New Ideas in Software Development for Linguistics and Language Learning

 

Vance Stevens
Steve Spurling
Don Loritz,
Roger Kenner
John Esling
Moya Brennan

Abstract:
Language teachers and linguists have been developing their own computer-assisted language learning (CALL) software for use in research and language learning. This report describes several such programs which meet a need for communicative, inductively-based software: that are controlled by the students using them; and that can he used as tools in research as well as having applications in the classroom.

Perhaps the greatest hindrance to widespread use and early acceptance of computers in language learning has been the limited availability of commercial software and the difficulty it has had in meeting the varied specific needs of a wide range of language teachers. Early offerings in the educational software market have often been aimed at the lowest common denominator, and the high price of much software has also restricted its use. A number of linguists and language instructors have been responding to the need for programs that suit their own purposes by creating their own software. What is unique about such software is that it is not created primarily for its marketability, but rather in order to meet a professional or pedagogical need not addressed in available commercial courseware. A sample of this type of software is presented here as an indication of the kinds of programs that several language learning professionals have found useful and practical.

The contributors to this article have either authored or been involved in developing one set of programs described here and have implemented them in instructional settings. The authors intend that the ideas presented in this report might serve as examples or points of departure for other software developers, and thereby suggest ways in which computers can be used in language education as an extension to methods found in currently available commercial material.

1. Gerunds and Infinitives with Stop, Remember, Regret, and Forget

commercial software publishers are often and understandably reluctant to market software that relies of peripheral devices which users would first have to purchase in order to run the program. Hofstetter (1985) takes the position that greater use of peripherals, such as touch screens and joysticks, allows students to focus all their attention on the screen without having to divert attention away to the keyboard. Although touch screens have been developed for some microcomputers, and joysticks and game paddles are common in arcade games, the commercial sector has been slow to capitalize on such devices for education.

Using Apple PILOT, Vance Stevens has created a lesson to teach choice of complement (infinitive or gerund) after the matrix verbs stop, remember, regret, and forget. After any of these verbs, either a gerund or infinitive complement is grammatically possible, but meaning changes depending on which is chosen. These choices have been observed to pose problems for ESL students.

The lesson inductively teaches appropriate choice of gerund or infinitive with the above verbs in a contextualized situation. In the lesson, a character named Max has attended a party the night before. During the course of the lesson, it becomes apparent that Max may have exhibited either normal or perhaps rude and offensive behavior at the party. For example, he possibly forgot to thank the hostess, or perhaps he just forgot thanking her. If Max was not entirely sober the whole time, perhaps this was because he stopped to drink at 3:00 a.m. (when he should have stopped drinking). During the course of the evening, he may have either stopped to talk to a pretty girl or stopped talking to her; the latter being the wisest choice since her husband turns out to be a boxer. In any case, choice of scenario is up to the student, and there are 32 possible scenarios that students may pursue by selecting verb combinations from a chart.

By manipulating the game paddles, students form sentences from a chart of possible sentence components. The chart appears on the screen with the four matrix verbs listed down the first column, and four complements listed in the next two columns: gerunds in the middle column and infinitives in the last. Students manipulate game paddle knobs (or joystick) to make cross hairs meet over their selections, and then push buttons on the game paddles to indicate their selection to the computer. The resulting sentence appears at the bottom of the monitor screen along with a paraphrase or elaboration of the semantic import of that sentence, followed eventually by questions as to which happened first, the action in the complement or that in the matrix verb.

The lesson is conceived as an attempt to create a lively learning environment for student users by bringing a variety of perspectives (as per Moore & Anderson, 1969), senses (via graphics), and cognitive mechanisms (via the story, puzzle and final recapitulation formats) to bear on the language learning process. The lesson is also designed in an effort to help students become involved in their own learning. It is possible for students to slip through many drill and practice exercises without really engaging themselves mentally (Howe & DuBoulay, 1979), while with this lesson students have been observed to ponder moves at length while trying to predict the outcome.

Mastery of the lesson was tested empirically, and the results suggest that this design is most effective with low-intermediate ESL students (Stevens, 1984a; 1984b).

2. L'Accord du Participe Passe (Past Participle Agreement)

Another disappointing aspect of much widely marketed software is the lack of meaningful feedback provided to the student. User input is often limited to a single keystroke or, where a word or words are allowed, these tend to be judged either right or completely wrong, with little information given as to the reason for the mismatch. Language learners may require more informative feedback; for example, pattern mark-up with proofreader's marks to indicate problems with the answer, as available on PLATO (Hart, 1981) or with DASHER (Pusack, 1982). Pattern mark-up, however, does not provide feedback in grammatical terms where students have been exposed to those terms in the target language.

