Spontaneous Speech Collection for the ATIS Domain with 
an Aural User Feedback Paradigm 
Christine Pao and Jay Wilpon 
AT&T Bell Laboratories 
600 Mountain Ave. Office: 2D-464 
Murray Hill, NJ 07974 
ABSTRACT 
This paper describes the AT&T ATIS data collection system, 
with emphasis on the development of the speech-in, speech- 
out interaction paradigm. The ATIS task involves providing 
air travel information to a user in the context of a interactive 
dialogue. Under the AT&T interaction paradigm, informa- 
tion retrieved from a travel information database in tabular 
form is automatically transformed into sentences, which are 
spoken to a user by a speech synthesizer. To date, we have 
collected over 1800 sentences from subjects who used the sys- 
tem to solve travel planning scenarios. We present a compar- 
ison of the ATIS data collected at AT&T with the ATIS data 
collected at other sites (BBN, CMU, MIT, and SRI), and 
discuss what we have learned in this prehminary effort. 
1. INTRODUCTION 
In support of our research toward developing telephone- 
based spoken language systems, we have joined BBN, 
CMU, MIT, and SRI in collecting speech and language 
data for the ATIS (Air Travel Information Service) do- 
main. The task of a spoken language system can be 
broken into three parts. First, the system must under- 
stand and interpret what the user says in the context 
of the human-machine dialogue. If the user's request is 
understood, the system must retrieve the requested in- 
formation. The information source for the ATIS domain 
is a relational database that represents a 10-city subset 
of the OAG, or Official Airline Guide. Finally, the sys- 
tem must convey the retrieved information to the user 
in an appropriate format, or provide some other coop- 
erative response, such as a request for more information 
or an error message. In the AT&T ATIS data collec- 
tion system, the first and second portions of the task 
are handled by components of the MIT ATIS system \[1\]. 
Our efforts have been directed toward dealing with the 
third part of the task: information presentation and sys- 
tem feedback under a speech-in, speech-out interaction 
paradigm. Our goal is to begin to address the problems 
of dialogue and information control which will affect the 
performance of interactive spoken language systems. 
In this paper, we will first describe our data collection 
system and data collection procedure. Then, we will 
present analyses of AT&T ATIS data and of ATIS data 
collected at other sites. Finally, we will discuss what we 
have learned in this preliminary effort. 
2. DATA COLLECTION 
As at most of the other sites collecting ATIS data, data 
were collected from subjects' interactions with a partially 
simulated spoken language system. As in the MIT data 
collection setup, a human experimenter is substituted 
for the speech recognition component of the system to 
provide a transcription of the subject's speech for the 
natural language (NL) component. In this section, we 
discuss the system's development, describe the system 
hardware, and describe the collection procedure. 
2.1. System Development 
The AT&T data collection system was designed to 
closely simulate a real, telephone-based, human-machine 
interaction. Building on the framework of the MIT ATIS 
collection system, we directed our development effort 
towards controlling the presentation of information re- 
trieved from the database, providing feedback to the user 
on the state of the system and the discourse, and ex- 
ploring areas where system initiative would help users 
achieve their goals efficiently. We also modified the sys- 
tem control loop to transfer recording control away from 
the subject. 
Information Presentation. The collection systems 
at all other sites make use of a visual display to present 
information retrieved from the database in a tabular 
format. Because of our choice of an audio interaction 
paradigm, the AT&T system does not present informa- 
tion in a tabular format, but instead translates the re- 
trieved information into sentences. In some cases, in- 
formation is converted into sentences using an entry-to- 
phrase, template-based approach (Figure 1). When more 
information is retrieved from the database than can be 
reasonably presented by the template-based approach, 
summarization functions are used to present some sub- 
set of the information (Figure 2). In other cases, spe- 
cialized functions are used to avoid excessive repetition 
(Figure 3) or to select information based on the discourse 
history. 
43 
Sentence: I'LL TAKE DELTA FLIGHT NINE SEVENTY FIVE 
Ent,ryType Entry Sentence Fra~nent 
FLIGHTJD 105595 
AIBLINE~ODE DL delta 
AIRLINE3LIGHT 975 flisht nine seventy five 
FROM~IRPORT BOS from Boston 
TO_AIRPORT ATL to Atlanta 
DEPARTURE~IME 1520 departs at three tgenty P M 
ARRIVAL~INE 1804 and arrives at six oh four P M 
STOPS 0 gith no stops 
ATIS: delta flight nine seventy five from Boston to 
Atlanta departs at three tgenty P N and arrives at six oh 
four P M glth no stops. 
