Communicative CALL With Artificial Intelligence: Some Desiderata

Posted: December 23, 2010 in CALL related to Linguistics

 

AUTHOR:

Richard Barrutia

LINK:

https://calico.org/html/article_264.pdf

Abstract:
The ideal CALL courseware development procedure will take advantage of many or all of the latest hardware, pedagogical, and theoretical advances in the language teaching field. These advances are presented in this paper both as a description o the ideal CALL courseware and, where applicable, as actual design implementations in a Spanish course being developed by the author.

KEYWORDS: artificial intelligence, branch programming, Spanish, hardware, design, development, courseware, software, desiderata, expert system, language acquisition, communicative drill.

This article is written with the purpose of presenting a desiderata for the best possible features in a foreign language computer courseware program which meets the criteria of the most modern communication oriented language acquisition methodologies. Prior to establishing such a desiderata the author conducted extensive research of existing and available foreign language courseware in order to see how closely such programs adhere to the newest second language acquisition principles and how much branching and AI (artificial intelligence) was being used. 

It is well documented in many recent articles, reviews and books on CAI in foreign languages (e.g. Baker, 1984) and (Underwood, 1984)1 that there exists precious little programming which even pays glancing attention to the basic common denominators of the newest FL teaching methods. Further, there are few, if any, programs which employ branching techniques that act upon the students’ choices in order to effect simulations, artificial intelligence, or some other interactive analysis of the student’s progress. Indeed, the commercial courseware available today has taken a conformist step backward in imitation of other programs and/or standard textbook material, in many cases making a mere page turner out of the computer.

Such branch programming will provide a possibility of many different routes through the program; thus each student virtually writes his own program as he goes. Since each branch is programmed with supplementary information that is applicable to the student’s response, the individual learning style of the student is analyzed by the computer and it can therefore alter its own course accordingly.

With regard to the analysis of learning styles, the interaction can take a form that the computer has never presented before by combining graphics, written work, sound, or traditional technique as dictated by the unique and individual learning style of a given student.

An expert system thus arises from a computer program that mimics a human expert using the methods and information acquired and developed by a human expert. An expert system will be able to solve problems, make predictions, suggest positive treatment and offer advice with a degree of accuracy closely equivalent to that of its human counterpart.

This is done in part by programming the computer to first keep a record of all right, wrong and fuzzy answers. Then to record what style of presentation (script, graphic, voice, or combination etc.) rendered the best answers for a given student. With this knowledge the computer can then emphasize the type of display that empirically works best for that student.

37

A clarification of the above statement concerning errors (wrong and fuzzy answers) is essential here before going on. The so-called errors built into the program are not errors at all in the normal sense of the word. All the multiple-choice alternatives should always be correct in syntax, intonation, and pronunciation. That is, the student should never hear a wrong form in itself. These correctly phrased and pronounced utterances, however, might be put by the student into the wrong environment. The errors occur because some rejoinders are more fitting than others. They are more or less aptly satisfy the needs of the environment framed by the initial question. Only one rejoinder, of course, is completely correct in the given context. The expert system takes this fuzzy consensus of not totally wrong answers into account in deciding what to present next.

Courseware/Software Desiderata

The courseware should utilize many items, methods, options, etc.

Extended silent listening (pre-production phase)

Natural Approach Techniques (Terrell)

Spoken and illustrated dialogs (VCR)

TPR with spoken written and graphic commands (Asher)

Vocabulary-building segments with graphics

Pictionary (extensive photo inventory of actions, places, things etc.)

Digitized voice questions and answers

Constant Interaction (Student instigated movement of the program)

Simulation sequences (Student with teacher-with another student-with policeman, etc.)