One CALL program, written for the IBM PC by Lea Penney and programmed by Roger Kenner, aims at tutoring the complexity of past participle agreement in French using a sophisticated system of feedback along lines succinctly described by Chapelle & Jamieson (1983). The system used here, while not actually parsing the students' responses, can produce a detailed grammatical analysis of the answer. For example, the lesson can produce the following diagnostic comments:

— The past participle appears incorrect.

— The agreement appears incorrect.

— The auxiliary verb appears incorrect.

— The pronoun is incorrect.

— The pronoun is wrongly placed.

— The negation is incorrect.

— You do not have enough words.

These comments, of course, appear in French, as does the rest of the material in the lesson. (English translations of key points are available to the student on request.) Learning is thus taking place at two levels. Students are getting practice in monitoring this difficult aspect of the French verb system; plus they are entirely immersed in a target language environment in a manner that focuses their attention on the activity rather than on the language of interaction.

Another issue in CALL is the degree of student control allowed. One occasionally finds commercial software which seems to be based on the use of the computer as a regulator rather than a facilitator of learning. In the Penney & Kenner lessons, progress through the material is completely under the students' control. They can enter any activity or can move backward or forward at will. Students having difficulty with a concept are invited to review the rules and examples for that concept. Once into the grammar section, students can page around at will. At any point, they can decide to try some exercises, and they will be shifted to exercises pertinent to the grammar section they have just left.

3. IPA Tutor, AWFE, Miss Fidditch, and Henry Higgins: Computational Linguistics and CALL

In a different vein from the foregoing, Don Loritz has written four programs designed to explore the application of algorithms originally developed in the field of computational linguistics. The first program converts text to IPA and vocalizes the text via speech synthesis; the second program converts spoken text to its graphic waveform representation; the third program allows a student to try to emulate the waveform graph of a model speaker; and the fourth program uses LISP to parse sentences in English and to deliver a syntactic analysis.

The programs all require an Apple II+ or Ile (DOS 3.3) and a single disk drive, plus additional peripherals and memory as noted below.

Ipa Tutor: The simplest of the four programs, lpa Tutor, redefines the Apple keyboard to produce IPA characters on the high-resolution screen, and produces the corresponding sounds through a speech synthesizer. The user is prompted to enter free text (appearing in IPA) above a graphics-map display of the IPA keyboard. At <RETURN>, the ASCII codes in a GETLIN buffer are mapped onto the phoneme codes of a Votrax SC-01 speech synthesizer chip (on a board manufactured by John Bell Engineering), and IPA input is synthesized as speech. Ipa Tutor has been used in introductory linguistics classes and has proved a painless (indeed, even enjoyable) way to learn IPA. Only English can by synthesized, and then only to an approximation of native human English, largely because of the ethnocentric and idiosyncratic "phoneme" inventory of the SC-01 chip. Synthesis quality is, however, adequate to train students in quite narrow "transcription."

AWFE: the Acoustic Waveform Editor: AWFE functions as a storage oscilloscope and as such is primarily a research utility: it digitally records speech input and displays the speech waveform on the high-resolution graphics screen. Cursors can be directed about the screen to make precise timing measurements on the waveform. For example, VOT can be measured to within one millisecond. Additionally, segments of the waveform can be digitally spliced out of the waveform, saved, and/or spliced back into the waveform. In this way, quasi-synthetic speech stimuli such as VOT continua can be created from natural speech.

AWFE uses the Mountain Computer ADA (analog-digital-analog) board, although any similar product with an A-D conversion speed faster than 100 microseconds per sample can be used. AWFE requires the computer to be interfaced to standard audio input-output equipment (e.g., a cassette recorder) through an ADA converter.

Henry Higgins: Henry Higgins is a "biofeedback" intonation tutor. It requires 64K of RAM (i.e., an Apple II+ and a 16K language card). Like AWFE, it also requires an ADA interfaced to standard audio I/O equipment. It operates as follows:

First, a teacher digitally records and saves a set of model sentences as a pronunciation exercise. A standard Apple diskette can accommodate approximately 15 2.4-second sentences. Upon entry to Henry Higgins, the student receives standard CAI instructions, and then hears the first model sentence played back. Simultaneously, the intonation contour is mathematically extracted and displayed on the screen immediately below the model contour. The student can repeat the exercise until his or her contour satisfactorily matches the model contour(s). Student responses can be saved to disk.

