, write and communicate through the computer in any area of our lives. It is sufficiently clear that multimedia technology offers, no doubt, many advantages. It encourages the process of foreign language learning, which always requires a long and continuing effort by the student, in the sense that it provides many opportunities and facilities to get a better performance in this effort, while adapting to the individual learning pace of each student. It helps to develop, especially, oral and written comprehension, vocabulary acquisition and retention, and it also helps to improve pronunciation. On the other hand, we must also recognize that multimedia technology still has many limitations, mainly in regard to value and correct the very language productions of the student. In this sense, the development of oral and written expression, fundamental in achieving communicative competence in foreign languages, is an issue that CD-ROM programmes only deal with superficially. For this reason, the higher the level of foreign language skills by the student, the greater the need is to increase face to face teaching which would allow you to practise something as important as conversation. The progress in voice recognition software and oral interaction with the computer, already being developed, will contribute progressively to remedying this deficiency. Overcoming what we call INTEGRATIVE CALL will come from the hand of what is becoming known as Intelligent CALL, and it will give more and more satisfactory answers to the challenges that arise in the near future to get an increasingly active, simple and straightforward interaction with the computer and its usefulness. It is necessary to investigate the use of multimedia technology taking into account what is known about language acquisition and especially learning strategies. Thus, we will be able to identify to what point, to what extent and how new technologies promote the teaching and learning of foreign languages. It would also be necessary to delve into issues such as the type and amount of interaction that is generated using the computer, what students and teachers think about 26

technology and how they use it, what are their attitudes towards the media, and investigate about their effectiveness in developing the four language skills. Meanwhile, we must continue working with the many media that technology has put at the service of education. And it seems that, ultimately, the question before us is not “what is the role of multimedia in foreign language class”, but rather just the opposite, “what is the role of the foreign language class in this technological age of information?” Perhaps the only possible answer is that we must prepare our students to work and function in a digitally connected society, where most communications will take place in the target language. New technologies applied to education should, therefore, help to develop different learning opportunities for students and, consequently, teachers must be the first to accept them as an increasingly indispensable tool for our educational work, but without fear that in no case can they become a substitute for language teachers.

References Allan, M. (1999). Language awareness and the support role of technology. In R. Debski and M. Levy (Eds.). In WORLDCALL. Global perspectives on computer-assisted language learning. Pp. 303–18. Lisse: Swets & Zeitlinger.

Allen, J. (1995). Natural language understanding. Redwood City: Benjamin/Cummings.

Bernstein, J., A. Najmi y F. Ehsani. (1999). Subarashi: Encounters in Japanese spoken language education. CALICO Journal 16 (3), 261–384.

Chapelle, C. (2001). Computer application in second language acquisition. Foundation for teaching, testing and research. Cambridge: Cambridge University Press.

Chapelle, C. (2003). English language learning and technology. Philadelphia, PA: John Benjamins B.V.

Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press.

Chun, M.D. (1998). Signal analysis software for teaching discourse intonation. Language 27

Learning & Technology 2 (1), 61–77 Available at http://llt.msu.edu/vol2num1/article4/.

Corder, P. (1967). The significance of learner errors. International Review of Applied Linguistics 5, 161–70. Doughty, C. (2001). Cognitive underpinnings of focus on form. In P. Robinson (Ed.). Cognition and second language instruction. Pp. 206–57. Cambridge: Cambridge University Press.

Doughty, C. & Williams, J. (1998). Pedagogical choices in focus on form. In C. Doughty & J. Williams (Eds.). Focus on form in classroom second language acquisition. Pp. 197–262. New York: Cambridge University Press.

Egan, K.B. (1999). Speaking: A critical skill and challenge. CALICO Journal 16 (3), 277–94.

Ehsani, F. & Knodt, E. (1998). Speech technology in computer-aided language learning: strengths and limitations of a new CALL paradigm. Language Learning & Technology 2 (1), 45–60. Available at http://llt.msu.edu/vol2num1/article3

Ellis, N. (2001). Memory for language. In P. Robinson (Ed.). Cognition and second language instruction. Pp.33–68. Cambridge: Cambridge University Press.

