LEXICAL ACCOMMODATION IN MACHINE-MEDIATED 
INTERACTIONS 
Laurel Fais 
ATR Interpreting Telecommunications Research Laboratories 
2-2 Hikaridai, Seika-cho Soraku-gun 
Kyoto, Japan 619-02 
email: fais@itl.atr.co.jp 
Abstract 
We report results of lexical accommodation studies 
involving three different interpretation settings: 
human-human monolingual; human-interpreted 
bilingual; and machine-interpreted bilingual. We 
found significant accommodation in all three 
conversational settings, with the highest rate in the 
human-interpreted setting. There is evidence for long- 
range mutual accommodation in that setting, as 
compared to short-range accommodation in the 
machine-interpreted setting. Motivations discussed in 
the accommodation literature, including speakers' 
concern for social standing and communicational 
efficiency, are examined in the light of the results 
obtained. Finally, we draw implications for the 
design of multimedia human-computer interfaces. 
1 Introduction 
For real-time, real-situation, human-computer 
interaction to approach reality, the burden of 
understanding and conveying information cannot be 
shared equally between the two interaetors. Humans 
need to make allowances for features of the computer 
interface such as synthesized speech, limitations on 
range of knowledge base, and imperfect speech 
recognition. However, in order for naive users to 
accept and use computers effectively in an interactive 
format, the restrictions placed upon them need to be as 
minimal and as natural as possible. Exploring the 
linguistic behavior of humans interacting with 
computers in an unrestricted environment allows us to 
determine how humans are naturally inclined to 
accommodate to the current limitations of human- 
machine interaction. Encouraging those natural 
inclinations in real human-machine systems will have 
a better chance of success than imposing artificial 
restrictions. In addition, systems designers can take 
advantage of the accommodations that humans make 
naturally to improve the performance of their systems 
(Fais et al., 1995) 
Below, we will discuss a particular kind of 
accommodation in conversational interaction, namely 
lexical accommodation. In lexical accommodation, 
one conversant adopts the lexical items used by the 
other conversant. This type of accommodation is one 
way in which users adapt to the limitations of 
computer interfaces, i.e., they converge to the limited 
lexicon of the computer. Thus, it has important 
implications for the design of workable human- 
computer interfaces. We discuss experiment results 
from three conversational settings: human-human 
monolingual; human-interpreted bilingual, and 
machine-interpreted bilingual. 
1.1 Background 
Of course, there is no a priori reason why interactors 
could not conduct conversations in completely 
different styles, using different phonologies, sentence 
structures, vocabulary, etc. However, it has been 
widely demonstrated that they do not. Lexical 
accommodation is only one instance of a diverse range 
of convergence behaviors that humans display in 
conversation. Giles etal. (1987) cite studies 
demonstrating convergence of speech styles, dialect, 
non-verbal behavior, vocal intensity, prosody, speech 
rate and duration, and pause length. Garrod and 
Doherty (1994) report on a study in which 
conversational interactors in a language community 
showed a high level of convergence on a particular 
description language over the course of the task (a 
maze game). Fais (1994a) discusses both lexical and 
syntactic accommodation over a range of natural 
conversational styles. 
Accommodation has also been shown to be present 
in human-computer interactions. Zoltan-Ford (1991) 
and Leiser (1989) demonstrate accommodation by 
users to the phrasing and vocabulary of the 
confirmation output of a computer in information 
manipulation and retrieval tasks. Speakers also 
unconsciously adapt their speaking behavior to the 
limitations of a speech recognition system, as 
demonstrated by Mellor and O'Connor (1995). 
But while the phenomenon of accommodation, both 
between humans and between human and computer, is 
amply demonstrated, the motivations behind the 
phenomenon are less often discussed. Those in the 
field of human-computer interactions usually note 
simply that accommodation exists. They express 
relief when it is found to act as a natural constraint on 
the user's vocabulary or syntax (e.g., Leiser, 1989; 
Zoltan-Ford, 1991), and distress when accommodation 
to the "natural speech" style of some computer output 
encourages matching casual speech from the user 
which is difficult to process (e.g., Spitz et al., 1991). 
