|
IATEFL Poland A Journal for Teachers of English ISSN 1642-1027 Vol. 5, Issue 1 (January 2005) |
|
|
IATEFL Poland A Journal for Teachers of English ISSN 1642-1027 Vol. 5, Issue 1 (January 2005) |
|
|
Articles |
||||||||||||||||
| |
|
|
| Ÿ
"Pronunciation in EFL CALL" by Wlodzimierz
Sobkowiak Ÿ
"Language
Teaching and Culture: Australian Language and Culture on the Web" by Renata Setmajer-Chylinski PRONUNCIATION IN EFL CALL [1] by Wlodzimierz Sobkowiak Poznan, Poland Abstract State of the art in pronunciation-oriented EFL CALL is reviewed from the
pedagogical perspective. Discussion
touches upon CALL flexibility, coverage, declarative vs. procedural knowledge,
L1-sensitivity, multimedia employment and automatic speech recognition (ASR).
Six different CALL programs are briefly evaluated from these points of view: Fluency,
Pronunciation Power, Connected Speech, Better Accent, ISLE and Tell Me
More. Future promises and
challenges in speech-enabled EFL CALL are outlined, such as speech synthesis,
multimodality in man-machine communication and (speech-to-speech) machine
translation. 1. Introduction All three terms appearing in the title of
this text beg for definitions, explanations and discussion. They are in the mainstream of current
foreign language pedagogy, on the one hand, and they are all multiply
ambiguous, on the other. To illustrate
the former claim, one would only need to mention the heated dispute on the
standards of English-as-a-Foreign-Language (EFL) pronunciation teaching, which
has been raging over To start this discussion at this point,
however, would be to jeopardize the main aim of the paper by getting swamped in
layers upon layers of metalinguistic and methodological details and
distinctions. And the main aim of the paper is, after all, not to conduct
methodological discourse, but rather to present a critical snapshot of the
current state of art at the interface of the three areas listed in the
title. This interface area itself is
enormous and it tends to exhibit a breathtaking pace of innovation, mostly due
to the rapid development of computer technology (some authors prefer ICT –
Information and Communication Technology – but I will stick with the simpler
term). In this situation, I can
entertain no hope of ever managing to provide a comprehensive and completely
up-to-date account of the whole area.
The sampling I made is of necessity subjective and fragmentary. For example, mainly for reasons of space, I
am not going to venture into the exciting world of on-line Internet
pronunciation teaching and learning, even though distance education is among
the most fashionable themes in current foreign language pedagogy. The discussion will thus be restricted to
'localized' CALL, which could also be called 'off-line' or CD-ROM‑ (or
DVD‑) based. Even so dramatically
circumscribed, the area is still too large to treat representatively. Out of many potentially interesting issues I
will select only a few. Out of hundreds
of available CALL programs, I will present but a handful. Out of their many components and
functionalities, I will concentrate on those which I regard as central to my
theme. The organization of the remainder of this
text is as follows: first I will discuss some key issues in EFL CALL
pronunciation, then some relevant software will be presented and briefly
analyzed from the point of view of the preceding discussion, finally a rather
informal glimpse of the future will close the paper. 2. Computer-assisted pronunciation
teaching and learning 2.1. Flexibility CALL researchers have successfully argued
(e.g. Kaliski 1992, Warschauer 1996, Warschauer & Healey 1998, Kern &
Warschauer 2000) that one of the main strengths of CALL is its didactic
flexibility. Unlike some other
educational technologies which have been implemented in schools over the ages
(from blackboards to video, say), computers will fit any didactic approach,
method or technique, if used skillfully. Grammar-translation supporters may use them to expedite
translation practice from L2 to L1, for example, with machine-translation
software. Audio-lingualists will be
able to control their students' structure drilling with the computer in much
more sophisticated ways than they could in the traditional language
laboratory. Cognitivists will sit their
learners in front of adventure games, where they will have to navigate an
unknown territory using their linguistic competence. Communicatively-minded teachers may pit learners against each
other in a simulation game to make them negotiate meanings to reach agreed-upon
goals. This flexibility of CALL is true also on
phonetic ground. Practically all
multimedia programs presented below, though to varying degrees, can be
accommodated into different pronunciation-teaching philosophies. Specifically, both the 'know-that'
declarative knowledge component, and the 'know-how' procedural skill component
of phonetic competence can be appealed to in various ways, for example through
multiple-choice exercises and listen-repeat-compare tasks, respectively. Despite common belief,
pronunciation-oriented CALL has not given up on the old techniques in favour of the razzle-dazzle display of vibrant
hypermedia. Rather – quite wisely in my
opinion – it has accommodated the new technological achievements such as speech
recognition, for example, into a versatile framework of structures and
functionalities where each learner and teacher can find something to fit his
needs and preferences. 2.2. Coverage An issue related to CALL's flexibility is
its coverage. The classic core of
pronunciation training in the traditional syllabus includes segments,
suprasegmentals, fast speech phenomena, grapho-phonemics, accentual
variation. On the level of particular
textbooks, courses and materials there is enormous variation, of course, both
in choice and priority of these elements.
Communicative language learning, for example, brought with it the
preference for prosody in pronunciation teaching, with some courses actually
beginning from rhythm, stress, juncture and intonation. This is now changing, with the advent of the
post-communicative era in foreign language teaching. Regardless of fashions and vacillations, however, the canon is
reasonably well defined. Contemporary pronunciation-oriented CALL
is able to deliver instruction in all those canonical areas. There are programs concentrating mostly on individual
sounds of English, as well as those which cater predominantly for
suprasegmentals. Some would
specifically target natural (fast) speech, while others would proudly (and
politically-correctly) offer different accents from speakers of different skin
colour. This is not to say that a
particular piece of software will necessarily include the full phonetics
syllabus. Unlike with the
methodological flexibility of section 2.1., coverage of phonetic substance is
of necessity much more 'hard-wired' in the structure of the package. It will be obvious from the short software
presentations below that programs tend to specialize in certain sub-areas of
the pronunciation syllabus. What is
crucial, however, is that there is now no technological obstacle to using CALL
in any of the canonical components. 2.3. Declarative versus procedural
knowledge This dichotomy was mentioned above in
2.1. It appears to be among the most
fundamental distinctions in all foreign language teaching, including teaching
pronunciation. It captures the
intuitively rather obvious truth that in order to do anything one must have –
in varying proportions depending on the actual activity – both the
'theoretical' and 'practical' competence.