Multiple-choice tests (text, voice, graphics)

Branching frames (according to degree and nature of error in test frames)

Intrinsic programming (nonlinear sequencing)

Cloze exercises (graduated from every 6th, 5th, 4th, 3rd word blanked)

Traditional grammar approaches (where learning style patterns indicate potential)

Dictation exercises (with immediate correction feedback)

Writing practice (with immediate correction feedback)

Visual Phonetics (Displaying native to students’ formant wave patterns)

Functions of An Expert System

The above components are to be organized and supported by an expert system to alter the presentation of items and frames by analyzing the student’s feedback and reacting with an appropriate response equivalent to a human teacher. If the student’s aural comprehension is weak, it should reinforce listening skills. If the syntax is weak, it should present grammar drills. If the vocabulary control is limited, it should deliver comprehensive vocabulary-building exercises through speech test and graphics.

Beginning The Research

The initial research for the instructional software should always begin with comparison and analysis of all available empirical data gathered in previous research experiments on second language acquisition. Where empirical evidence is not available, programmers should study the most advanced scientific hypothesis about optimal conditions for language acquisition. Comparisons should be made of orientations and evidence as provided by such researchers as Earl Stevick, A Way and Ways; Charles Curran, Counseling-Learning; James Asher, Total Physical Response; Caleb Gattegno, The Silent Way; Tracy D. Terrell, The Natural Approach; Georgi Lozanov, Suggestopedia; and others. Also instructional experiments as conducted by researchers at the State University of New York, Brigham Young University, Stony Brook, and software languages such as PLATO IV, ELIZA, DART, LOGO and LISP should be reviewed and evaluated. In addition to this comparison of linguistic methodologies and software, programmers should conduct their own search for computer-based foreign language programs already using branching systems.

The present research for a better program is being pursued with a rekindled interest in an original branching project (1964-70)2 by this author who has now been encouraged by the many advancements in both applied linguistics and digital speech synthesization, as well as cheaper and more extensive memory systems available since that early research. It is also being pursued because of the enhanced possibilities of using artificial intelligence not only to alter the program, but to monitor students’ learning styles and thus to present sequences that were not programmed into the courseware originally. The potential of CALL becomes renewed and more exciting when one considers the growing need of teaching foreign languages in the U.S. today.

Design of Ideal Courseware

The ideal courseware described here introduces a much needed approach to computer-assisted study of foreign language. The development of such an innovative project should be implemented on an interactive microcomputer-based system. The courseware should include branch programming using high quality voice reproduction recorded digitally directly on the chip and recently developed programming techniques based on expert systems, (a branch of Artificial Intelligence). An expert system database will allow for the monitoring of the language acquisition process and will dynamically modify the branching logic to deliver the appropriate information frames according to students’ needs. Using the criteria the initial courseware is now being developed by the author for the Spanish language, but the methodology could be used for other languages.

The instructional needs of the program must be given priority over the tools currently available to the profession.

38

Consequently, special computer software should be developed whenever needed to conform to the courseware rather than vice versa. We favor a small individually-computerized system which will perform nearly all of the feedback and branching operations that can be done by the large computers. What will result will be a courseware program that will give birth to computer software and a configuration of hardware devices rather than a program written for a specific piece of equipment and/or software.

The end result will be the production of an economical interactive microcomputer-based system to aid in teaching foreign (or any) language which will include a voice digitizer/synthesizer and courseware based in branching methodology and AI to recognize and react to observed and recorded learning patterns. The major implication of this development is, of course, the production of a device which will greatly aid in the selfstudy of languages.

Branch Programming

The CALL program described here should not only anticipate errors based on pressures from the learners’ native speech, but it should also further utilize those errors and others so as to signal itself where to move next. By comparison a linear program is extrinsic, could be presented by a book, is passive and cannot alter itself or function with a student in this way. Linear programs are often boring and even insulting to the intelligence of a bright student, mainly because the steps are made extremely miniscule in order to avoid error. Nevertheless, some linear programs can and do build up behavioral patterns as per their objectives in certain cases. On the other hand, a branching program will receive signals from one student or another and interact with that student for as many calculated errors as are written into the program.