Figure 1: Template-based conversion of flight informa- 
tion to sentence format. 
Sentence: 
Table: 
FLIGHT~D 
105584 
105586 
105588 
I MANT TO GO FROM BOSTON TO ATLANTA ON MONDAY 
AIRLINE~ODE ... DEPARTURE~IME ... 
DL ... 630 ... 
EA ... 700 ... 
DL ... 815 ... 
Summary: There are flights departing between six thirty 
A M and eight tgenty four P M. 
ATIS: There are seventeen flights from boston to atlanta 
on Monday August nineteenth. 
There are flights departing betgeen six thirty A N and 
eight tgenty four P M. 
What time gould you like to go7 
Figure 2: Output of the table summarizer. The sum- 
mary is one part of a three sentence response to the 
user. 
Sentence: WHAT AIRLINES FLY FROM BOSTON TO ATLANTA 
Table: 
AIRLINE NAME AIRLINE CODE FROM TO 
DELTA AIR LINES, INC. DL BOS ATL 
EASTERN AIR LINES, INC. EA BOS ATL 
USAIR US BOS ATL 
ATIS: The airlines gith service between Boston and 
Atlanta are Delta, Eastern, and U S Air. 
Figure 3: In this example, a table with three rows is 
compressed into a single sentence. 
The information presentation component of the system 
was developed with two goals in mind. The first was 
to present information so it could be easily understood. 
Toward this end, the presentation component was de- 
veloped to format information into coherent sentences, 
to expand or hide all codes and abbreviations, and to 
maximize the intelligibility of the speech synthesizer out- 
put. The second goal was to minimize the amount of 
irrelevant information presented to the user. Towards 
this goal, the presentation component includes the above 
mentioned facilities for summarizing, compressing and 
filtering information retrieved from the database. 
Sentence: I WANT TO 60 ABOUT 3 P M 
ATIS: There are no flights leaving beteeen teo forty five 
P N and three fifteen P N. 
The next earliest flight is eastern flight six forty five 
departing at tgo tgenty one P M. 
The next latest flight is delta flight nine seventy five 
departing at three tgenty P M. 
Please refer to these flights by flight number or 
departure time. 
Figure 4: AT&T system initiative. 
System Feedback. The MIT ATIS system provides 
feedback to the user on the state of the discourse in the 
form of text and synthesized speech. However, the sup- 
porting text produced by the MIT system is intended 
to complement and to direct the user's attention to a 
tabular display. This capability was modified to com- 
plement the summarization facility mentioned above. In 
the example shown in Figure 2, the MIT system would 
generate the text "Here are the flights from Boston to 
Atlanta" to accompany the table listing seventeen flights, 
while the AT~T sysiem would generate "There are sev- 
enteen flights from Boston to Atlanta" to accompany 
the following summary sentence. The system error re- 
sponses (NL failure, database access failure, etc.) were 
also modified to fit the audio feedback paradigm. 
System Initiative. The MIT system takes initiative 
in two contexts: when the system does not have enough 
information to access the database, and when guiding a 
user through the flight booking process \[3\]. The AT&T 
system takes initiative in an additional context, when 
the subject is selecting a flight on the basis of departure 
or arrival time. First, the system prompts the user for 
a departure time if the departure time is summarized, 
as in Figure 2. Second, the system volunteers the next 
earliest and next latest flight when the subject requests 
a flight at a certain time, and there isn't one. An ex- 
ample is shown in Figure 4. This second capability was 
developed to address a problem that was causing a great 
deal of user frustration. Because the system would not 
provide complete flight information for more than three 
flights, subjects were forced to play a guessing game to 
find out flight departure times. The need for this type of 
system initiative is a result of the limits imposed by the 
interaction paradigm. The more a system restricts the 
flow of information, the more assistance it must provide 
to help the user access the information. 