In pilot testing, it is clear that many native English speakers have difficulty imitating the intonation contours of English models; hence, learners might need to be trained to use Henry Higgins. Henry Higgins can be easily reconfigured to accept live rather than pre-recorded model sentences, thus serving as an enhancement to classroom pronunciation activities. Poor native speaker mimics might also be poor language learners—at least with respect to spoken accent ability. Thus, adaptations of Henry Higgins might assist language learners if they do nothing more than train them to be better mimics. Most promising, however, is the observation that the poor intonation mimics are nevertheless good timing mimics, which implies that the pedagogic value of Henry Higgins could be improved by abstracting timing information from the basic analysis and highlighting it. Further research in this area at least seems warranted.

Miss Fidditch: Miss Fidditch is a prototype syntactic, error-detecting, augmented transition network parser. it is written in LISP and so requires a LISP interpreter, such as P-LISP.

In theory, Miss Fidditch prompts the user to create and input any English sentence. It detects and identifies errors of agreement and of word order. In practice, it is constrained, principally by the 64K available in the Apple Ile, to simple sentences with a lexicon of a hundred words. In its present (unoptimized) implementation, it is also rather slow, requiring something like 40 seconds to parse an eight-word sentence. It is, however, capable of quite complex analyses of simple sentences, as well as of modest semantic analyses.

Although more complete grammar parsers will require machines with larger memories than the Apple II, Miss Fidditch demonstrates that within the restricted domain of a particular language lesson or unit, simple microcomputers already have sufficient processing capacity to allow students to engage in computer-supervised creative construction.

4. Visible Speech

Along similar lines to the programs described in sections 3.2 and 3.3, Visible Speech is a software package designed originally by Craig Dickson and Roy Snell to provide visual feedback on pitch, amplitude, and voicing in real time for hearing-impaired children; and applied in a language laboratory setting for ESL and linguistics students by John Esling. Original applications are described in Dickson, Ingram & Snell (1984). Visible Speech is implemented on an Apple II+ or Ile through a peripheral interface board, and requires two disk drives. Pitch contours representing tone or intonation are displayed in real time, from microphone or pre-recorded input, as waveforms across the upper half or lower half of the screen. Either trace may be redrawn, in any order, in a variable time frame of 1, 2, or 4 seconds, and saved to disk. Alternatively, the upper trace can be stored as a model while the lower trace is repeatedly redrawn in real time for student practice, using the <SPACE BAR> as trigger. Amplitude and voiced/voiceless displays also appear in split-screen mode for modeling or practice, in the time frame of 1, 2, or 4 seconds. These displays are helpful for observing syllable stress patterns or rhythmic patterns found in instructional ESL materials such as Jazz Chants (Graham, 1978).

Theoretically, the pedagogical applications of Visible Speech can be viewed from either an "input" or "output" perspective (Esling, Warbey, & Scholefield, 1984). Output theory focuses on imitation, repetition and memorization, while input theory emphasizes exposure to large amounts of language, often complex, but whose meaning (intention) is understood. The format of Visible Speech conforms superficially to output theory in the sense that learners are given small amounts of modeled language to practice. In this respect, the present version of this program is perhaps best suited to the linguistic training of teachers and advanced students as an analytical/descriptive tool. There are also students of all levels who prefer this more controlled approach. There is no reason, however, why such a device cannot be used to provide input, of itself and through user interaction. From an input perspective (Smith, 1984), presenting pronunciation material to beginner or intermediate ESL students would entail collecting large inventories of examples for display, comparison and manipulation in interactive group activities on the computer. Access to large amounts of data, quickly, is exactly what the computer facilitates.

Examination of pitch contours of contrasting languages and of English dialect varieties has indicated that subtle details of pronunciation can be isolated and brought to students' awareness using Visible Speech. Applications where students wish to practice styles of delivery by observing their voice response characteristics have also been requested. Recent enhancements to the program include the addition of speech storage (digitally encoded) for random-access replay.

5. ANALIT

One educational task that can be greatly enhanced by the use of computers is test development. Statistical analysis points out malfunctioning test items in a way our linguistic intuitions cannot, and computers (properly programmed) are particularly good at statistical analysis. Therefore, anyone serious about reliable and valid test instruments would find his or her work facilitated by computer-assisted test analysis. Charged with the construction of a new admissions test by the director of the ESL program at the College of Marin, Steve Spurling was dissatisfied with the item and test analysis program in use by the data processing department. Therefore, he produced a microcomputerized item and test analysis program that rivals mainframe programs, but which is easier to use and more appropriate to the needs of a small department. Called ANALIT, the program is implemented on an Osborne computer.