Unlike syntax and vocabulary, pronunciation in a foreign language has
traditionally been regarded as the exclusive province of the latter; hence
murderous drilling in the audio-lingual method and little explicit teaching in
the cognitive-communicative era. The
pendulum now appears to be swinging in the other direction, so that the
declarative, explicit, 'know-that' meta-competence is back in the picture, with
researchers trying to feed it into the process of phonetic proficiency building
(see, e.g. Dziubalska-Kołaczyk 2002).
In the academic context of EFL, for example, this means that the
so-called 'descriptive grammar' of English should be well integrated with the
practical phonetics syllabus, so that students practicing, say, the intricacies
of English obstruent voicing could fall back on their knowledge concerning
laryngeal excitation source-filter models, as well as 'external' evidence from
their L1 interference, speech errors, speech play, and the like. CALL supports both types of
knowledge. Most multimedia
pronunciation programs are not content to provide the learner with ample
opportunities to use his articulators, whether for simply recording
utterances or for actual simulated dialogues.
There is usually also explicit instruction concerning such matters as
correct articulation and voicing, keeping the right rhythm, varying the pitch
for intonation, using appropriate lexical and sentential stress patterns, and
the like. This instruction may appear
in many different forms in the program: as mini-lectures, glossaries,
multimedia presentations, help files, task prompts, error messages, and many
others. Some packages offer manual-like
functionalities which can be used more or less like traditional textbooks,
complete with comprehension questions and suggestions on further reading. 2.4. L1 sensitivity Where there is still a lot of room for
improvement is how CALL relates to the native tongue of the user. The sad truth is that in very many cases it
simply does not. There appear to be two
main reasons for this state of affairs, one commercial, the other
linguistic. The former has to do with
cost-effectiveness mostly: to prepare a large CALL package with all the
currently expected multimedia bells and whistles is an extremely expensive
undertaking, much more so than, say, a traditional course with manuals,
exercise books and audio tapes. The
large investment will only pay off if the package can be used on a global scale
with all thinkable L1 learners.
Investing in a multi-CD EFL course for Poles only, for example, is
hardly commercially viable. Exceptions
to this rule are small programs made locally or localized versions of the
leading packages made in the West. Even
these, however, seldom go beyond simply translating the metalanguage and
fitting local translations to the existing monolingual built-in dictionary. This appears to be due to the other reason
mentioned above. While we may know a
lot about L1 transfer and interference on the theoretical level, there are huge
lacunae of knowledge in the actual application of this information in the speech-enabled
CALL setting. For example, the CALL
craze of the last few years – speech recognition (see below) – has hardly
reached a stage where it would have a viable model of the learner with a given
L1, hence a particular interlanguage.
The technique is hastily transferred from native speaker applications
such as dictation or dialoguing expert systems into the world of EFL with
little recognition of the need to make it sensitive to non-native speech. Thus, what one often observes is either
disastrous recognition results with highly demotivating end-effects, or the
anything-goes principle where any learner input is happily accepted. Both these extremes are thoroughly
a-pedagogical, of course, as noticed by many researchers (e.g. Chen 2001). I can but agree with Ehsani and Knodt
(1998:56) at this juncture that "one of the most needed resources for
developing open response conversational CALL applications is large corpora of
non-native transcribed speech data, of both read and conversational speech". 2.5. The growth of multimedia 'Multimedia' is one of the modern
buzz-words, on a par with CD-ROM, SMS, hypertext, DVD, mp3, video-clip, and
dozens of others. Few young people in
the developed countries would be completely ignorant of it, and most would agree
that the term has positive connotations with novelty, movement, sound, colour,
fun, etc. It is these connotations that
are exploited in the contemporary saturation of CALL with multimedia. CALL, after all, is supposed to be
edutainment, it must motivate, it must attract. And what better attraction to a young mind can there be than a
full-colour video with good quality sound?
Most current CALL packages are built on this premise (one reason why
they are expensive to make). Multimedia has grown gradually. First, (still) pictures were added to sheer
text, then sound of initially rather poor quality, then simple animations,
finally video. Pronunciation-oriented
CALL jumped on each band-wagon soon after they appeared. In the age of CGAs and Hercules graphics no
phonetic transcription (or accented letters, for that matter) could be shown on
screen, so simplified systems had to be used[3]. With the first graphics showing articulators
in cross-section, vowel diagrams or lip shapes added to text CALL resembled
good old pronunciation manuals. With
the advent of animation and audio the true era of multimedia began and
pronunciation-CALL finally got its added value. Sounds could now be illustrated in various media: in
transcription, articulatory diagram or recording. Finally, with the improvements of sound cards, sound recording
algorithms, processor speed and memory, sounds could also be visualized as
waveforms or (for the more intrepid learners) as spectrograms. It is this last idea that has become
enormously popular among the designers of CALL software and among the less
critical users. It is usually
implemented in the context of the listen-record-compare task: the learner
listens to the model (native-speaker) recording, records his own rendition of
the text trying to mimic the original closely, finally compares the two
recordings along both channels: aural and visual (example screen-dumps appear
below, e.g. Figure
12). The former – aural comparison
– is the traditional method used since the very inception of foreign language
learning. The latter is new (motivating
element!): the eye is supposed to guide the tongue, to put it
metaphorically. While the technique
obviously has the required commercial potential, there are serious pedagogical
objections to its use (see e.g. Ehsani & Knodt 1998). To enumerate them briefly: (a) no two recordings, even of the same
person, are exactly acoustically alike, (b) no instruction is normally provided on
how to align the two waveforms for best result, (c) speech tempo and loudness will
interfere with the correct reading of the waveform information, (d) considerable acoustic knowledge
and skill are needed to be able to benefit from comparing waveforms, (e) the technique is very sensitive to
hardware quality. This is hardly the end of multimedia
development, of course. New ideas and
technologies hit the news lines weekly.