The writing of a communicative branch program is a difficult and painstaking task. A great number of questions must be created and recorded in segments. These questions are often the same type of questions asked by students in language classes or questions asked of the class by an expert human teacher. The questions must then be followed by alternative answers or reactions which may be presented to the student with frequency in a multiple choice or some other format, e.g., fill-ins. The alternative answer or branch can be chosen by experience gained in teaching a foreign language or another means is by anticipating or predicting through contrastive linguistic analyses what types of errors might be made concerning certain foreign language learning problems. A sample multiple choice set of questions might be the following:

:Q: Quienes son esas dos chicas? (with visuals and synthesized voice)

:Rejoinders:

A. Son amigos de mi hermana.

B. Es amiga de Alicia.

C. Voy a la biblioteca a estudiar.

D. Son amigas de mi hermana.

This little test, like all others in the program, should be thought out very carefully where both the rejoinders and the initial question are concerned. In fact, the alternative responses—those with the highest degree of interference and predictability as possible wrong answers—should be selected from an even longer list of examples, some predictable, some arbitrary. As will be shown in the following paragraphs, this selection can be made by a simple but fuzzy scale of candidacy where NR equals Nearly Right Answer, CW equals Common Wrong Answer, and VW equals Very Wrong Answer.

When an expert teacher analyzes this small test, it is seen that a student selecting answer “A” (CW track) has missed hearing the two gender markers in the question. That student will need more work on the a-o sound contrasts and the computer will consequently deliver such a branch of instructional material to that student. A student who selects answer “B,” (NR track) though nearly right has amore grievous error than the first, because he has missed hearing the number agreement which was marked five times (in fact, in every word) of the original question. That student will work with plural agreement exercises. A student picking the “C” answer (VW track) has not, of course, understood a single thing in the question and will be routed back for practice on the original material that included the frame question. He will see more graphics illustrating vocabulary meanings and have more listening comprehension work than either of the other students. All three students will, at some point, either on the mainline track or in some other branching frame, be tested by the same question again until passing it successfully. Answer “D” (RA track) will generate a whole new set of material but with the computer deciding what, by taking into account the student’s previous learning record (number of right scores), learning pattern (closeness of wrong answers), and style (speed and ease of interaction and choice of sound, pictures etc.)

Communicative CALL

In discussing communicative CALL as Stevick and Underwood point out that, “Mere contextualization of languages is not in itself communication. Communication is the exchange of information. Describing what is obvious in a picture adds no new information to the picture, hence nothing is communicated” (1982, 130). Such an activity would be an example of what Paulston (1980) calls ‘meaningful drill’: the learner must understand what she is saying, but the content of the response is already known.

In contrast, in a ‘communicative drill’ the learner adds new information about the real world, even though the structures she uses may be controlled. Although most of the CALL activities described in this chapter would fall into the ‘communicative’ category, there are also some useful exercises of the ‘meaningful’ type (Underwood 1984).

The Role of CALL

The reader could well ask, if adhering to the newest pedagogy and, for instance, to Krashen’s first hypothesis of the Acquisition-Learning Distinction; How can one justify the use of CAI at all given that the computer is more effective for teaching learning principles than for the more human communicative acquisition approaches. This is a good question that quite correctly addresses the heart of the problem where CALL is concerned. Obviously, we are decades, centuries or perhaps even an eternity away from having devices that are equal to human beings and a human society in all the nuances of natural communication as suggested in i+1. But on the other hand one should also ask if we are using CALL in ways that best approximate natural language interactions in order to approach the best acquisition principles presently known. Since the answer to this last question is a resounding negative, we c an see that the limited thinking, effort, and imagination that have gone into CALL programs has only aggravated the problem posed in the first question. The microprocessor hasn’t even been given a chance yet where branching communicative techniques are concerned.

