Language Engineering and the Pathway to Healthcare: A user-oriented view
Harold Somers
School of Informatics
University of Manchester
PO Box 88
Manchester M61 0QD, England
Harold.Somers@manchester.ac.uk
Abstract
This position paper looks critically at a
number of aspects of current research into
spoken language translation (SLT) in the
medical domain. We first discuss the user
profile for medical SLT, criticizing de-
signs which assume that the doctor will
necessarily need or want to control the
technology. If patients are to be users on
an equal standing, more attention must be
paid to usability issues. We focus briefly
on the issue of feedback in SLT systems,
pointing out the difficulties of relying on
text-based paraphrases. We consider the
delicate issue of evaluating medical SLT
systems, noting that some of the standard
and much-used evaluation techniques for
all aspects of the SLT chain might not be
suitable for use with real users, even if
they are role-playing. Finally, we discuss
the idea that the “pathway to healthcare”
involves much more than a face-to-face in-
terview with a medical professional, and
that different technologies including but
not restricted to SLT will be appropriate
along this pathway.
1 Introduction
The doctor–patient consultation is a central element
of the “pathway to healthcare”, and with language
problems recognised as the single most significant
barrier on this pathway, spoken-language translation
(SLT) of doctor–patient dialogues is an obvious and
timely and attractive application of language tech-
nology. As Bouillon et al. (2005) state, the task
is both useful and manageable, particularly as inter-
actions are highly constrained, and the domain can
be divided into smaller domains based on symptom
types. In this position paper, we wish to discuss a
number of aspects of this research area, and suggest
that we should broaden our horizons to look beyond
the central doctor–patient consultation to consider
the variety of interactions on the pathway to health-
care, and beyond the confines of SLT as an appropri-
ate technology for patient–provider communication.
In particular we want to stress the importance of
the users – both practitioners and patients – in the
design, especially considering computer- and con-
ventional literacy. We will argue that the pathway to
healthcare involves a range of communicative activ-
ities requiring different language skills and implying
different technologies, not restricted to SLT. We will
comment on the different situations which have been
targeted by research in this field so far, and the im-
pact of different target languages on research, and
how the differing avilability of resources and soft-
ware influences research. We also need to consider
more carefully the design of the feedback and verifi-
cation elements of systems, and the need for realistic
evaluations.
2 Who are the users?
We start by looking at the assumed profile of users
of medical SLT systems. Systems that have been
developed so far can be divided into those for use in
the doctors office – notably, MedSLT (Rayner and
Bouillon, 2002), CCLINC (Lee et al., 2002), and
(honourable mention) the early work done at CMU
(Tomita et al., 1988)1 – and those for use for first
contact with medical professionals “in the field”, de-
veloped under DARPA’s CAST programme:2 MAS-
TOR (Zhou et al., 2004), Speechalator (Waibel et
al., 2003), Transonics (Narayanan et al., 2004) and
SRI’s system (Precoda et al., 2004). This distinction
mainly motivates differences in hardware, overall
design, and coverage, but there may be other more
subtle differences that result especially from the sit-
uation in which it was envisaged that the CAST sys-
tems would be used.
Some descriptions of the systems talk of “doc-
tors” and “patients” though others do use more in-
clusive terms such as “medical professional”. A sig-
nificant common factor in the descriptions of the
systems seems to be that it is the doctor who con-
trols the device. This may be because it can only
handle one-way translation, as is the case of Med-
SLT, “. . . the dialogue can be mostly initiated by the
doctor, with the patient giving only non-verbal re-
sponses” (Bouillon et al., 2005), or may be an ex-
plicit design decision:
There is, however, an assymmetry in the
dialogue management in control, given the
desire for the English-speaking doctor to
be in control of the device and the primary
“director” of the dialog. (Ettelaie et al.,
2005, 89) [emphasis added]
It is understandable that as a regular user, the
medical professional may eventually have more fa-
miliarity with the system, but this should be re-
flected in there being different user-interfaces (see
Somers and Lovel 2003). We find regrettable how-
ever the assumption that “the English speaker [. . . ]
is expected to have greater technological familiar-
ity” (Precoda et al., 2004, 9) or that
the medical care-giver will maintain the
initiative in the dialogue, will have sole
access to the controls and display of the
translation device, and will operate the
1We give here one indicative reference for each system.