Ellis, R. (1985). Sources of variability in interlanguage. Applied Linguistics 6, 118–31.

Eskenazi, M. (1999). Using automatic speech processing for foreign language pronunciation tutoring: some issues and a prototype. Language Learning & Technology 2 (2), 62– 76. Available at http://llt.msu.edu/vol2num2/article3

Ferreira, A. (2006). Estrategias Efectivas de Feedback Positivo y Correctivo en Español como Lengua Extranjera. In Revista Signos 39 (62), 309-406. Valparaíso: Universidad Católica de Valparaíso.

28

Ferreira, A. (2007). Estrategias efectivas de feedback correctivo para el aprendizaje de lenguas asistido por computadores. In Revista Signos 40 (65), 521-544. Valparaíso: Universidad Católica de Valparaíso.

Ferreira, A. & Kotz, G. (2010). ELE-Tutor Inteligente: Un analizador computacional para el tratamiento de errores gramaticales en Español como Lengua Extranjera. In Revista Signos, 43 (73), 211-236. Valparaíso: Universidad Católica de Valparaíso.

Garrett, N. (1987). Modern media in foreign language education. Lincolnwood: National Textbook.

Garrett, N. (1991). Technology in the service of language learning: Trends and issues. Modern Language Journal 75 (1), 74-101.

Garrett, N. (1995). ICALL and second language acquisition. In V.M. Holland, J.D. Kaplan & M.R. Sams (Eds.). Intelligent language tutors. Pp. 345–58. Mahwah: Lawrence Erlbaum.

Gregg, K. R. (2001). Learnability and second language acquisition theory. In P. Robinson (Ed.). Cognition and second language instruction. Pp. 152–82. Cambridge: Cambridge University Press.

Heift, T. & Schulze, M. (2003). Error diagnosis and error correction in CALL. CALICO Journal 20 (3), 437–50.

Heylighen, F. (1995). Evolutionary epistemology. Available at Principia Cybernetica Web http:// pespmc1.vub.ac.be/EVOLEPIST.html

Holland, M., Maisano, R., C. Alderks & J. Martin. (1993). Parsers in tutors: What are they good for? CALICO Journal 11 (1), 28–46. 29

Holland, V.M. (1995). Introduction: The case for intelligent CALL. In V.M. Holland, J.D. Kaplan & M.R. Sams (Eds.). Intelligent language tutors. Pp. 7-16. Mahwah: Lawrence Erlbaum.

Holland, V.M. (1999). Tutors that listen. CALICO Journal 16 (3), 245–50.

James, C. (1998). Errors in language learning and use: exploring error analysis. London: Longman.

Krashen, S.D. (1987). Principles and practice in second language acquisition. London: Prentice-Hall.

LaRocca, C.S.A., Morgan, J. J. & S.M. Bellinger. (1999). On the path 2X learning: exploring the possibilities of advanced speech recognition. CALICO Journal 16 (3), 295–310.

Levy, M. (1997) Computer-assisted language learning: context and conceptualization. Oxford: Oxford University Press.

Levy, M. (1998). Two concepts of learning and their implications for CALL at the tertiary level. ReCALL Journal 10 (1), 86–94.

Levin, L. & Evans, D. (1995). ALICEchan: A case study in ICALL theory and practice. In V. Holland, J. Kaplan, & M. Sams, (Eds.). Intelligent language tutors: theory shaping technology. Lawrence Erlbaum Associates.

Liou, H. C. (1991). Development of an English grammar checker: A progress report. CALICO Journal 9 (1), 57–71.

Long, M. H. & Robinson, P. (1998). Focus on form: Theory, research and practice. In C. Doughty y J. Williams (Eds.). Focus on form in classroom second language acquisition (pp. 15–41). New York: Cambridge University Press. 30

MacWhinney, B. (2001). The competition model: the input, the context and the brain. In P. Robinson (Ed.). Cognition and second language instruction. Pp. 69–90. Cambridge: Cambridge University Press.