Applications of Research

In keeping with the principles established by the above mentioned researchers it is important to build into the courseware frames, items and sequences that respond to the research findings (e.g. Krashen’s Acquisition Hypothesis). We know also today, for instance, that intensive listening periods are essential. Even if presented within the most minute steps, the target language may be misheard and mispronounced many times before it is ultimately acquired. Error and inaccuracies will occur until there has been sufficient comprehensive input + 1 to overcome the fixed patterns of one language system while adding to it all the new distributions of the phonemes, morphemes, and syntactic complexities of another.

Language acquisition, like with some other subjects such as music, first requires much listening and discrimination of contrasts. It is this discrimination of contrasts, then, that could subsequently be taught but always in a comprehensive context.

In more specific terms, we can state that certain phonemes in Spanish are similar to their counterparts in English and have a similar distribution. In such cases, less difficulty arises and, consequently less or perhaps only a minimum of practice, which will probably occur with fewer errors, is needed. Here, learning occurs by simple transfer of one item in one system into the other if the context is meaningful. It is well to note that this is not a transfer of a like system into another like situation but rather, a situation that the student understands and wants to know about. One problem with a simple transfer is that some phonemes are often identical, and even in closely related languages the distribution is often similar but seldom identical in all its positions. Similarity can sometimes be of great help by using it as a key to systematizing the target language and thereby motivating the student by lowering the affective filter of apprehension. For instance, a literate Portuguese speaker studying Spanish needs to know among other things that the Spanish intervocalic /s/ is always [s] like his own double ss or c or c+i or c+e.

Similarities between the target and native languages also present moderately difficult problems of distributional adjustment as witnessed by a different aspect of the above situation. A Spanish speaker does not have an intervocalic /z/ as the Portuguese speaker does and will need practice in meaningful and interesting contexts to learn to separate the Portuguese /s/, /z/ distribution.

Unfortunately, in most unrelated languages, there are a good many phonemes that are not only differently distributed but also have differing points and manners of articulation. Persistent difficulty of many types will plague the student in these cases until there has been enough comprehensive input + 1 so that the new phonology has become a part of the student’s nervous system. This comprehensive input in an interesting and non-threatening context can be repeatedly presented through graphics, sound, text and/or any combination of the three.

Computer Hardware

Computer hardware for ideal CALL courseware must take advantage of high quality visuals and sound. We know that today, for instance, the availability of hard discs or laser discs can give us better visuals for meaningful, comprehensive input and that computer graphics and keyboard interface are advanced enough for the elimination of the paper puller for teaching writing skills. The most exciting advances. Of course, as mentioned above, will come from the computer’s new ability to record, disassemble, and reassemble speech as recorded from the outset in digital form. This new ability allows for the comparative study of the newest expert systems and how we can best relate them to the advanced branching techniques which are emerging from research.

Since all the advancements of the preceding decade concerning second language acquisition emphasize a communication based approach, it is necessary that high quality sound be a major component of the student terminals. Further, in order to maintain the linguistic principle of primary of communication, a branch program of spoken language should be the first desideratum.

Following is a list of pertinent hardware and a brief introduction to some of the factors that need to be considered in choosing hardware for your CALL courseware and curriculum. Digitizer/synthesizer speech board with microphone and speaker. This item is one of the central pieces to the successful delivery of the courseware and represents an important breakthrough where language instruction is concerned given that early and the limited attempts at AI (as built in Barrutia’s 1964-70 incipient expert system).3 There are basically two types of speech synthesis boards:

– Phoneme speech synthesis, which constructs words using a set of approximately 64 unique sounds

– Digitizer/synthesizer speech boards

The digitizer/synthesizer speech circuit boards record speech waveforms as digital signals which are then reconstructed to produce speech. It converts a voice message into a digital recording that can be stored on disk. A high fidelity, natural sounding reproduction of the speaker’s voice and intonation can then be generated from the computer file.

There are many different techniques currently being used to implement speech digitization. Some common speech synthesis techniques include linear predictive coding (LPC), pulse code modulation (PCM), delta pulse code modulation (DPCM), and adaptive delta pulse code modulation (ADPCM). The major advantages of this last technique are its excellent reproduction quality, moderate storage requirements, and the ability to easily and inexpensively implement digitizing as well as synthesizing circuits.