Recording Control. At all the other ATIS data 
collection sites, the subject controls the recording pro- 
cess using a push-to-talk or push-and-hold to talk mech- 
anism. We chose not to use such a subject-controlled 
recording mechanism in order to more closely simu- 
late an actual telephone dialogue. Instead, the exper- 
44 
imenter who transcribed the subject's speech also con- 
trolled the start and end of recording from the key- 
board. The control loop was designed to keep the in- 
teraction flowing as smoothly and efficiently as possible 
in the hope of eliciting more natural speech from our 
subjects. Many subjects were initially uncertain about 
when to start and stop talking, but most of them ad- 
justed to the interaction after the first scenario. Some 
effects of experimenter-controlled recording on subjects' 
speech are discussed in section 3.4. 
2.2. Recording Environment and Sys- 
tem Hardware 
Data were collected in a walled-off corner of a computer 
laboratory. The subjects were seated at a desk with a 
telephone, and provided with paper and writing imple- 
ments. All system feedback to the subject was provided 
over the telephone by the AT&T TTS speech synthesizer. 
Speech data were captured simultaneously using (1) a 
Sennheiser HMD-410 close-talking microphone amplified 
by a Shure FPll microphone-to-line amplifier, and (2) 
a standard carbon button microphone (in the telephone 
handset) over local telephone lines. Digitization was per- 
formed by an Ariel Pro-Port A/D system. 
2.3. Data Collection Procedure 
Before a recording session began, the experimenter pro- 
vided the subject with a brief verbal explanation of the 
task and a page of written instructions. The subject 
also received a summary of the task domain and two 
sets of travel planning scenarios. The first set of sce- 
narios included a number of simple tasks (referred to 
below as "short scenarios") and the ATIS common sce- 
nario (used at all five ATIS data collection sites). The 
second set contained more complicated tasks (referred to 
below as "long scenarios"), and subjects were permitted 
to attempt to book flights while working on these scenar- 
ios. Initially, the subjects selected which scenarios they 
wanted to try; because of problems with uneven scenario 
distribution, the experimenter began selecting an initial 
set of scenarios (two short, one long) for each subject. 
Subjects were asked to speak as they would to a human 
being, and to speak in single sentences. They were not 
told that someone was listening to them and typing in 
what they said until after the entire recording session 
was over. A complete session lasted about an hour, in- 
cluding initial instruction, a two part recording session 
with a five minute break, and a debriefing questionnaire. 
During the recording session, the experimenter listened 
to the subject's speech and the system's response. The 
system initiated the dialogue with the prompt, "I'm 
ready to begin a scenario," and responded after every 
utterance with information or an error message. An ex- 
ample of a typical series of interactions is given in Fig- 
ure 5. The experimenter controlled recording from the 
keyboard, starting recording as soon as the system re- 
sponse ended, and stopping recording when the subject 
appeared to have completed a sentence. The experi- 
menter was asked to transcribe exactly what the sub- 
ject said, excluding false starts. However, because of 
(perceived) pressure on the experimenters to get answers 
to the subjects, especially after repeated system failure, 
the session transcriptions sent to the interaction log files 
were not always accurate. Most of the time, the ex- 
perimenter interacted with the subject only through the 
system. However in cases of complete system failure and 
severe subject confusion, the experimenter could com- 
municate directly with the subject, either by sending a 
message through the speech synthesizer, or by speaking 
directly to the subject. 
Subjects for data collection were recruited from local 
civic organizations, and collection took place during 
working hours. As a result, 82% of the subjects were 
female, and subjects ranged in age from 29 to 77, with 
a median age of 55. Approximately 60% of the subjects 
came from the New York City dialect region; all were 
native speakers of English. In return for each subject's 
participation, a donation was made to the civic organi- 
zation through which he or she was recruited. 
Four summer students served as experimenters for al- 
most all of the data collection sessions. They were 
trained for two weeks during pre-collection system de- 
bugging. The system was debugged and intermittently 
upgraded during and after the 2 1/2 week collection. All 
of the data was then transcribed and submitted to NIST 
for distribution. 
3. COMPARATIVE ANALYSES 
In the following section, we present analyses of data col- 
lected at AT&T and the four other data collection sites. 
We will pay particular attention to comparing the AT&T 
data with the MIT data, since the two collection systems 
had identical natural language and database access com- 
ponents, and the two sites used the same travel planning 
scenarios. 
3.1. General Characteristics 
Table 1 presents some statistics which can be used to 
characterize each site's collection system and procedure. 