Speech Accommodation Theorists, who fall under a 
broad category which might be called socio- 
linguistics, tend to ascribe one of three motivations to 
speakers who accommodate: "evoking listeners' 
social approval, attaining communicational efficiency 
between interactors, and maintaining positive social 
identities." (Giles et al., 1987) p.t5). These 
motivations can be grouped into two major categories: 
concern with social standing and identity, and concern 
370 
with communicational efficiency. 2 Methods 
1.2 Hypotheses 
Specifically, then, if we base our predictions on these 
standard accounts in the literature, we should expect 
the following results concerning both level and 
direction of accommodation: 
• In human-human interaction, we should find 
significant lexical accommodation. Because this is 
essentially an information-giving and -receiving task, 
we expect that the receiver of the information will 
accommodate to the giver, adopting the lexical items 
used by the speaker who imparts information. 
. The human-interpreted setting constitutes 
both a human-human interaction and a more stressful 
communication environment, one in which 
communicational efficiency is a concern. For that 
reason, we expect an even greater level of 
accommodation in the human-interpreted setting than 
in the human-human setting. In the human- 
interpreted setting, we examine the accommodation 
between client and interpreter. The client is 
inibnnation receiver, and the interpreter is the imparter 
of information, not the originator; thus, neither client 
nor interpreter is in a dominant role. For this reason, 
we cannot predict whether the client will accommodate 
to the interpreter or vice versa. 
• The machine-interpreted setting only 
indirectly involves human-human interaction; all 
dialogue is mediated by the "machine" interpreter. 
Therefore, we conjecture that interactors in this setting 
will not be concerned with social standing. On the 
other hand, this is the most difficult communication 
environment of the three, involving as it does not 
only the limited understanding of the machine 
translator but also limited speech recognition, a 
difficult-to-understand modulated speech signal, and 
rigid turn-taking constraints. For this reason, 
communicational efficiency will be a concern. 
However, whether this will generate more or less 
accommodation than concern for social standing 
generates in the human-human case is an open 
question. 
Since users in the machine-interpreted setting should 
not be concerned with social standing, we might 
predict a lower rate of accommodation than in the 
human-interpreted setting. However, again, the 
greater difficulties in communication in the machine- 
interpreted setting might make up for a lack of 
concern with social standing, resulting in a rate of 
accommodation comparable to that in the human- 
interpreted setting. Results discussed below will shed 
some light on the interaction of these two factors. 
, We expect that clients will accommodate to 
the machine to some extent, that clients' word choice 
will be affected by their perception of "what works," 
or "what the machine knows." Given this possibility, 
then, we predict that the results will show client 
accommodation to the machine-interpreter, but at a 
lower level than in the human-interpreted setting. 
We conducted three experiments to evaluate lexical 
accommodation in various interpretation settings. In 
the first experiment, native English-speaking subjects 
acting as "clients" were instructed that their task was 
to get directions to the site of a conference they were 
"attending" by engaging in a cooperative dialogue 
with a native English-speaking "conference agents." 
Their interaction was humml-human. In two other, 
similar experiments, English-speaking clients 
interacted with Japanese-speaking agents, both to get 
directions and to make hotel reservations. In one of 
these experiments, speech was translated by human 
interpreters; in the other, by a simulated automatic 
machine translation system ('Ngizard of Oz" style; see 
(Fais, 1994b) and (Fais et al., 1995) for further details 
of these two experiments). We will refer to this latter 
setting as the "machine-interpreted" setting; keep in 
mind, however, that translation was actually done by 
trained interpreters mimicking a computer-based 
system.1 The experimental configurations are shown 
in Figure 1. 
2.1 Measures 
Lexical Accommodation: We measured lexical 
accommodation by exmnining the number of lexical 
items which were used by both interaetors in the 
course of a conversation. The accommodation rate for 
each conversation in the three experiments was 
calculated by dividing the number of (different) lexical 
items the two speakers had in common by the total 
number of (different) lexical items in the conversation. 
We calculated lexical accommodation rates for client 
and agent in the same-language, human-human 
experiment setting; for client and (Japanese-to- 
English) interpreter in the human-interpreted 
experiment setting; and for client and (Japanese4o- 
English) "Wizard" interpreter in the machine- 
interpreted experiment setting. Although the actual 
measurement of lexical items was done for the 
English speech of the Japanese-to-English interpreters, 
these interactors in the conversation will be referred to 
below as "agents," to conform to the human-human 
setting in which we assessed the lexical 
accommodation of the agents directly. 
Direction of Accommodation: Accommodation 
is not necessarily a mutual phenomenon (Giles, 
1987). In order to determine if one conversant was 
accommodating more than the other, we examined the 
number of words used first by the client and the 
number used first by the agent. We reasoned that the 
subject who used a particular lexical 
1At no time did any subject doubt that he/she was 
interacting with an actual machine translator. 