Boards using this technique are built by several companies. Choose the best board which will meet the minimum requirements of quality, price and adaptability to selected hardware, including microprocessor system and storage requirements. Quality and price of microphone and speaker are generally included as part of the digitizer/synthesizer board.

Microcomputer Chip. The choice of a microcomputer chip depends on various factors: Any one particular chip is already available in the marketplace as part of a complete computer system, e.g., Intel 8088 used in the IBM PC and compatible machines, the Motorola 68000 used in the Apple Macintosh, Lisa Fortune Systems and others, the Intel 80186 used in the MAD-1 and Tandy computers, the Zilog Z80 used in 8-bit computers, which although older are cheaper, the Synertek 6205 used in Apple IIe computers, etc. Each system has its own capabilities, advantages and disadvantages ranging from price to graphics capabilities to memory addressability. All these factors should be studied in order to choose the mot suitable system to support the courseware software. The choice of a monitor or graphics terminal should depend on several factors: price, graphics support required by courseware, and the support required by the system selected above.

RAM Memory (Random Access Memory). The size of main Memory available in the computer for processing, RAM, is an important component for two reasons. The first is that the quality of digitized voice depends heavily on th4e amount of memory used to digitize the voice, which in turn depends on the technique used for digitizing. Depending on the digitizer/synthesizer board selected, varying amounts of RAM may be required. Secondly, the same is true as regards RAM requirements for courseware. Both these factors and possibly others should be taken into consideration in selecting the optimal amount of memory (RAM) required. This selection in turn will affect the choice of the microcomputer chip and the computer system.

Floppy Disk Drives. Floppy disk drives are another important component of a computer system. The total amount of secondary storage required to store the digitizer phrases will depend on the courseware design. It is already known that the storage requirements will be substantially large. It’s recommended, at this writing, that the system include a combination of floppy disk drives and hard disks (Winchester disks). Floppy disk drives vary in capacities from 140K bytes for the Apple IIe to 2.5 Megabytes for IBM PC, Kaypro and other personal computers. The distribution of the courseware will require a flexible medium, therefore a floppy disk drive is a must.

Winchester hard disks. The larger magnetic storage requirements for the digitization of a whole courseware will certainly require the addition of a large capacity Winchester hard disk drive. The choices are varied: 10 megabyte, 20 megabyte, up to 140 megabyte or 300 megabyte.

An additional medium which is relatively new is the laser disk which can contain a much larger amount of information but which has its own inherent advantages and disadvantages.

Access times of these different disk storage devices vary as do their prices and capabilities.

Operating systems. There are several choices of operating systems: apple DOS for Apple II and IIe computers; PC-DOS and MS-DOS for the IBM PC and IBM PC compatible machines; CPM for 8-bit computers; CPM-86 for 16-bit computers; UNIX, and others. The most appropriate operating system should be chosen to support the hardware configuration required by the courseware (speech board, memory, storage, etc.) and the computer programming, including interactive requirements.

System software. These elements must include a high level system design which will contain the needed support:

To drive the IO speech board and to build phrases dynamically

To support the graphics requirements as specified in the courseware

To build the expert system data base which will contain the information required by the courseware to support dynamic changes in the branching instructions. This data base will also contain information to monitor the language acquisition progress of the students.

A Final Desideratum

In considering then, all of the combined components as presented in this brief paper. We should be careful not to be so impressed by the latest electronic gadget that we forget what it is we are trying to do. We need to think not in terms of grander hardware, but rather in terms of making the hardware conform to grander and more humanistic programs (Barrutia 1970, 361).

ENDNOTES

1. Probably the very best study of its kind on the subject to date is John H. Underwood’s Linguistics Computers and the Language Teacher: A Communicative Approach. This small volume is a must reading for anyone interested in working on any aspect of CALL.