Some of them have serious ramifications for the future of
pronunciation-aware CALL. These will be
briefly presented at the end of this paper in the 'science fiction' section. 2.6. Automatic Speech Recognition (ASR) It was at the beginning of the last decade
of the 20th century that computer hardware reached a stage where it
could support speech recognition in real time.
This was of course an enormous breakthrough in computer-human
communication, which had so far been multimedial in the output, but monomedial
(keyboard) in the input (see Aist 1999 for an overview and literature). As mentioned above, ASR was immediately
implemented in a number of consumer applications, starting with those where
single-word input was sufficient, and gradually spreading to other functionalities,
such as dictation, for example. To
fully appreciate how complex computer speech recognition is one would have to
have a large aside here explaining the intricacies of Hidden Markov Models
(HMMs), Fast Fourier Transforms (FFTs), Dynamic Time Warping, and a bunch of
other algorithms verging on higher math.
The reader is referred to some introductory sources, such as Deroo,
Bernstein & Franco 1996 or Cole & Zue 1997. Here, we will briefly concentrate on CALL applications of ASR. Notice first that the term itself may be a
misnomer if it is applied to such functionalities as the mentioned waveform
display, for example, or the facility which tries to automatically assess the
acoustic fit of the learner-recorded speech with the model (as some of the
programs analyzed in section 4. do). No
true recognition is required in either process, rather acoustic pattern
matching, which is technologically much simpler. For fully-fledged ASR to take place, the software must
meaningfully react to the content of the input speech. This meaningful reaction may simply be
visualizing speech as computer-readable text on screen, or adjusting the flow
of the simulated dialogue according to learner input, for example. This is the current stage which ASR has
reached in (the more advanced) CALL, as illustrated below (see Tell Me More
in 3.6.). Notice that the dialogue is simulated,
i.e. the learner can control the flow of the conversation within limits imposed
by the ASR algorithm, which will only allow the choice of a few communicatively
relevant responses to the computer's recording (a low-perplexity
closed-response task for the ASR engine).
With no true AI (Artificial Intelligence) in view, the simulation can
proceed no further: any attempt to enter into natural conversation with the
machine quickly ends in linguistic-communicative disaster, as clearly
illustrated by the many 'chatterbots', i.e. chatting robot-like agents on the
web (see selection of links on www.chatterbot.tk,
for example). ASU (Automatic Speech Understanding),
unlike ASR, is still a long way off. With all these caveats, ASR did give CALL
an added measure of face validity. To
be able to speak to the machine in the foreign language and have it
react in meaningful ways is certainly exciting to most learners, especially to
the new generation of children, who take the 'traditional' modes of keyboard
communication for granted. Also the
speech assessment routines now built into some ASR-equipped CALL can initially
be quite thrilling. From them the learner
will get the extra metalinguistic feedback on his pronunciation, as if from the
teacher. This may initially motivate
the learner to actually try harder and pronounce the given sentence again,
hoping to push the match indicator to an even higher level. This repetition is no bad thing, of course,
as far as it goes. Learners quickly
discover, however, that the mechanism can hardly offer robust and detailed
evaluation and guidance, as a teacher would: all that can be expected is a
global yes/no measure of phonetic achievement, with completely mysterious
evaluation criteria and no explanation whatsoever[4]. With the user-customized acceptance
threshold set to low, this functionality will accept virtually any spoken input
and grade it as good; with high settings even native speakers will have
problems getting satisfactory notes.
Thus, somewhat analogically to the visual feedback routines discussed
earlier, the technique is highly pedagogically questionable at this stage of
ASR development (see Chen 2001 for similar conclusions). 3. Pronunciation-oriented CALL software: a
sample Just like the selection of issues in EFL
CALL pronunciation in section 2 of this paper, so the choice of software for
presentation and analysis is unavoidably fragmentary and subjective. With the line of CALL merchandise now
running into hundreds, it could hardly be otherwise. Few comprehensive overviews of (EFL) CALL products exist, also
because it is almost impossible to keep pace with the growth of the field: new
packages hit the market every week.
Finally, serious CALL journals, can only carry a few reviews in every
issue, thus sampling a tiny proportion of the whole market. In this situation, the only feasible
resource for CALL software information, analysis and advice is the
Internet. And indeed, there is CALL
info galore on the web, with the characteristically unavoidable disadvantage of
uncertain reliability and expertise, volatility, advertising hype, selectiveness,
repetition and dispersal. With the
practical unavailability of other sources, however, the web remains the best
place to go for CALL software research. In what follows I will briefly present six
pronunciation-oriented EFL CALL programs, relating the discussion to the issues
discussed in section 2. The programs
are listed in the table. As can be seen from the table, despite the
tiny size of the sample, the packages do represent a certain variety along the
four dimensions of (a) visual and (b) corrective feedback, (c) transcription
use, and (d) general didactic focus.
Each one has some www presence in the form of their dedicated
websites. They are enjoying a certain
amount of critical reviewing attention, as well. The following is not to be construed as actual reviews of the
programs, however, as explained earlier. 3.1. Fluency According to the blurb on the Fluency's
website, "Fluency is designed to let you speak, then give you
feedback as to how you did – what to correct and how to correct
it. Using state-of-the-art speech recognition technology, SPHINX from Carnegie
Mellon University, this interactive software allows you to speak, to get
corrections, to listen to yourself and a native speaker and try again, over and
over, as many times as you want".