The statistics were computed for each speaker, and the 
speaker values averaged for each site. AT~zT is in the 
middle of the pack in utterances per scenario, collection 
rate (number of utterances collected divided by recording 
session duration), and human-machine interaction time 
/45 
Scenar~\[o: "Determine the type of aircraft used on a 
flight from one city to another that leaves before (or 
after) a certain time of the day." 
ATIS: l'm ready to begin a scenario. 
Subject: BEGIN SCENARIO THREE 
ATIS: Okay, go ahead. 
Subject: <truncated utterance> 
ATIS: Could you repeat that please? I didnJt catch 
everything you said. 
Subject: YES I WOULD LIKE TO FIND WHAT FLIGHTS ARE 
AVAILABLE FRON ATLANTA TO BOSTON ON THE 
MORNING OF JULY SEVENTH 
ATIS: I'm confused. Could you try to rephrase your 
question? 
Subject: I WANT TO FLY FROM BOSTON TO ATLANTA 
ATIS: There are seventeen flights from boston to 
atlanta. 
There are flights departing between six thirty A H 
and eight twenty four P M. 
What time .ould you like to go? 
Subject: SIX THIRTY A M 
ATIS: There is one flight from boston to atlanta leaving 
six thirty A N. 
delta flight two twenty five from Boston to Atlanta 
departs at six thirty A M and arrives at nine 
thirteen A N with no stops. 
Subject: WHAT TYPE OF AIRCRAFT WILL BE USED 
ATIS: There is aircraft for one flight. 
delta flight two twenty five from Boston to Atlanta 
departs at six thirty A M and arrives at nine 
thirteen A M with no stops. The plane for this 
flight is a Mcdonnell Douglas H D eighty. 
Subject: I WOULD LIKE TO BOOK THAT FLIGHT 
ATIS: There is one flight. 
delta flight two twenty five from Boston to Atlanta 
departs at six thirty A M and arrives at nine 
thirteen A M with no stops. 
I~m sorry, I can help you with travel information, 
but I can't book flights. 
Subject: END SCENARIO 
Figure 5: A typical interaction between a subject and 
the AT&T data collection system. 
(interval from the beginning of one utterance to the be- 
ginning of the next). The average AT&T speaker used 
significantly more words per utterance than the average 
speaker at any other site. This may be due to our sub- 
jects' response to NL system failure, which is discussed 
in section 3.2. AT&T subjects also had a lower aver- 
age speaking rate than speakers from other sites. This 
may be related to the higher disfluency rate (discussed in 
section 3.4) and increases in the frequency of occurrence 
and durations of silent pauses. 
3.2. NL System Failure 
One of the effects of the audio interaction paridigm was 
a higher NL system failure rate (MIT 33.4%, AT&T 
42.9%), where NL system failure is defined as the fail- 
Variables 
# Utterances 
Avg utterances/scenario 
Avg utterances/hour 
Av interaction time 
Av 6 words/utterance 
Av words minute 
~t~m mW~.2i!! mmB~im/ 
ram/ram mmmmm 
/mmmB mmRumm 
mmmmE 
iF~ii~|ill~nn~ | 
Table 1: Summary of general characteristics of data from 
each site. 
ure to completely process an utterance because it con- 
tains unknown words or fails to parse. The change in 
the interaction paradigm changed the NL task; since the 
NL system was designed based on a visual display, the 
NL system failure rate was expected to increase. The 
response of subjects to NL system failure was also af- 
fected by the change to the audio interaction paradigm. 
Table 2 shows the subjects' responses to NL failure at 
AT&T and MIT. Subjects at both AT&T and MIT spoke 
longer sentences and slowed their speaking rates. How- 
ever, the effects of NL failure on subjects' speech are 
more dramatic in the AT&T data: the number of words 
per utterance increased by over 50% (MIT 20%), the 
speaking rate dropped by 15% (MIT 5%), and the utter- 
ance duration increased by over 75% (MIT 25%) when 
compared with utterances which did not follow an NL 
system error. 
Variables AT&T MIT 
Average words/minute: 
Post NL failure 106.8 153.3 
Non-post NL failure 123.8 160.8 
Average words/utterance: 
Post NL failure 14.8 10.,2 
Non-post NL failure 9.8 8.6 
Average seconds/utterance: 
Post NL failure 8.3 4.0 
Non-post NL failure 4.7 3.2 
Table 2: These statistics reflect the subjects' response to 
system failure. 