2. With grants from the University of California, the Modern Language Association of America, and with the Ampex Corporation providing the hardware R&D, Barrutia was able in 1969-70 to provide a fully intrinsic branching course that was offered experimentally at UCI.

The proposed approach of branch programming using voice interaction with the learner has already been proven feasible through work done at UCI (Barrutia, 1970). At that time the voice reproduction was obtained through tape recordings which made branching difficult, resulting in a final hardware system that, though functional, proved to be cumbersome and difficult to maintain (impossible after the demise of the funding from Ampex Learning Division).

That self-instructional course produced either very good results or the opposite extreme. There was a significant lack of any sizable segment with only average results which is generally not the case in teacher-taught classes. This points up that the program either works well when properly attended to or fails completely with high absentee students. All the statistics refer to the experimental group as AID (Ampex Intrinsic Device).

Although the second quarter exam reflected mainly the writing skill, there was still strong evidence that the experimental group’s speaking ability exceeded that of the teacher-taught classes. To confirm or dispel this suspicion, separate written and oral tests were administered at the end of the third quarter. The results of these tests seem to bear out the original assumption in that the programmed students did equally well in writing and significantly higher in or al skill.

In order to continue with the final phase of Program production and the completion of the project’s second year (1969-70), further financial support has to be sought. For this we turned again to the MLA. With a generous gift from our sister association and additional help from the Regential Fund we were finally able to bring the project to a successful conclusion in the Spring Quarter of 1970.

Following are the empirical results of the comparative test scores for self-instructional student results versus traditionally taught students for the three quarters of the 1969-70 school year.

Results of final exam given to first year (third quarter) Spanish students of a teacher-taught class, compared with AID students at the same level:

Written

Test

Oral

Test

(180)

(100)

Control Group (17 students)

162.6

63.9

AID Group (5 students)

163.6

86.2

Average

162.9

68.3

Results of final exam given to first-year Spanish students (first quarter). Comparison is made between a class taught by a professor, one taught by a teaching assistant, and those students taught by AID:

Written

Test

Oral

Test

(180)

(50)

Professor (10 students)

101.7

37.5

T.A. (9 students)

88.2

31.8

AID (10 students)

80.3

32.1

Average

90.1

33.9

Results of final exam given to beginning Spanish students (first quarter). Comparison is made between the three groups with the addition of a new AID group of nine students:

Written

Test

Oral

Test

(180)

(50)

Professor (10 students)

101.7

37.5

T.A. (9 students)

88.2

31.8

AID (10 students)

96.2

43.3

Average

91.6

36.1

It is natural to ask, With results like these why was the project abandoned? In the Fall of 1970 the principal investigator (Barrutia) took on a two-year directorship assignment for the University of California in Mexico City, and the Ampex corp. closed its section of instructional devices R&D. Consequently, the program lost its momentum and leadership. After surviving for another full three quarters at CUE, the already cumbersome analog software became faulty and lack of maintenance left it in disuse.

3. Linguistic Theory of Language Learning As Related to Machine Teaching. Richard Barrutia, The Center for Curriculum Development Inc., 1969.

This is a precursor study of the field of computer assisted language learning as developed prior to the availability of modern microprocessors. It affords a perspective of CALL emphasizing branch programming and functional load of error analysis as a first and foremost consideration. It includes the use of sound, phonetic diagrams, an extensive array of slides with random order of presentation and limited decisions made by the machine as available through the diode logic circuitry of that period.

Today we can record the program digitally on a synthesizer board without the interface of an analog recording for either the input or the output functions. This represents an important breakthrough where language instruction is concerned given that the limited AI (as built in Barrutia’s 1964-70 incipient expert system) can now not only be reactivated but be dramatically improved. The plans are to upgrade that system by whatever insights, findings, or methods that continued study show will enhance the new approach.

Author’s Address

Richard Barrutia

Dept. of Spanish and Portuguese

University of California, Irvine

Irvine, California 92717

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s