ANALIT provides a number of test and item statistics; for example, means, standard deviations, histograms, and correlations. Traditional item statistics such as reliabilities, item-by-option breakdowns, point biserial correlations, and p values are also provided. There is also a module to handle Rasch item analysis. The program produces a person-by-item matrix filled with mean square fits for unexpected responses (Wright & Stone, 1979, p. 74). At the far right of each row of the matrix is each person's raw score, percent correct, the associated Rasch ability measure and the mean square fit statistic, with misfits flagged. Under each matrix column is each item's score in terms of the number of examinees who answered it correctly, the percent of the group they represent, the Rasch item difficulty measure, and the mean square fit statistic with misfits once again flagged. While these statistics may seem esoteric to the uninitiated, to the test developer they provide a wealth of information.

The section of the program used most at the College of Marin is the standardized score placement module. In this module, the two multiple choice sections of the college's admission test are entered item by item. Then the total score for the writing section is entered along with each examinee's name. After all examinees have been entered in this fashion, the placement criteria (cutoff scores and class names) are called up from a separate file and the group is automatically placed into classes with placement listed to either the screen or a printer. The option exists of listing students to separate class files on disk. The examiner can then go back and print out class lists for each instructor as well as an overall placement list. These facilities also exist in a raw score placement module where the same functions can be performed on raw rather than standardized scores.

The bottom line is this: ANALIT makes a difficult and time-consuming task less difficult and less time consuming. It removes one of the chief barriers to the development of optimal language tests--the gap between intuition and analysis. ANALIT is one of a new generation of productivity tools for educators; tools which should dramatically alter what is possible for educators to accomplish given constraints on time and resources.

6. Interactive Video: CALL Using a Computer/Video Interface

Interactive Video is one of the most exciting areas of development in CALL, albeit one not yet ripe for wholesale commercial exploitation. Interactive video combines the advantages of the learning media of the video cassette (or videodisc) and computer-assisted learning, essentially by means of an interface card. The computer controls the video and guides the student in offering options, allowing input, and reacting to that input. Depending on the student's response, the computer will offer appropriate material (i.e., text or video) for a given level of understanding, ideally taking into account different learning styles. The program should not only guide each student to the best exercises for his or her level, but (in being able to search automatically a tape or videodisc) should also allow a student to go to a particular section of video without having to manipulate the machine.

Truly innovative interactive video implementations require relatively generous institutional backing. Reports of interactive video projects inspired by teachers wanting simply to try out the medium are rare, but often insightful (see Stevens, 1983, for example). Accordingly, an interactive video project undertaking at Hong Kong Polytechnic by Moya Brennan, Eveline Caldwell, Tim Lockyer, and Eddie Tse, is of interest in that it shows a use to which teachers have elected to put this medium when given the opportunity.

In the resulting experimental lesson, a televised panel discussion was used. The teaching materials aimed to help the students to note the form of introductions, to get the gist of arguments, to become sensitive to nonverbal cues, and to interpret the strength of various statements in the discussion. A sequence of presentation was suggested, but the students could choose exercises from the menu and plan their own sequences of instruction.

The materials were programmed in BASIC, which allowed the use of graphics and variety in exercise types. Subsequently, a McGraw Hill authoring system was used with the same materials and, although lesson creation was much less time consuming, the result was less stimulating, visually and intellectually.

The group found that interactive video can be used with groups, but that it also offers effective instructional possibilities for those who cannot attend classes, who prefer a self-paced, private learning experience, or who feel they need more individual help than they receive in regular classes. But the medium is only as good as the materials which are available, and production of materials for interactive video in particular requires an altered perspective on the part of the language teacher, as well as the cooperation of a computer specialist with an awareness of language learning theory.

7. Summary

The software described here represents the developments undertaken by a particular group of language teachers, all of whom have made some effort to apply computers to their own professional situations, as an alternative to currently available commercial courseware. These materials reflect a search for alternatives to the discrete point drill and practice software often associated with computer-assisted language learning. Furthermore, these programs reflect a desire on the part of some language teachers to have available to them more responsive and inductively-based software which is largely under student control, and a desire on the part of the linguists to have access to software which can be used as a research tool as well as have applications in the classroom. It is hoped that by publicizing this work, further efforts may be inspired on the part of practicing language teachers toward the development of even more responsive CALL software.