2Formerly known as Babylon. See www.darpa.mil/ipto/ pro-
grams/cast/.
push-to-talk controls for both him or her-
self and the [P]ersian patient. (Narayanan
et al., 2004, 101)
In fact, although the early use of computers in
doctor–patient consultations was seen as a threat,
more recently the help of computers to increase
communication and rapport has begun to be recog-
nised (Mitchell and Sullivan, 2001). This may be at
the expense of patient-initiated activities however,
and many practitioners are suspicious of the nega-
tive impact of technology on relationships with pa-
tients, especially inasmuch as it increases the per-
ceived power imbalance in the relationship.
Figure 1, a snapshot from Transonics demo,3
leaves in no doubt who is in control.
Figure 1: Snapshot from Transonics’ demo movie.
The patient is not even allowed to see the screen!
Equipment whose use and “ownership” can be
equally shared between the participants goes some
way to redressing the perceived power-balance in
the consultation. We have evidence of this effect in
ongoing experiments comparing (non-speech) com-
munication aids on laptops and tablet PCs: with the
laptop, controlled by a mouse or mouse-pad, the
practitioner tends to take the initiative, while with
the tablet, which comes with a stylus, the patient
takes the lead. Bouillon et al. (2005) comment that
“patients [. . . ] will in general have had no previ-
ous exposure to speech recognition technology, and
may be reluctant to try it.” On the other hand, pa-
tients also have suffered from failed consultations
3http://sail.usc.edu/transonics/demo/transedit02lr.mov
which break down through inability to communi-
cate, and in our experience are pleased to be in-
volved in experiments to find alternatives. In our
view, one should not underestimate patients’ adapt-
ability, or their potential as users of technology on
an equal status with the practitioners.
This being the case, we feel that some effort needs
to be devoted to usability issues. We will return to
this below, but note that text-based interfaces are not
appropriate for users with limited literacy (which
may be due to low levels of education, visual im-
pairment, or indeed the lack of a written standard for
the language). Use of images and icons also needs
to be evaluated for appropriateness, an issue not ad-
dressed in any of the reports on research in medical
SLT that we have read. For example, Bouillon et al.
(2005) show a screenshot which includes the graphic
reproduced in Figure 2. The text suggests that the
user (i.e. the doctor?) can click on the picture to
set the topic domain. It is not clear why a graphic is
more suitable for the doctor-user than a drop-down
text menu; there is no mention of whether the patient
is encouraged to use the diagram, but if so one won-
ders for what purpose, and if it is the best choice of
graphic. Research (e.g. by Costantini et al. 2002)
suggests that multimodal interfaces are superior to
speech-only systems, so there is some scope for ex-
ploration here.
Figure 2: Graphic taken from screenshot in Bouillon
et al. (2005)
Incorporating more symbolic graphics into an in-
terface is an area of complexity, as Johnson et al.
(2006) report. Iconic text-free symbols, for exam-
ple to represent “please repeat”, or “next question”,
or abstract concepts such as “very” are not always
as instantly understandable as some designers think.
Considering the use of symbols from AAC (augmen-
tative and alternative communication) designed for
speech-impaired disabled users by patients with lim-
ited English, we noticed that AAC symbol sets have
a systematic iconicity that regular users learn, but
which may be opaque to first-time (or one-time) un-
trained users (Johnson, 2004).