The website is that of the Language Technologies Institute (http://www.lti.cs.cmu.edu/index.html),
which does most of its research in machine translation and information
retrieval, and Fluency appears to be an offshoot of that research. Going a little deeper into the LTI website, we
will encounter a slightly more precise description of the Fluency's
operation: "The system detects pronunciation errors, such as duration
mistakes and incorrect phones, and offers visual and aural suggestions as to
how to correct them". As seen in the screenshot in Figure 1, the
program's functionalities are rather modest, exactly like stated above. The sagittal cross-section of the vocal
tract and the frontal lip view are now practically standard features of EFL
(pronunciation-oriented) CALL software (sporting better graphics in most
cases). The obscure non-IPA phonetic
transcription system, on the other hand is definitely non-standard (even if
used in practically all CMU Natural Language Processing (NLP)
applications). The segment illustrated,
i.e. AX is of course a schwa. Unlike
most EFL CALL programs, Fluency does try to evaluate and correct the
pronunciation of single segments, but the user would be hard-pressed to find
out something substantial about the criteria the software is using in the
process, so self-correction remains a hit-and-miss procedure. There is a lot of the
declarative-knowledge material in the program: the transcription, the diagrams,
the corrective advice are all phrased in the notorious 'descriptive grammar'
parlance. The speaking skill component
is restricted to simply reading a reply in the simulated dialogue. The learner can also listen to the model
utterance, of course, but this would hardly count as a communicative activity,
say practicing listening comprehension, because there is effectively none. Neither is there any pretense of a
communicative setting, of course, with the short dialogue highly stylized and
unnatural. 3.2. Pronunciation Power Pronunciation Power is marketed by English Computerized Learning Inc. (ECL), which,
according to their website, "was founded in 1995 in Edmonton, Alberta,
Canada [and] operates as a developer and distributor of professional high
quality, interactive, multimedia ESL/EFL CALL materials". The package was developed, however, by
Blackstone Multimedia Corporation, which is a privately held company also based
in Pronunciation
Power is designed for
over a dozen L1s. It includes a variety
of game-like activities and functionalities, such as (a) listen-record-compare,
with single words, minimal pairs and full sentences, (b) listening
discrimination, (c) vocal-tract cross-section and lip animations (Figure 2), (d)
waveform display and comparison (Figure 3), (e)
STAIR exercises: stress, timing, articulation, intonation, rhythm, (f) a 40-page
manual, (g) on-board illustrated audio dictionary of over 7,000 words with a
variety of search keys. Thus, it is a
large and fully professional package, entirely devoted to pronunciation
training, enjoying a considerable commercial success. With the multitude
and variety of material and activities it offers, Pronunciation Power is
suitable for pronunciation teaching and learning under different conditions and
in different settings, thus eminently proving that flexibility is a great asset
of CALL. Its coverage, in terms of the
canonical pronunciation/phonetics syllabus is large, although fast-speech
phenomena and accentual differences of English appear to be
underrepresented. The
declarative-procedural balance is much better than in Fluency, with a lot
traditional textual exposition on the one hand (see Figure 4, for
example, where heavily technical terminology is used to annotate the diagram),
but also the extremely rich skill component containing varied activities for
practicing both listening and speaking.
In going through the demo of the package, which is freely dowloadable
from its website, I did not notice any L1-sensitivity beyond the localization
of the dictionary. This is, as
discussed above, the sad norm in native-made EFL CALL. While the waveform matching technique is
used to its limit (with the learner being able to drag-align his recording to
better fit the model; see Figure 3), no
attempt is made to automatically evaluate the learner's pronunciation. Considering the criticism I voiced earlier,
this is probably to the good of the package.
While the system of phonetic transcription used is thoroughly simplified
compared to IPA, it is still generally IPA-ish, with ashes, schwas, engs,
thetas, and the like[5].
ASR is used for most exercise types, so that there is relatively little
keyboard input. 3.3. Connected Speech Connected
Speech was made by an
Australian firm Protea Textware. Like Pronunciation
Power, it is an integrated CALL package exhibiting professional design and
programming, with noticeably less content, versatility and multimedia
interaction, however. From the entry
screen (see Figure
5), the learner can be taken to one of three proficiency levels (starting
with lower intermediate) in one of the accentual versions, American, Australian
or British English, placed on separate disks.
There, he should first listen to a few minutes of video-recorded
narrative monologue, whereupon he can enter a suite of tasks and exercises
mostly focused on "the suprasegmental features of English, with mark up,
recording, practice activities, tests
and tutorials. It has speech
recognition that gives specific feedback on the suprasegmental features of the learner's production" (from http://www.proteatextware.com.au/). As can be seen in Figure 6,
separate components are dedicated to such phonetic areas as pause groups,
stress, pitch change, linking, segmentals and syllables. More traditional exercises are also
included, such as listening comprehension tests prompting the learner to fill
in critical vocabulary items from the keyboard, or IPA training. The program does
appear to focus on connected speech, which makes it unique among those under
scrutiny here, although the monologues are rather far from the native norm of
"natural informal", even at the advanced level, the web blurb
notwithstanding. The pros of Connected
Speech are: (a) the good balance between the declarative and
procedural element, (b) the wide variety of voices, tempos and accents, (c) the
skillful use of ASR on all phonetic levels, i.e. for segmentals, pitch, stress
and duration, and (d) a well-designed, uncluttered and intuitive user
interface. Among the cons one should
certainly mention: (a) the rather disappointing use of video to record 'talking
heads' only[6], (b) complete L1 insensitivity, and (c) the
generally rather uninspired design of the tasks and exercises. 3.4. Better
Accent Tutor American-made Better Accent Tutor is "pronunciation training
software based on instant audio-visual feedback of intonation, stress and
rhythm". This is indeed the
primary focus of the package: suprasegmentals.