The large increase in utterance length after NL failure, 
combined with the high NL failure rate, is the main rea- 
son the average AT&T sentence is so much longer than 
the average MIT sentence, both in number of words and 
in duration. However, the reason behind the post-NL- 
failure increase in sentence length is not entirely clear. A 
qualitative examination seems to indicate that the sys- 
tem is not effectively communicating the reason for its 
failure. The NL system usuMly fails as a result of an un- 
familiar, unusual, or u~agrammatical syntactic construc- 
tion. During the initial task familiarization, the subjects 
46 
were told that the system failure was triggered by prob- 
lems with a sentence's grammatical construction, and 
not by any type of recognition problem. Subjects were 
also informed of the system's discourse capabilities. Yet 
when a sentence failed to parse and the subject was asked 
to rephrase his or her request, he or she frequently re- 
sponded by simply tacking on a summary of the previ- 
ous discourse without modifying the syntactic structure 
of the original sentence. In these cases, the subjects ap- 
peared to respond to the NL failure as a discourse failure 
instead of a syntactic failure. Subjects did appear to ad- 
just their speech to the constraints imposed by the NL 
system, as the NL system failure rate decreased "from 
51% in the first scenario to 39% in subsequent scenarios. 
3.3. Vocabulary Comparisons 
Table 3 contains statistics on the increase in lexicon size 
as a function of the number of sentences collected. The 
breakpoint of 600 sentences collected was chosen because 
it was the point at which the vocabulary growth rate re- 
mained less than 30 words/100 sentences for all sites. 
MIT has reached a terminal vocabulary growth rate of 
8.7 new words/100 sentences collected; the vocabulary 
growth rates at the other four sites continue to decrease 
as the number of collected sentences increases. Figure 6 
is a graph of lexicon size vs. number of sentences col- 
lected for each site. 
Figure 7 shows the overlap of the different sites' lexicons. 
Excluding the AT&T data, the percentage of words in 
lexicon X that are found in lexicon Y appears to be 
proportional to the size of lexicon X. However, the per- 
centages of words in the AT&T lexicon that appear in 
the BBN, CMU, and SRI lexicons are lower (by about 
5 percentage points) than predicted, though the over- 
lap with the MIT lexicon matches the prediction fairly 
well. One explanation is that, although the change in 
the interaction paradigm does affect the vocabulary, the 
effect of the change in paradigm is similar to the ef- 
fects of other inter-site system variations. The AT&T 
system differs from the MIT system only in the inter- 
action paradigm, but differs from the BBN, CMU, and 
SRI systems in other ways, in addition to the different 
interaction paradigm. 
Variables AT&T MIT BBN CMU SRI 
Total utterances 1885 4247 1616 1543 1055 
Above 600 sentences* 13.75 12.64 19.70 14.78 15.75 
*For MIT, between 600 and 2000 sentences. 
MIT above 1000:8.72 new words/100 sens 
Table 3: Vocabulary growth: New words per 100 sen- 
tences collected. 
700 t i 
• 00 i 
500 _ .~ ~ 
~z ~.4. .'"'" ! AT&T: ooo 
200- .-,,+o;, . .'" i BBN: rams 
• "~ ~.*.."" i CMU: *~ 
lOO ~.":" i ~ "' 
L , , , _ , , , _ 
0 200 4GO (6{~0 8~0 1000 1200 1400 1600 1800 2~0 
Calls rzl 
Figure 6: Number of sentences vs. vocabulary size. 
J 
85 
8O 
75 
7O 
65 
55 
45 
4~ 
3OO 
i I I ! i 
i i i i 
350 400 450 fRO 550 600 650 700 750 800 
VocJbulmy S~ 
Figure 7: Percentage of inter-site vocabulary overlap. 