In each experiment, subjects interacted in two modes: 
via a standard telephone, and via a computer-based, 
multimedia environment in which subjects could interact 
by voice, text, and drawing (Loken-Kim et al., 1993). 
The results discussed below did not differ for these two 
modes; thus, we will not distinguish them here. 
371 
Eng glish 
agent c.ent 
Htm~an-human setting 
Interpreter 
Human-interpreted experiment 
item first was not accommodating, and, by extension, 
then, the subject who did not use an item first, was 
accommodating. 
The following objection to this definition might be 
made: the fact that interactors use words that their 
partners have used does not necessarily mean that they 
are accommodating to the other's prior use of that 
word. But what other justification could there be for 
saying that accommodation is taking place? Because 
even lexical accommodation is rarely a conscious act, 
speakers' intuitive judgments are not helpful. On the 
other hand, there is no other external evidence 
available. Thus, we will use the quantitative criterion 
described above. We will argue that accommodation 
is a real phenomenon in dialogue; it follows, then, 
that at least some of the instances in which 
conversants use lexical items previously used by their 
partners are instances of accommodation. 
Relative Importance of Words-in- 
Common: We also looked beyond the initial use of 
lexical items to determine what role was played in 
subsequent conversation by the words that agent and 
client both used. That is, once one conversant 
accommodates to the other by adopting a lexical item, 
does that conversant continue to use that lexical item 
in a significant way in the remainder of the 
conversation? In order to explore this question, we 
estimated the percent of usage for each worddn- 
common, for both client and agent. That is, for each 
word-in-common, we divided the number of times 
each subject used the word by the total number of 
words uttered by that subject in order to determine 
what proportion of the subject's conversation 
consisted of the uses of that word. By comparing 
these proportions for the roles of client and agent, we 
ascertained the relative frequency of the word for each 
role, giving us some idea of the "importance" of the 
word for that role. 
Coincidental Overlap: Of course, a certain 
amount of lexical overlap is inevitable as a simple 
artifact of cooperative conversation. In order to 
i 
Machine-interpreted experiment 
I I LMJ 
E/itzlish 
/61ient 
Figure 1. Speech and visual data configurations for 
each experiment. Computers were used only in the 
multimedia condition. Telephones replaced 
headphones in the telephone condition. Subjects 
could not see one another or the interpreter/Wizards. 
determine file extent of coincidental overlap, we 
measured the lexical overlap in the speech of clients 
and agents from the first experiment who had not 
participated in the experiment together. That is, the 
speech of clients who had participated in the 
experiment with Agent A was compared with the 
speech of Agent B. Likewise, the speech of clients 
who had participated in the experiment with Agent B 
was compared with the speech of Agent A. Because 
these conversants were not talking to one another, the 
lexical overlap in their speech could not be a result of 
accommodation to one another. However, because the 
overlap was calculated for speakers engaging in 
cooperative dialogues concerned with the same task 
and via the same media, it reflects the extent to which 
overlap occurs simply because these are speakers in 
similar situations talking about similar topics. 
3 Results 
3.1 Lexical Accommodation 
.4 
.35 .~ 
.3 
.25 
(9 
(9 .2 
~d 15 
O9 
m .05 
o 
Figure 2. Rates of accommodation for coincidental 
overlap and all three settings. 
372 
Table 1. Significance levels for differences in lexical accommodation (two-way ANOVA). 
Coincidental overlap 
Human-human 
Ihlman-interpreted 
human-human human-interpreted machine-inte~re~d 
p<.03 p<.0001 p<.001 
p<.00m P<.0? 
p<.01 
"\[%e rates for lexical accommodation for all three 
experiments were significantly different from the level 
established for coincidental lexical overlap (Figure 2, 
Table 1; data were subjected to two-way analyses of 
variance). In addition, the lexical accommodation 
rates for each experiment also differed significantly 
from those for each of the other two experiments. 
Human-human accommodation was higher than 
coincidental overlap, but lower than both of the 
interpreted settings. The human-interpreted setting 
had the highest rate of accommodation. 
3.2 Direction of Accommodation 
When we examined the percentage of words-in- 
common used first by each role (agent or client), the 
following patterns emerged (Figure 3; data were 
subjected to three-way analyses of variance). In the 
hmnan-human setting the agent used a significantly 
higher percentage of words first (p<.03); the client 
accommodated to the agent. In the human-interpreted 
setting, there was no difference. It is not possible to 
say that one or the other interlocutor was responsible 
for the accommodation found. In the machine- 
interpreted setting, the agent used a significantly 
higher percentage of words first (p<.005); the client 
accommodated to the agent. 