3 Feedback and verification
Translation accuracy is of course crucial in the med-
ical domain, and sometimes problematic even with
human interpreters, if not trained properly (Flores,
2005). Both speech recognition (SR) and translation
are potential sources of error in the SLT chain, so it
is normal and necessary to incorporate in SLT sys-
tems the provision of feedback and verification for
users. The standard method for SR is textual repre-
sentation, often in the form of a list of choices, for
example as in Figure 3, from Precoda et al. (2004).
Figure 3: Choice of recognizer outputs, from Pre-
coda et al. (2004:10)
For translation output, some form of paraphrase
or back-translation is offered, often facilitated by the
Figure 4: Choice of recognizer outputs, from Precoda et al. (2004:10)
particular design of the machine translation (MT)
component (e.g. use of an interlingua representa-
tion, as in MedSLT, Speechalator). In the Transonics
system, the SR accuracy is automatically assessed
by the MT component: SR output that conforms to
the expectations of the MT systems grammar is pre-
ferred.
For the literate English-speaking user, this ap-
proach seems reasonable, although an interface such
as the one shown in Figure 4, detailing the output of
the parse must be of limited utility to a doctor with
no linguistics training, and we must assume that the
prototype is designed more for the developers’ ben-
efit than for the end-users.
For the patient with limited or no English, the is-
sue of feedback and verification is much more diffi-
cult. As mentioned above, and reiterated by Precoda
et al. (2004), the user may not be (wholly) liter-
ate, or indeed the language (or dialect) may not have
an established writing system. For some languages,
displaying text in the native orthography may be an
added burden. Figure 5 shows Speechalator’s Ara-
bic input screen (Waibel et al., 2003). It is acknowl-
edged that the users must “know something about
the operation of the machine”, and although it is
stated that the display uses the writing system of
the language to be recognised, in the illustration the
Arabic is shown in transcription.
Another issue concerns the ease with which a lay
user can make any sense of a task in which they
are asked to judge a number of paraphrases, some
ungrammatical. This is an intellectual task that is
difficult for someone with limited education or no
experience of linguistic “games”. For example, for
this reason we have rejected the use of semantically
unpredictable sentences (SUS) (Benoˆıt et al., 1996)
in our attempts to evaluate Somali speech synthesis
(Somers et al., 2006). This leads us to a considera-
tion of how medical SLT can best be evaluated.
4 Evaluation
MT evaluation is notoriously difficult, and SLT eval-
uation even more so. Most researchers agree that
measures of translation fidelity in comparison with a
gold-standard translation, as seen in text MT evalu-
ation, are largely irrelevant: a task-based evaluation
is more appropriate. In the case of medical SLT this
presumably means simulating the typical situation
that the technology will be used in, which involves
patients with medical problems seeking assistance.
Since SLT is a pipeline technology, the individ-
ual components could be evaluated separately, and
indeed the effects of the contributing technologies
assessed (cf. Somers and Sugita 2003). Once again,
literacy issues will cloud any evaluation of speech
recognition accuracy that relies on its speech-to-text
function, and evaluation of speech synthesis must
simulate a realistic task (cf. comments on SUS,
above).
Evaluations that have been reported suggest us-
ing real medical professionals and actors playing
the part of patients: this scenario is well established
in the medical world, where “standardized patients”
(SPs) – actors trained to behave like patients – have
been used since the 1960s. One problem with SPs
for systems handling “low density” languages like
Persian, Pashto and so on, is the need for the vol-
Figure 5: Speechalator’s Arabic input screen
(Waibel et al., 2003, 372)
unteers to understand English so that they can be
trained as an SP, in conflict with the need for them
to not understand English in order to give the sys-
tem a realistic test. Ettelaie et al. (2005) for exam-
ple report that their evaluation was somewhat com-
promised by the fact that two of their patient role-
players did speak some English, while a third partic-
ipant did not adequately understand what they were
supposed to do.