The decision to circumscribe the content area so narrowly is supported
on the website of the package (http://www.betteraccent.com/papers/quotes_on_pronunciation.htm)
with quotes from a number of experts in the field, such as Joan Morley,
Marianne Celce-Muria, Joanne Kenworthy, even Alexander Graham Bell! This is very much in the spirit of the
communicative language teaching approach, whereby communication is supposed to
be maximized even at the expense of (phonetic) correctness. It is claimed that ill-pronounced individual
segments will rarely hamper mutual understanding as much as incorrect
prosodies: hence the emphasis on the latter. The curriculum covered by the Tutor includes: word stress, simple
statements, wh-questions, general questions, repeated questions, alternative
questions, tag questions, commands, exclamations, direct address, series of
items, long phrases, tongue twisters.
The approach is heavily 'know-that'-oriented, with a lot of 'explanation'
carried out in rather dense phonetic jargon; all of these features illustrated
in Figure
7. ASR is used to display the
learner's pitch (intonation) and intensity (loudness) graph alongside the model
ones for visual inspection and comparison.
No automatic evaluation is attempted. With such a narrow focus and modest content, the package must compromise
flexibility of application, of course.
Notice also that the multimedia technology does not reach beyond audio
play-back and input, with no graphical animation or video. Likewise, there is no phonetic transcription,
articulatory diagrams, waveform comparison or traditional phonetic exercises
(cloze, dictation, multiple choice, and the like). Thus, the package projects a rather austere image, as also
transpires from the screenshot in Figure 7. 3.5. ISLE Unlike all the CALL packages so far discussed, the Interactive Spoken
Language Education (ISLE) was not a privately owned commercial venture, but
rather a multinational (German-Italian-British) project, running between 1998
and 2000, heavily subsidized by EC funds, and coordinated by Educational
Concepts, the R&D department of Ernst Klett Verlag. Three of the partners were universities of According to the website, "The main objective of ISLE is to provide
technical solutions to support training of spoken language communication. This
will be achieved by developing computer based tools to support the training of
speaking skills and by integrating such tools into existing multimedia-based
language teaching software systems".
The deliverable, however, whose demo can be freely downloaded from the
Internet, is a stand-alone package apparently targeted at the busy manager in
whom "the social climate of a classroom can easily produce psychological
barriers to the training of elementary speaking skills in a foreign
language". This can be seen even
in the names of the entry screen components, such as "Travel
Arrangements", "At the Airport", "In the Office",
"At the Hotel" or "In the Restaurant". The content of the whole package, including
dialogues, exercises and glossaries, is also adjusted accordingly. For example, we have a Paolo Rossetti
arranging for his business trip to After entering the program the learner must first calibrate the ASR
engine by reading a story of the conquest of The "Oral Exercises" section offers read-and-repeat,
listen-and-repeat, Q&A, Build the Sentence and Free Choice. The first of these is illustrated in Figure 8, the
last but one – in Figure
9. What is of particular interest
here is the 'improve' option: when the program decides that the learner
mispronounced some sounds (no suprasegmental practice in this package), it
offers advice and provides corrective practice, as shown in Figure 10, first
kindly asking the learner "How strict should I be?". Upon testing, it turns out – somewhat
expectedly – that with the 'strict' setting there is no way to make the ASR
wizard satisfied. There is – again not
unexpectedly – no guidance on how best to approach the recorded model; it is
all the matter of hit-and-miss. As can be seen from this short description of the functionalities of the
package, as well as from the screenshots, despite claims to the effect that the
program focuses on natural communication, the tasks are rather traditional,
with repetition galore, comprehension questions and phonetic drills. While the learner can listen to a
conversation conducted in a natural setting, he cannot himself engage in
one in any form. And why should he ever
need 'tanks' and 'ants' in the business context (see Figure 10)? The ASR evaluation of sentences is characteristically unreliable, and
the segmental ASR-assisted practice – tedious and unhelpful. The overall balance is in favour of
procedural knowledge, with close to no explanation, no (phonetics) manual and
no phonetic terminology. IPA
transcription is used (sparingly) in the 'improve' menus. No technological gimmicks with waveform
display and adjustment are in sight.
Generally, with no animation or video movement on the screen, the
impression is that it is very traditionally rendered, despite the use of
ASR. This, in turn, leads to the guess
that the EC funding was not adequate to elaborate the content and function of
the program any further. 3.6. Tell
Me More Finally, a CALL
package where ASR technology has been used most effectively: to actually
simulate a spoken dialogue between the learner and the computer. Auralog's Tell Me More is heavily advertised
on the web as "the reference in foreign language learning, developing all
linguistic skills: oral and written expression, comprehension, grammar and
vocabulary". As far as
pronunciation is concerned, it boasts "the exclusive S.E.T.S. technology
(Spoken Error Tracking System) automatically detecting errors in
pronunciation" as well as "3D phonetic animations" (see Figure 11). The program is sold in nine different
language (L2) versions, including both British and American English, and three
proficiency levels. There are also
networked modalities of the software, with functionalities allowing teacher
control and class management as well as student-teacher messaging and other
tools. In what follows, however, I will
describe the 'traditional' CD-ROM-based package. As mentioned
above, in Tell Me More the ASR technology is pushed to its current
limits: (a) waveform display, (b) pitch tracking (see Figure 12),
(c) learner input evaluation, (d) dialogue simulation. The latter proceeds by the program offering
the learner a few printed options to read off the screen in response to the
computer-initiated contextualized dialogue in an authentic setting, e.g. travel
arrangements (Figure
13). The ASR engine tries to figure
out which option was actually spoken, and reacts accordingly by responding to
this user input. While this is far
from an actual conversation, of course, the technique is reasonably robust and
very motivating: the learner at last feels that what he says will change
the following flow of communicative interchange. As mentioned above (2.6.), to achieve more along this path, ASR
would have to feed into a functional AI component with L1 sensitivity and
learner modeling. Such packages will
not be available for... some time to come. Tell Me More is definitely balanced towards procedural
knowledge, with heavy emphasis on pronunciation (speech communication),
although it is hard to make blanket statements for this package which appears
on the market in so many different versions: proficiency-, L2-,
learner-group-wise (there are dedicated business courses, for example). Even the title of the whole series changed
over time, from Talk to Me in 1997; some older versions of the package
are still available under this title.