6O 
50 
~: 40 
"~ 3O © 
2O 
10 
--1 Agent 
Client -- 
i~g,~¥i~- -- - ........... i!!~i!!!iii~i!~!~ii!!i ::::::::::::::::::::::: 
i~iiiiii~i/i;:i:::. ~!iii~ ~ i~Ji~,~.~ 
~iiiiiiiiiiiiNii 
Human 
iiii==iiiii~{!ii% 
H-interp. 
....:~,.... .............. 
~ ~ !~ ~J~:-:: ~i}ii~ ~#~{ziiii::::::::::::::::::::::: 
M-interp. 
Figure 3. Percent of first use of words-in-common for 
agent and client in each setting. 
3.3 Relative Importance of Words-in- 
common 
An examination of the use of each word-in-common 
with respect to overall word use for client and agent, 
i.e., the word's "importance," showed the following 
results (Figure 4; three-way ANOVA). In the human- 
human setting, client use of words-in-common made 
up a significantly greater percent of total word use 
than agent use of these words (p<.0003). There was 
no significant difference between client and agent in 
the hnman-interpreted setting or in the machine- 
interpreted setting. 
1.6 £1.4 
=1.2 
o 1 
'S .8 
cD c~ .4 
.2 
0 
I-'1 Client 
\[\] Agent 
Human H-interp. M-interp. 
Figure 4. Frequency of use of words-in-common for 
agent and client in each setting. 
4 Discussion 
We analyzed lexical accommodation in a variety of 
interactions in order to determine how accommodation 
can be expected to operate in a machine-interpreted 
context, and learn ways in which to support lexical 
accommodation in the design of human-machine 
interfaces. It is encouraging that lexical 
accommodation happens spontaneously. As our 
initial results show, it is not simply a coincidental 
byproduct of conversing about common topics. There 
is a significant difference between that case (what we 
have called "coincidental overlap") and the case of two 
people talking about the same topic to one another. 
While lexical accommodation has been shown for 
typed human-computer interactions (Leiser, 1989), and 
lexical and structural accommodation has been 
demonstrated for typed and spoken human-computer 
interactions in constrained contexts (Zoltan-Ford, 
1991), these results demonstrate lexical 
accommodation for unconstrained, spoken, machine- 
mediated human-human interactions. Given that 
lexical accommodation is a "real" phenomenon, then, 
how can we characterize the patterns of 
accommodation for each experimental setting, and 
what can we learn from them? 
In the human-human setting, there was a non- 
trivial, but low level of accommodation. The client 
accommodated to the agent, using the words-in- 
common more frequently in subsequent conversation 
than did the agent. This is consistent with the 
interpretation that the agent acted as information- 
provider and the client acted as information-receiver in 
a non-stressful communication environment, as our 
initial hypothesis stated. The interactors maintained a 
level of accommodation high enough to satisfy their 
373 
concern for social standing, but since they were native 
speakers of the same language and the communication 
channel was clear and direct, they had minimal concern 
for communicational efficiency. There was no 
incentive to extend lexical accommodation. 
On the other hand, the human-interpreted setting 
presented a more difficult communication environment 
in which concern for communicational efficiency was 
present. Since the interaction was also human- 
human, social standing continued to be a concern. We 
expected that the addition of the concern for efficiency 
to that for social position would result in a higher 
level of accommodation; in fact, the level of 
accommodation observed in the human-interpreted 
setting was the highest in all three experimental 
settings. 
More specifically, the human-interpreted setting 
involved speakers from two different language 
backgrounds, both of whom were capable of 
recognizing the differences in their linguistic 
behaviors, and of reducing those differences to 
facilitate communication. Lexical choice is a surface 
level phenomenon, open to manipulation. Thus, 
lexical accommodation is an important conversational 
strategy for speakers who do not share linguistic 
conventions. The interpreters in the human- 
interpreted setting were native speakers of Japanese, 
and, though they were fluent in English, the range of 
overlap between their English linguistic habits and 
those of the native English-speaking clients was much 
smaller than that between two native speakers. 
Lexical choice was an obvious strategy for 
establishing shared linguistic behavior, and thus 
promoting effective communication. So, concern for 
social standing and communicative efficiency 
combined to generate a high rate of mutual 
accommodation. 