Another problem is that there is no obvious base-
line against which evaluations can be assessed. One
could set up “with and without” trials, and mea-
sure how much and how accurately information was
elicited in either mode. But this would be a waste of
effort: it is widely, although anecdotally, reported
that when patients with limited English arrive for
a consultation where no provision for interpretation
has been made, the consultations simply halt. It is
also reported, as already mentioned, that human in-
terpreters are not 100% reliable (Flores, 2005). Of-
ten, an untrained interpreter is used, whether a fam-
ily member or friend that the patient has brought
with them, or even another health-seeker who hap-
pens to be sitting in the waiting room. The potential
for an unreliably interpreted consultation (or worse)
is massive.
Ettelaie et al. (2005) mention a number of metrics
that were used in their evaluation, but unfortunately
do not have space for a full discussion. The principle
metric is task completion, but they also mention an
evaluation of a scripted dialogue, with translations
evaluated against model translations using a modi-
fied version of BLEU, and SR evaluated with word-
error rate. These do not seem to me to be extremely
valuable evaluation techniques.
Starlander et al. (2005) report an evaluation in
which the translations were judged for acceptability
by native speakers. Given the goal-based nature of
the task, rating for intelligibility rather than accept-
ability might have been more appropriate, though it
is widely understood that the two features are closely
related. On the positive side, Starlander et al. used
only a three-point rating (“good”, “ok” or “bad”):
evaluations of other target languages might be sub-
ject to the problem, reported by Johnson et al. (in
prep.) and by ADD REF that rating scales are highly
culture-dependent, so that for example Somali par-
ticipants in an evaluation of the suitability of sym-
bols in doctor–patient communication mostly used
only points 1 and 7 of a 7-point scale.
Another evaluation method4 is to assess the num-
ber and type of translation or interpretation errors
made, including whether there was any potential or
actual error of clinical consequence.
As Starlander et al. (2005) say:
In the long-term, the real question we
would like to answer when evaluating the
prototype is whether this system is practi-
cally useful for doctors
to which we can only add, reiterating our comments
in Section 2, “. . . and for patients”.
5 The Pathway to Healthcare
Let us move on finally to a more wide-ranging is-
sue. “Medical SLT” is often assumed to focus on
doctorpatient consultations or, as we have seen in
4Thanks to the anonymous reviewer for pointing this out.
the case of systems developed under the CAST pro-
gramme, interactions between medical professionals
and affected persons in the field. Away from that
scenario, although it is natural to think of “going to
the doctor” as involving chiefly an interview with a
doctor, and while everything in medical practice ar-
guably derives from this consultation, the pathway
to healthcare in normal circumstances involves sev-
eral other processes, all of which involve language-
based encounters that present a barrier to patients
with limited English. None of the medical SLT sys-
tems that have been reported in the literature address
this variety of scenarios, although the website for the
Phraselator (which is of course not an SLT system as
such) does list a number of different scenes, such as
the front desk, labour ward and so on.
In this section, we would like to survey the path-
way to healthcare, and note the range of language
technologies – not always speech or translation ori-
ented – that might be appropriate at any point. The
purpose of this is both to make a plea to widen our
vision of what “medical SLT” covers, but also to
note that SLT is not necessarily the most appropri-
ate technology in every case.
The pathway might begin with a person sus-
pecting that there may be something wrong with
them. Many people nowadays would in this situa-
tion first try to find out something about their con-
dition on their own, typically on the Web, though
of course there is still a major “digital divide” for
racial and ethnic minorities, and the poor, partly
due to the langauge barriers this research is address-
ing. If you need this information in your own lan-
guage, and you have limited literacy skills, tech-
nologies implied are multilingual information ex-
traction. MT perhaps coupled with text simplifica-
tion, with synthesized speech output. For specific
conditions which may be treated at specialist clin-
ics (our own experience is based on Somalis with
respiratory difficulties) it may be possible to iden-
tify a series of frequently asked questions and set
up a pre-consultation computer-mediated help-desk
and interview (cf. Osman et al. 1994). See Somers
and Lovel (2003) for more details.