Unlike Pronunciation Power and Connected Speech, there is
thus much less formal exposition of matters phonetic, no structural division of
the program into phonetic fields such as segmentals, stress, intonation, and no
phonetic terminology. In these respects
Tell Me More resembles ISLE: speech communication in a naturalistic
setting is at the centre of the package.
Unlike in ISLE, however, the ASR engine does not attempt to
identify specific errors in the learner speech input; rather the assessment is
global. The acceptance level can be
adjusted by the learner himself, with all the disadvantages described above
(too strict or too lenient). The
program is lavishly illustrated with good quality photos and videos (in newer
versions) and it shows all signs of professional graphics design. While the semi-transparent animation of the
articulators (what is called "3D phonetic animations" on the web) may
be little more than a gimmick at this stage of human-computer interaction, it
does show us a glimpse of things to come in terms of educational applications
of virtual reality (see Baldi below). 4. The
future of pronunciation-oriented (EFL) CALL Automatic Speech
Recognition (ASR) is by far not the last word in human-computer interface
design. As mentioned in the
introduction, the pace of technological innovation in computer technology
generally, and in natural language processing (NLP) in particular, is
breathtaking. In this last part of the
paper I can only try to briefly speculate about the impact on (EFL) CALL of
some recent inventions and developments.
Like before, the selection is of course heavily subjective, but – it is
hoped – not quite irrelevant for the discussion above. 4.1. Text-to-Speech Synthesis One area where the
impact of technology on CALL is going to be felt soon is that of speech
synthesis. Text-to-speech (TTS)
synthesis by rule, whereby no previous human recording is necessary in any
form, has reached human-like quality (cf. e.g. Dutoit 1997 and 1999). The high-end TTS engines are rather
expensive now, and research to improve especially the prosodic properties of
synthesized speech are still under way, but the technology is now reaching the
stage where it can be applied to CALL, as the synthesized speech can actually
function as a model of pronunciation, as well as in the now trite capacity of
information deliverer[7]. The added bonus,
compared to pre-recorded human speech, is that it is under stringent control of
the designer in terms of practically all phonetic criteria: pitch (responsible for
the impression of gender), tempo, intonation, timbre of voice, accent,
loudness, etc. It would be technically
rather easy to simulate a foreign accent, if need arose, for example to better
demonstrate to the learner the areas which need improvement (e.g. final
devoicing in Polish English). Keller
& Zellner-Keller (2000a) note that "speech synthesis allows [...] the
creation of sound examples that could not be produced by a human being (e.g.,
speech with intonation, but no rhythm)". Because TTS
engines are tiny compared to audio recordings, the CD space recovered could be
used for other multimedia components of CALL, such as video, for example (see
below for video synthesis). And,
naturally, speech synthesis is by far less expensive than recording a team of
highly trained human speakers. 4.2. Face
animation One of the most
successful applications of cutting-edge computer technology to CALL has been
the University of Colorado Center for Spoken Language Understanding (CSLU)
"Baldi" project (http://cslr.colorado.edu/toolkit/main.html). In brief, it is an NLP environment focused
on the use of TTS synthesis and ASR enhanced with the animated face
("Baldi", Figure 14 )
simulating phonetically realistic articulatory movements in real time. Visual object programming, speech
spectrography and many other components are integrated in the Rapid Application
Developer which makes it possible to create a simple dialogue schema in
minutes, which can then be build into another application, such as CALL for
example (see http://www.haskins.yale.edu/haskins/heads.html
for a comprehensive interactive overview of many other 'talking head'
projects). What is most
exciting in the package (which is free for educational purposes) is the novelty
of using the animated face to enhance speech synthesis and make the spoken
exchange more realistic. Baldi not only
moves his lips and eyes to provide the much needed – especially in the context
of learning a foreign language – visual information to aid
intelligibility. It can also 'go
transparent', exposing the realistically rendered inner articulators in full
motion, down to the root of the tongue (see Figure 15). This is an incredible resource for
pronunciation learners, of course: they can listen to natural (if synthesized)
speech and see how it originates in the mouth.
The head is quasi-3D; it can be rotated in all three dimensions with the
mouse, and the amount of transparency can also be adjusted at will, the extreme
leaving just the articulators on screen. The CSLU toolkit,
where Baldi lives, has so far been used mostly to assist speech and language
therapy of native American children, but its application to EFL CALL (and other
L1's – Baldi can be programmed for any language whatsoever) is just a matter of
time. Also, it is enough to go to the
movies nowadays to see the level of realism which animation of human-like
synthetic actors has achieved (e.g. "Shrek" or "Lord of the
Rings"; see also Thalmann & Thalmann 1990). In a few short years animated anthropomorphic agents will be used
in CALL which will be hard to tell apart from video-recorded real human
speakers. One technical consequence of
this will be – like with the TTS synthesis – that more CD space will be freed
from the enormously memory-hungry current video files. It is much harder to predict learner
reactions to (semi-intelligent) speaking and animated human-like agents acting
as conversation partners in settings which are now only available in video
conferencing. Learners may relate to
these artificial personas to the extent which may be pedagogically relevant,
with both its pros and cons. 4.3.