Because neither client nor interpreter had a dominant 
role in the conversation, we could not predict the 
direction of accommodation. In fact, our results show 
that it was not possible to single out a primary 
accommodator in the human-interpreted setting, either 
in terms of proportion of words used first or the 
frequency with which words-in-common were used. 
Does this mean that both speakers accommodated or 
that neither did? Considering the high accommodation 
rates for this setting (Figure 1), we conclude that, in 
fact, both client and agent were accommodating to one 
another. 
In the machine-interpreted setting, we saw a rate of 
accommodation higher than that of the human-human 
setting, but lower than that of the human-interpreted 
setting, as expected. The machine-interpreted setting 
is an even more stressful communication environment 
than the human-interpreted setting; concern for 
communicational efficiency resulted in a higher level 
of accommodation than concern for social standing did 
in the human-human setting. However, we do not 
expect humans to be concerned about their social 
standing with a machine, unlike in the human- 
interpreted setting. This explains why the rate for 
lexical accommodation in the machine-interpreted 
setting is lower than that of the human-interpreted 
setting. The greater concern for communicational 
efficiency in the machine-interpreted setting was not 
enough to generate as high a level of accommodation 
as that found in the human-interpreted setting, where 
there was the additional factor of concern for social 
position, though it did generate a higher level of 
accommodation than did concern for social standing 
alone (the human-human setting). 
As we conjectured above, clients accommodated to 
the machine as part of a strategy for effective 
communication. However, given the fact that there 
was a lower rate of accommodation than in the 
human-interpreted setting, coupled with the strong 
directionality observed, we conclude that this is not a 
case of mutual accommodation. Instead, as in the 
human-human setting, clients were the primary 
accommodators. Clients in the machine-interpreted 
setting may have perceived the machine to be in the 
dominant role, just as the agent played the dominant 
role in the human-human setting. 
Recall that clients in the machine-interpreted 
setting, unlike those in the human-interpreted setting, 
did not use words-in-common more than the agent did 
in subsequent conversation. Since clients were not 
concerned with social standing, including establishing 
mutual linguistic conventions, accommodation in the 
machine-interpreted setting was a local phenomenon 
which did not extend throughout the conversation. 
5 Conclusions and Future Directions 
What does this tell us about the design of human- 
computer interfaces? Recall that these conversations 
were unconstrained; neither agents, clients nor 
interpreters, whether human or "machine," were under 
instructions to limit or modify their speech in any 
way. Thus, what we see in these results is the natural 
tendency of humans to accommodate to their 
interlocutors in a variety of communication 
environments. This tendency resulted in the highest 
level of accommodation in the human-interpreted 
setting. That level was achieved as a result of mutual 
accommodation between the two humans involved, 
both of whom felt a concern for both social standing 
and communicational efficiency. The level of 
accommodation observed in the machine-interpreted 
setting was both lower and less extensive, i.e., it did 
not persist throughout the conversation. 
We can take advantage of even the moderately high 
level of accommodation found in the machine- 
mediated setting by building into a language 
processing system a preference for the lexical items 
used by the machine. Coupled with accommodations 
in other aspects of language such as discourse and 
syntactic complexity, fluency, and speaking rate (Fais 
et al., in press), lexical accommodation can inform a 
language model to improve language processing 
performance by exploiting the relationships between 
human speech and the speech of the machine interface. 
Specifically, candidates for speech recognition or 
parsing could receive higher preference scores if they 
include lexical items or structures previously 
encountered in the discourse. Alternatively, 
374 
preferences based on accommodation could be built 
directly into a language model for speech recognition 
or decision tree for parsing. 
We would like to investigate the possibility of 
increasing the level of accommodation in the machine~ 
mediated setting. Ideally, we would like users' 
accommodation to a machine interface to be as high as 
possible so that the lexical variability of users' speech 
can be as constrained as possible. One way is to use 
the resources of a multimedia environment to replicate 
the effect of the human-interpreted setting by 
providing the machine interface with a human-like 
persona. A number of human-computer systems 
already include such a feature (e.g., Ball and Ling, 
1995; Bertenstam et al., 1995; Webber, 1995); it 
remains to be seen if it will have the desired effect on 
lexical accommodation. On the other hand, there is 
evidence that encouraging users to interact with 
machines as if they were humans may actually 
undermine the quality of the users' speech from the 
point of view of language processing. Work in the 
area of disfluencies in human-to-machine speech 
suggests that humans do, in fact, "clean up" their 
speech for machines (Suhm et al., 1994). These 
advantages may be lost if humans are encouraged to 
treat a machine interface as if it were human. 