Having decided that a visit to the doctor is indi-
cated, the next step is to make an appointment. Ap-
pointment scheduling is the classical application of
SLT, as seen in most of the early work in the field,
and is a typical case of a task-oriented cooperative
dialogue. Note that the “practitioner” – the recep-
tionist in the clinic – does not necessarily have any
medical expertise, nor possibly the high level of ed-
ucation and openness to new technology that is often
assumed in the literature on medical SLT which talks
of the “doctor” controlling the device.
If this is the patient’s first encounter with this par-
ticular healthcare institution, there may be a process
of gathering details of the patient’s medivcal his-
tory and other details, done separately from the
main doctor–patient consultation, to save the doc-
tor’s time. This might be a suitable application for
computer-based interviewing (cf. Bachman 2003).
The next step might be the doctor–patient consul-
tation, which has been the focus of much attention.
For no doubt practical purposes, some medical SLT
developers have assumed that the patients role in this
can be reduced to simple responses involving yes/no
responses, gestures and perhaps a limited vocabu-
lary of simple answers at the limit. This view un-
fortunately ignores current clinical theory. Patient-
centred medicine (cf. Stewart et al. 2003) is widely
promoted nowadays. The session will see the doctor
eliciting information in order to make a diagnosis as
foreseen, but also explaining the condition and the
treatment, and exploring the patients feelings about
the situation. While it may be unrealistic at present
to envisage fully effective support for all these as-
pects of the doctorpatient consultation, we feel that
its purpose should be explicitly appreciated, and the
limitations of current technology in this respect ac-
knowledged.
After the initial consultation, the next step may
involve a trip to the pharmacist to get some drugs or
equipment. Apart from the human interaction, the
drugs (or whatever) will include written instructions
and information: frequency and amount of use, con-
traindications, warnings and so on. This is an ob-
vious application for controlled language MT: drug
dose instructions are of the same order of complexity
as weather bulletins. For non-literate patients, “talk-
ing pill boxes” are already available:5 why can’t
they talk in a variety of languages?
Another outcome might involve another practi-
tioner – a nurse or a therapist – and a series of meet-
5Marketed by MedivoxRx. See Orlovsky (2005).
ings where the condition may be treated or managed.
Apart from more scheduling, this will almost cer-
tainly involve explanations and demonstrations by
the practitioner, and typically also elicitation of fur-
ther information from the patient. Hospital treat-
ment would involve interaction with a wide range
of staff, again not all medical experts. If a commu-
nication device is to be used, it makes more sense
for it to be under the control and “ownership” of the
person who is going to be using it regularly: the pa-
tient.
6 Conclusion
Some of the comments made in this position paper
may seem critical, but it has not been my intention
to be negative about the field.6 It has been my inten-
tion in this paper to draw attention to the following
aspects of medical SLT which I believe so far have
been somewhat neglected:
• What is the ideal user profile for medical SLT?
Should the doctor control the system, or could
it be seen as a shared resource?
• If the patient is also a user, devices need to be
more user-friendly, taking into account cultural
differences, and problems of low literacy.
• This particularly applies to feedback and veri-
fication modules in the system.
• Evaluation should focus on the ability of the
technology to aid the completion of the task,
from the perspective of both the practitioner
and the patient.
• Evaluation methods should not involve partici-
pants in meaningless or incomprehensible tasks
(such as rating nonsensical output), nor rely on
skills (such as literacy) that they may lack.
• The pathway to healthcare involves more than
the one-way doctor–patient dialogues covered
by most systems. A wide range of technologies
can be brought to bear on the problem.
6In particular, it should perhaps be acknowledged that in
terms of practical accomplishment we have yet to match oth-
ers in the field.
Acknowledgments
I would like to acknowledge the contribution to this
paper of my colleagues on the CANES project, Ann
Caress, Gareth Evans, Marianne Johnson, Hermione
Lovel, Zeinab Mohamed. Some of the work re-
ported here is funded by the Economic and Social
Research Council (ESRC), project reference RES-
000-23-0610. Thanks also to the anonymous refer-
ees for their very useful comments.