Multimodal man-machine communication For truly
multimodal human-machine interaction the machine would have to progress beyond
simple (?) ASR – into the realm of automatic recognition of audiovisual
speech. The AI TTSS ASR agent would be
able to recognize and act upon (at least) the facial expressions of the
computer user. This would not only aid
communication generally through taking advantage of gaze, eye-brow movement, head
positioning, and the like, but also – in the context of pronunciation teaching
– make it possible to provide additional articulatory feedback to the learner
concerning his lip position and movement in labial(ized) sounds, tongue-tip
control in apico-dental fricatives or labio-dental contact in /f/ or /v/. Of course, to achieve this level of video
sensitivity highly sophisticated systems would have to be employed. Contemporary prototypes are nowhere near the
required technological stage (see Figure 16
for a simple example). The area is full
of vibrant research activity, however (see e.g. Granström, House & Karlsson
2002, or Scott 2001 for an accessible introduction), and the feeling is that we
can expect significant breakthroughs quite soon. 4.4. Machine Translation (MT) Machine
translation might at first sight appear not to belong here, in a paper where
pronunciation-oriented (EFL) CALL is discussed. It is indeed true that MT is seldom used in this context, even
though one can envisage its creative applications in a grammar class, for
example, where learners would try to induce the rules of the foreign language
from the (usually risible) characteristic errors of an MT package. However, the impact of speech-enabled, or
Speech-to-Speech (StS) MT, once it is perfected, can be enormous, not only for
the business of pronunciation instruction, but for the whole world of foreign
language teaching (FLT) and learning.
In the words of Crystal (2001:227): "We can also envisage the
translating telephone, where we speak into a phone, and the software carries
out the required speech recognition, translation, and speech synthesis,
enabling the listeners to hear our speech in their own language [...] Such a
world is, of course, a very long way off". It is enough to have a look at Ectaco's web pages (http://www.ectaco.com/)
to appreciate that the envisaged world has already arrived: while the Russian
company's translator is still rather primitive (in terms of device size, range
of languages translated, vocabulary size, noise robustness, etc.) it does
translate speech to speech in real time with quality quite adequate for a
tourist or businessman ordering a meal in a restaurant or air tickets in a
travel agency. Thus, I believe,
Crystal's "way-off" nightmare is more immediately threatening than he
ever thought: "in a world where it is possible to translate automatically
from any one language into any other, we have to face up to the issue of
whether people will be bothered to learn foreign language at all" (ibidem;
see also Cribb 2000 for similar conclusions). 5.
Conclusions With the currently fashionable 'focus on form'
in foreign language teaching (EFL in particular; see, for example, Doughty &
Williams 1998 or Ellis 2001)
the role of pronunciation-oriented CALL software is bound to grow in the
process of phonetics teaching and learning, both in the classroom context and
outside it. Teachers will delegate onto
computers some of the more tedious tasks involved, such as drilling as one
technique of skill-getting. It seems
that the unconstrained smoothly-flowing spoken foreign-language dialogue with
the computer will not become reality for some time to come. So, until computer AI improves
significantly, truly communicative activities will not be used as a vehicle for
practicing pronunciation. But at least
there is a chance that the employment of (EFL-aware) speech recognition,
text-to-speech synthesis and certain elements of artificial intelligence will
gradually transform the boring phonetic 'drill-and-kill' procedure into an
exciting, multimedia, interactive 'drill-and-thrill' adventure. The audio channel of communication between
human and machine, which is now opening, both in FLT and outside it, is an
additional boost for the growing anthropomorphization of the computer. After all, with which other creatures,
natural or artificial, can we communicate by voice? The computer will unavoidably grow its own persona around
it. The FLT learner will take it more
and more for granted that he can intelligently communicate with this persona in
the foreign language. He will react to
it more and more on the affective level, as well as intellectually. He will like it, or hate it, as the case may
be. He will look to it for help,
advice, praise and criticism. He will
count on its inherent intelligence and wisdom.
What impact this attitude will have on foreign language learning and
teaching remains to be discovered. My
guess is that it will be enormous. Notes 1.
This text is based on my lecture to
the 2. In this
context a multitude of books (e.g. Jenkins 2000), articles (e.g. Sobkowiak
2003) and conferences (e.g. LM34's workshop on LFC: http://elex.amu.edu.pl/ifa/plm/2003/index.htm)
might be invoked. 3.
Somewhat paradoxically, as early as at the beginning of 1980s Sinclair's ZX
Spectrum could (with pains) flash user-designed IPA on the TV screen, which
functioned as its VDU (Video Display Unit). 4. This
may not be true of some speech assessment and practice software mostly used in
the (native-language) clinical setting.
This is, however, usually rather narrowly targeted at, say vowel
quality, whereby the user is trying to match vowel formant positions in
two-dimensional diagrams with the model ones. 5. The
shape of TH is somewhat nonstandard, though: /đ/. 6. Even if
"experiments have shown that a visual display of the talker improves not
only word identification accuracy [...], but also speech rhythm and
timing" (Ehsani & Knodt 1998:52). 7. See Filoglossia, a CALL package with Greek as a foreign
language, which already employs TTS synthesis: http://www.ilsp.gr/filoglossia_plus_eng.html,
or WordPilot from http://www.compulang.com,
which also has this feature. Bibliography Aist, G. 1999. "Speech recognition in
computer-assisted language learning". In K.Cameron (ed.). 1999. 165-81. Asher, R.E. & J.M.Y.Simpson (eds). 1994. The
encyclopedia of language and linguistics. Bernstein, J. & H. Franco. 1995. "Speech recognition by computer". In N.Lass (ed.). 1995.
408-34. Cameron, K. (ed.). 1999. CALL: media, design and
applications. Amsterdam: Swets & Zeitlinger. Carey, M. 1998. "But is it interactive and does
it work? A review of some CALL English pronunciation CD-ROMs". In In
T.Ottmann & I.Tomek (eds).1998. Chen, Hao-Jan H. 2001. "Evaluating five speech
recognition programs for ESL learners". Paper presented at ITMELT'2001,
Hong-Kong, Cole, R. & V. Zue. 1996. "Spoken Language
Input". In: R. Cole et al. (eds). 1996. Survey of the state of the art
in human language technology. [http://cslu.cse.ogi.edu/HLTsurvey/ch1node2.html#Chapter1;
last accessed 15.12.2004] Cribb, V.M. 2000. "Machine translation: the
alternative for the 21st century?". TESOL Quarterly 34.3, 560-9. Delcloque, P. (ed.). 2000. Proceedings of InSTIL:
Integrating Speech Technology in Learning. Deroo, O. A short introduction to speech
recognition. [http://www.babeltech.com/download/SpeechRecoIntro.pdf;
last accessed 15.12.2004]. Derwing, T.M., M.J. Munro & M. Carbonaro. 2000.