Empirical investigation is required to determine if an 
optimal balance can be reached. 
We have suggested that the design of speech 
recognition and language processing systems can take 
advantage of users' lexical accommodation to machine 
interfaces to improve system performance. This, in 
turn, would allow the construction of systems which 
make fewer demands on the willingness of users to 
adjust to misrecognitions and nfisunderstandings, and 
which encourage users to interact with computer 
interpreters as if they were interacting with human 
interpreters. This result would also have the effect of 
further increasing lexical accommodation from users. 
Acknowledgments 
The author would like to thank Kyung-ho Loken- 
Kim, Yoshinori Ohkubo, and Tsuyoshi Morimoto for 
their encouragement, help, and support. 
References 
Ball, J. Eugene, and Daniel T. Ling. 1995. 
Spoken language processing in the Persona 
Conversational Assistant. In Proceedings of the 
ESCA Workshop on Spoken Dialogue Systems, 
pages 109-112, Vigs0, Denmark, May-June. 
Bertenstam, Johan, et al. 1995. Tim Waxholm 
system: A progress report. In Proceedings of the 
ESCA Workshop on Spoken Dialogue Systems, 
pages 81-84, Vigs0, Denmark, May-June. 
Fais, Laurel. 1994a. Conversation as 
collaboration: Some syntactic evidence. Speech 
Communication, 15:231-242. 
Fais, Laurel. 1994b. Effects of communicative 
mode on spontaneous English speech. Technical 
Report of the Institute of Electronics, Information and 
Communication Engineers, NLC94-22(1994-10): 1-8. 
Fais, Laurel, Kyung-ho Loken-Kim, and Tsuyoshi 
Morimoto. (in press) Linguistic and para-linguistic 
differences between multimodal and telephone-only 
dialogues in English and Japanese. J. Acoustical 
Society of Japan (E), (in press). 
Fais, Laurel, Kyung-ho Loken-Kim, and Young- 
Duk Park. 1995. Speakers' responses to requests for 
repetition in a multimedia cooperative dialogue. In 
Proceedings of the International Conference on 
Cooperative Multimodal Communication, pages 129- 
144, Eindhoven, The Netherlands, May. 
Garrod, Simon, and Gwyneth Doherty. 1994. 
Conversation, coordination, and convention: An 
empirical investigation of how groups establish 
linguistic conventions. Cognition, 53:181-215. 
Giles, Howard, Anthony Mulac, James J. Bradac, 
and Patricia Johnson. 1987. Speech accomtnodation 
theory: The first decade and beyond, ill M.L. 
McLaughlin, editor, Communication Yearbook 10, 
pages 13-48. Sage Publications, London, UK. 
Leiscr, R.G. 1989. Exploiting convergence to 
improve natural language understanding. Interacting 
with Computers: The Interdisciplina<y Journal of 
Human-Computer Interaction, 1(3):284-298. 
Loken-Kim, Kyung-ho, Fumihiro Yato, Kazuhiko 
Kurihara, Laurel Fais, and Ryo Furukawa. 1993. 
ATR Technical Report TR-IT-0018. ATR Interpreting 
Telecommunications Research Laboratories, Kyoto, 
Japan. 
Mellor, Brian, and Cian O'Connor. 1995. User 
adaptation to voice input interfaces. In Proceedings of 
the ESCA Workshop on Spoken Dialogue Systems, 
pages 97-100, Vigs0, Denmark, May-June. 
Spitz, Judith. 1991. Collection and analysis of 
data from real users: hnplications for speech 
recognition/understanding systems. In Proceedings of 
the 4th Darpa Workshop on Speech and Natural 
Language, pages 164-169, Asilomar, California, 
USA, February. 
Suhm, B., L. Levin, N. Coccaro, J. Carbonell, K. 
Horiguchi, R. Isotani, A. Lavie, L. Mayfield, C. P. 
Ros6, C. Van Ess-Dykema, and A. Waibel. 1994. 
Speech-language integration in a multi-lingual speech 
translation system. In Proceedings of the AAAI 
Workshop on Speech and Language Processing, 
Seattle, Washington, USA. 
Webber, Bonnie. 1995. Instructing animated 
agents: Viewing language in behavioral terms. In 
Proceedings of the International Conference on 
Cooperative Multimodal Communication, pages 5- 
15, Eindhoven, The Netherlands, May. 
Zoltan-Ford, Elizabeth. 1991. How to get people 
to say and type what computers can understand. 
International Journal of Man-Machine Studies, 
34:527-547. 
375 