References
Bachman, J.W. 2003. ‘The patient-computer interview: a
neglected tool that can aid the clinician.’ Mayo Clinic
Proceedings, 78:67–78.
Benoˆıt, Christian, Martine Grice and Val´erie Hazan.
1996. ‘The SUS test: A method for the assessment of
text-to-speech synthesis intelligibility using Semanti-
cally Unpredictable Sentences’. Speech Communica-
tion, 18:381–392.
Bouillon, Pierrette, Manny Rayner, Nikos
Chatzichrisafis, Beth Ann Hockey, Marianne
Santaholma, Marianne Starlander, Yukie Nakao,
Kyoko Kanzaki and Hitoshi Isahara. 2005. ‘A generic
multi-lingual open source platform for limited-
domain medical speech translation’. In Proceedings
of the Tenth Conference on European Association
of Machine Translation, Budapest, Hungary, pp.
CHECK
Costantini, Erica, Fabio Pianesi and Susanne Burger.
2002. ‘The added value of multimodality in the NES-
POLE! speech-to-speech translation system: an exper-
imental study’. In Fourth IEEE International Confer-
ence on Multimodal Interfaces (ICMI’02), Pittsburgh,
PA, pp. 235–240.
Ettelaie, Emil, Sudeep Gandhe, Panayiotis Georgiou,
Robert Belvin, Kevin Knight, Daniel Marcu, Shrikanth
Narayanan and D. Traum. 2005. ‘Transonics: A practi-
cal speech-to-speech translator for English-Farsi med-
ical dialogues’. In 43rd Annual Meeting of the Asso-
ciation for Computational Linguistics: ACL-05 Inter-
active Poster and Demonstration Sessions, Ann Arbor,
MI, pp. 89–92.
Flores, Glenn. 2005. ‘The impact of medical interpreter
services on the quality of health care: a systematic re-
view’. Medical Care Research and Review, 62:255–
299.
Johnson, M.J. 2004. ‘What can we learn from drawing
parallels between people who use AAC and people
whose first language is not English?’ Communication
Matters, 18(2):15–17.
Johnson, M.J., D.G. Evans and Z. Mohamed. 2006. ‘A
pilot study to investigate alternative communication
strategies in provider-patient interaction with Somali
refugees’. In Current Perspectives in Healthcare Com-
puting Conference, Harrogate, England, pp. 97–106.
Johnson, M.J., G. Evans, Z. Mohamed and H. Somers (in
prep.) An investigation into the perception of symbols
by UK-based Somalis and English-speaking nursing
students using a variety of symbol assessment tech-
niques.
Lee, Young-Suk, Daniel J. Sinder and Clifford J. Wein-
stein. 2002. ’Interlingua-based English–Korean two-
way speech translation of doctor–patient dialogues
with CCLINC’. Machine Translation, 17:213–243.
Mitchell, E. and F. Sullivan. 2001. ‘A descriptive feast but
an evaluative famine: systematic review of published
articles on primary care computing during 1980-97’.
British Medical Journal, 322:279–282.
Narayanan, S., S. Ananthakrishnan, R. Belvin, E. Et-
telaie, S. Gandhe, S. Ganjavi, P. G. Georgiou, C.
M. Hein, S. Kadambe, K. Knight, D. Marcu, H. E.
Neely, N. Srinivasamurthy, D. Traum, and D. Wang.
2004. ‘The Transonics spoken dialogue translator: an
aid for English-Persian doctor-patient interviews’. In
Timothy Bickmore (ed.) Dialogue Systems for Health
Communication: Papers from the 2004 Fall Sympo-
sium, American Association for Artificial Intelligence,
Menlo Park, California, pp. 97–103.
Orlovsky, Christina. 2005. ‘Talking pill bot-
tles let medications speak for them-
selves’. NurseZone.com (online magazine),
www.nursezone.com/Job/DevicesandTechnology.asp
?articleID=14396. Accessed 15 March 2006.