Miller, T. (2003). Essay assessment with latent semantic analysis. Journal of Educational Computing Research, 28 (3). Available at http://www.dfki.uni-kl.de/~miller/ publications/miller03a.pdf.

Mitchell, R. & Myles, F. (1998). Second language learning theories. London: Arnold.

Morales, S. & Ferreira, A. (2008). La efectividad de un modelo de aprendizaje combinado para la enseñanza de inglés como lengua extranjera: estudio empírico. Revista de Lingüística Teórica y Aplicada, 46 (2), 95-118. Concepción: Universidad de Concepción, Chile.

Muñoz, J. (1991). La adquisición de las lenguas extranjeras. Mardid: Visor.

Nerbonne, J., Jager, S., & A. van Essen. (1998). Introduction. In J. Nerbonne, S. Jager y A. van Essen (Eds.). Language teaching & language technology. Lisse: Swets & Zeitlinger.

O’Grady, W., Dobrovolsky. M. & M. Arnoff. (1997). Contemporary linguistics. Boston: Bedford/St. Martin’s.

Pennington, M. (Ed.). 1989. Teaching languages with computers: the state of the art. La Jolla, CA: Athelston.

Rogers, E. (1983). Diffusion of innovations. London: Macmillan.

Salaberry, M.R. (1996). A theoretical foundation for the development of pedagogical tasks in 31

computer mediated communication. CALICO Journal 14 (1), 5–36.

Smith, G.W. (1991). Computers and human language. New York: Oxford University Press.

Stenson, N., Downing, B., J. Smith & J. Karen. (1992). The effectiveness of computerassisted pronunciation training. CALICO Journal 9 (4), 5–20.

Stevens, V. (Ed.). (1989). A direction for CALL: From behavioristic to humnaistic courseware. In M. Pennington (Ed.), Teaching languages with computers: the state of the art. Pp. 31-43. La Jolla, CA: Athelston.

Swain, M. (1998). Focus on form through conscious reflection. In C. Doughty y J. Williams (eds.). Focus on form in classroom second language acquisition. Pp. 64–82. New York: Cambridge University Press.

Tamokiyo, L.M. & Burger, S. (1999). Eliciting natural speech from non-native users: collecting speech data for LVCSR. ACL-ICALL Symposium available at http:// acl.ldc.upenn.edu/W/W99/W99–0402.pdf

Taylor, R. (1980). The computer on the school: tutor, tool, tutee. New York: Teachers College Press.

Tomlin, R. (1995). Modeling Individual Tutorial Interactions: Theoretical and Empirical Bases of ICALL. In Intelligent Language Tutors: theory shaping technology. Lawrence Erlbaum Associates.

Tschichold, C. (1999). Intelligent grammar checking for CALL. ReCALL special publication, Language Processing in CALL, 5–11.

Tschichold, C. (2003). Lexically driven error detection and correction. CALICO Journal 20 (3), 549–59.

32

Underwood, J. (1984). Linguistics, computers and the language teacher: A communicative approach. Rowley, MA: Newbury House. Van Lier, L. (1996). Interaction in the language curriculum: awareness, autonomy, and authenticity. New York: Longman. Wachowicz, K.A. & Scott, B. (1999). Software that listens: it’s not a question of whether, it’s a question of how. CALICO Journal 16 (3), 253–76. Warschauer, M. (1996). Computer-assisted language learning: an introduction. In S. Fotos (Ed.), Multimedia Language Teaching. Pp. 3-20. Tokyo: Logos International

Warschauer, M. (2000). The death of cyberspace and the rebirth of CALL. English Teachers' Journal 53, 61-67. Warschauer, M. & Healy, D. (1998) Computer and language learning: an overview. Language Teaching, 31, 57-71.

Warschauer, M. & Kern, R. (2000). Network-based language teaching: concepts and practice. Cambridge: Cambridge University Press.

Yoshii, R. & Milne, A. (1995). Analysis of and feedback for free form answers in English and Romanized Japanese. CALICO Journal 12 (2–3), 59–88.

33