"Does popular speech recognition software work with ESL speech?" TESOL
Quarterly 34.3, 592-603. Doughty, C. & J. Williams. 1998. Focus on form
in classroom second language acquisition. Dutoit, T. 1997. An introduction to text-to-speech
synthesis. Dutoit, T. 1999. A Short Introduction to
Text-to-Speech Synthesis. [http://tcts.fpms.ac.be/synthesis/introtts.html;
last accessed 15.12.2004] Ehsani, F. &
E. Knodt. 1998. "Speech technology in
computer-aided language learning: strengths and limitations of a new CALL
paradigm". Language Learning & Technology 2.1, 45-60. Ellis, R. (ed.). 2001. Form-focused instruction and
second language learning. Eskenazi, M. 1999. "Using automatic speech
processing for foreign language pronunciation tutoring: some issues and a
prototype". Language Learning & Technology 2.2, 62-76. Fotos, S. (ed.). 1996. Multimedia language teaching. Granström, B., D. House & I. Karlsson (eds). 2002.
Multimodality in language and speech systems. Hiller, S.M. & E.J. Rooney. 1994. "Teaching
pronunciation with computer aids". In R.E. Asher & J.M.Y. Simpson
(eds). 1994. 4528-35. Hunyadi, L. et al. (eds). 1998. ALLC/ACH'98
conference abstracts. Jenkins, J. 2000. The phonology of English as an
international language. Kaliski, T. 1992. "Computer-assisted language learning
(CALL)". In P.Roach (ed.). 1992. 97-109. Kaltenboek, G. 2001. "A multimedia approach to
suprasegmentals: using a CD-ROM for English intonation teaching". In J.A.
Maidment & E. Estebas-Vilaplana (eds). 2001. 19-22. Keller, E. & B. Zellner-Keller. 2000a. "Speech synthesis in language learning:
challenges and opportunities". In P. Delcloque (ed.). 2000. Keller, E. & B. Zellner-Keller. 2000b. "New uses for speech synthesis". The
Phonetician 81. 35-40. Kern, R. & M. Warschauer. 2000. "Theory and
practice of network-based language teaching". In M. Warschauer & R. Kern (eds). 2000. 1-19. Lambacher, S.G. 1996. "Spectrograph analysis as a
tool in developing L2 pronunciation skills". In M. Vaughan-Rees (ed.).
1996. 32-35. Lass, N. (ed.). 1995. Principles of experimental
phonetics. Lecumberri, G., M. Cooke & J.A. Maidment. 2001.
"Automatic feedback on phonemic transcription". In J.A. Maidment
& E. Estebas-Vilaplana (eds). 2001. 15-18. Maidment, J.A. & E. Estebas-Vilaplana (eds). 2001.
Proceedings of the Phonetics Teaching and Learning Conference, Nagy, T., P.
Furkó & A. Tóth. 1998. "Computers
and multimedia applications in teaching pronunciation". In L.Hunyadi et
al. (eds). 1998.187. Ottmann, T. & I. Tomek (eds). 1998. Proceedings
of ED-MEDIA/ED-TELECOM 98. Pellegrino, F., C. Fressard & G. Puech. 2001.
"Teaching phonetics with a multimodal Internet site". In J.A.
Maidment & E. Estebas-Vilaplana (eds). 2001. 31-34. Pennington, M.C. 1999. "Computer-aided
pronunciation pedagogy: promise, limitations, directions". CALL 12.5.
427-40. Roach, P. (ed.). 1992. Computing in linguistics and
phonetics. Rudnik, O. 1998. ASR in EFL CALL. Scott, N. 2001. "Intelligent user
interfaces". Information Impacts Magazine, March 2001. [http://www.smartnet.co.nz/res-articles/intelligent-interfaces.htm;
last accessed 15.12.2004] Sobkowiak, W. 2003. "Why not LFC?". In W. Sobkowiak
& E. Waniek-Klimczak (eds). 2003. Dydaktyka fonetyki języka obcego.
Zeszyty Naukowe Państwowej Wyższej Szkoły Zawodowej w Koninie. Proceedings of the Wąsosze Conference (http://www.pwsz.konin.edu.pl/ins-neo.lm?tresc=ins-neo/instytut_neofilologii_wasosze2.html)
on teaching foreign pronunciation, 10-12.5.2002. [English translation here: http://elex.amu.edu.pl/~sobkow/wasosob4.doc] Thalmann, N. & D. Thalmann. 1990. Synthetic actors in computer-generated 3D films. Vaughan-Rees, M. (ed.). 1996. Speak out: newsletter
of the IATEFL Pronunciation Special Interest Group (PronSIG). Warschauer, M. 1996. "Computer-assisted language
learning: an introduction". In S.Fotos (ed.).
1996. 3-20. Warschauer, M. & D. Healey. 1998. "Computers and language learning: an overview". Language Teaching 31. 57-71. Warschauer, M. & R. Kern (eds). 2000. Network-based language teaching. Concepts and practice. LANGUAGE TEACHING
AND CULTURE: AUSTRALIAN LANGUAGE AND CULTURE ON THE WEB by Renata Setmajer-Chylinski
"Culture and communication are inseparable because culture not only dictates who talks to whom, about what, and how the communication proceeds, it also helps to determine how people encode messages, the meanings they have for messages, and the conditions and circumstances under which various messages may or may not be sent, noticed, or interpreted... Culture...is the fou | ||||||||||||