Osman, L., M. Abdalla, J. Beattie, S. Ross, I. Russell,
J. friend, J. Legge and J. Douglas. 1994. ‘Reducing
hospital admissions through computer supported edu-
cation for asthma patients’. British Medical Journal,
308:568–571.
Precoda, Kristin, Horacio Franco, Ascander Dost,
Michael Frandsen, John Fry, Andreas Kathol, Colleen
Richey, Susanne Riehemann, Dimitra Vergyri, Jing
Zheng and Christopher Culy. 2004. ‘Limited-domain
speech-to-speech translation between English and
Pashto’. In HLT-NAACL 2004, Human Language
Technology Conference of the North American Chap-
ter of the Association for Computational Linguistics:
Demonstrations, pp. 9–12.
Rayner, Manny and Pierrette Bouillon. 2002. ’A flexible
speech to speech phrasebook translator’. In Proceed-
ings of the ACL-02 Workshop on Speech-to-Speech
Translation: Algorithms and Systems, Philadelphia,
PA, pp. 69–76.
Somers, Harold, Gareth Evans and Zeinab Mo-
hamed. 2006. ‘Developing speech synthesis for under-
resourced languages by “faking it”: an experiment
with Somali’. In Proceedings of LREC: 5th Confer-
ence on Language Resources and Evaluation, Genoa.
Somers, Harold and Hermione Lovel. 2003. ‘Computer-
based support for patients with limited English’. In As-
sociation for Computational Linguistics EACL 2003,
10th Conference of The European Chapter, Proceed-
ings of the 7th International EAMT Workshop on MT
and other language technology tools, Budapest, pp.
41–49.
Somers, Harold and Yuriko Sugita. 2003. ‘Evaluating
commercial spoken language translation software.’ In
MT Summit IX: Proceedings of the Ninth Machine
Translation Summit, New Orleans, pp. 370–377.
Starlander, Marianne, Pierrette Bouillon, Manny Rayner,
Nikos Chatzichrisafis, Beth Ann Hockey, Hitoshi Isa-
hara, Kyoko Kanzaki, Yukie Nakao and Marianne San-
taholma. 2005. ‘Breaking the language barrier: ma-
chine assisted diagnosis using the medical speech
translator’. In Proceedings of the XIX International
Congress of the European Federation for Medical In-
formatics, Geneva, Switzerland.
Stewart, Moira, Judith Belle Brown, W. Wayne We-
ston, Ian R. McWhinney, Carol L. McWilliam and
Thomas R. Freeman. 2003. Patient-Centered Medi-
cine: Transforming the Clinical Method (2nd ed.).
Radcliffe, Abingdon, Oxon.
Tomita, Masaru, Marion Kee, Hiroaki Saito, Teruko Mi-
tamura and Hideto Tomabechi. 1988. ’The universal
parser compiler and its application to a speech trans-
lation system’. In Second International Conference
on Theoretical and Methodological Issues in Machine
Translation of Natural Languages, Pittsburgh, Penn-
sylvania, pages not numbered.
Waibel, Alex, Ahmed Badran, Alan W. Black, Robert
Frederking, Donna Gates, Alon Lavie, Lori Levin,
Kevin Lenzo, Laura Mayfield Tomokiyo, J¨urgen
Reichert, Tanja Schultz, Dorcas Wallace, Monika
Woszczyna, and Jing Zhang. 2003. ‘Speechalator:
two-way speech-to-speech translation on a consumer
PDA’. In Proceedings of EUROSPEECH 2003, 8th
European Conference on Speech Communication and
Technology, Geneva, pp. 369–372.
Zhou, Bowen, Daniel D´echelotte and Yuqing Gao. 2004.
‘Two-way speech-to-speech translation on handheld
devices’. In INTERSPEECH 2004 – ICSLP, 8th Inter-
national Conference on Spoken Language Processing,
Jeju Island, Korea, pp. 1637–1640.