3.4. Disfluencies 
Table 4 contains statistics on the occurence of disflu- 
encies as transcribed by each site. Partial words and 
word fragments are counted as lexical false starts, and 
verbally deleted complete words are counted as linguis- 
tic false starts, as in \[2\]. The high percentage of utter- 
ances containing linguistic false starts and filled pauses 
in the AT&T data reflect the subjects' response to 
experimenter-controlled recording and their uncertainty 
about how to interact with the system. Since the sub- 
jects did not take any initiative in starting and stopping 
recording, they were less likely to compose their thoughts 
before they began speaking. The rate of filled pauses and 
false starts did decrease somewhat as the subject became 
more comfortable talking to the system: comparing dis- 
fluency rates for first scenario and last scenario utter- 
ances, the percentage of utterances containing linguistic 
false starts decreased from 13% to 11%, the percentage 
containing lexical false starts from 8% to 7%, and the 
47 
percentage containing filled pauses from 15% to 12%. 
NL system failure strongly affected the rate of linguis- 
tic false starts. The percentage of utterances containing 
linguistic false starts increased from 9.4% after a success- 
fully processed utterance to 14.4% after an NL system 
error. The absence of similar increases in the rates of 
lexical false starts and filled pauses indicates that the 
subjects' speech was disrupted primarily at the syntac- 
tic level. 
% of sentences with AT&T MIT CMU BBN SRI 
linguistic false starts 11.4 3.9 1.2 2.4 4.5 
lexical false starts 7.6 2.8 9.3 2.2 2.8 
filled pauses 13.7 3.1 3.0 1.9 1.5 
Table 4: Percentage of sentences containing various dis- 
fluencies. 
4. DISCUSSION 
Our initial effort in collecting ATIS domain data under a 
speech-in, speech-out interaction paradigm has produced 
some interesting results. A number of issues have come 
up during system development, data collection, and data 
analysis which need to be considered in the development 
of telephone-based spoken language systems and spoken 
language systems in general. 
The system with which we collected data was not ideal. 
Many of the subjects were able to get around the sys- 
tem's limitations, but others had a great deal of trouble. 
As a result, some of the speech we collected sounds per- 
fectly normal, and some sounds exceptionally unnatural 
and unusual, and not like normal conversational speech 
at all. Because of the system's inefficiency, some peo- 
ple found it difficult to keep track of the dialogue. Other 
subjects had problems because of the high system failure 
rate (NL and otherwise), ineffective communication of 
the discourse history, and confusion about system limita- 
tions. Although most subjects said they had no problems 
understanding the output of the speech synthesizer, it 
appeared that many subjects had trouble absorbing and 
remembering the information presented. Many poten- 
tially useful system components were only partially de- 
veloped, and sometimes caused new problems. Because 
the bounds of the task domain as defined by the database 
did not match the bounds inferred by the users based on 
the travel planning scenarios, subjects frequently wan- 
dered out of the domain. 
Developing an interactive system like ATIS under an 
audio-only paradigm is more difficult than develop- 
ing a similar system under a less restrictive feedback 
paradigm. The audio interaction paradigm demands 
more effort on the system's part in compressing and fil- 
tering information before it is presented to the user, and 
in making sure all the information the user needs is eas- 
ily accessible. It is more difficult to communicate system 
limitations and system status to the user, since informa- 
tion cannot be provided continuously or from more than 
one source at a time, and the quantity of information is 
limited to what a user can be expected to absorb and 
remember. Because the perceived waiting time is longer 
with no visual distractions, a system operating under an 
audio feedback paradigm must be efficient, so that the 
flow of the dialogue is maintained and the user remains 
attentive. There must be logical closure in the system's 
capabilities, and the limits of the task domain must be 
obvious to the user. These issues are critical in develop- 
ing telephone-based systems, and many are important in 
the development of any interactive system. 
We intend to continue the development of the ATIS sys- 
tem, particularly in the areas of dialogue and informa- 
tion control. We will also continue to collect data to 
support our research in telephone-based spoken language 
systems, and in support of the speech and language re- 
search community in general. 
5. ACKNOWLEDGEMENTS 
We would like to acknowledge the members of the MIT 
Laboratory for Computer Science Spoken Language Sys- 
tems Group who provided us with the MIT ATIS sys- 
tem and other software assistance. We would also like to 
thank all the people in AT&T Bell Laboratories Center 
1122 who provided technical support and who served as 
experimenters and transcribers. 
REFERENCES 
1. Polifroni, J., S. Seneff, V. W. Zue, and L. Hirschman, 
"ATIS Data Collection at MIT," DARPA SLS Note 8, 
Spoken Language Systems Group, MIT Laboratory for 
Computer Science, Cambridge, MA, November, 1990. 
2. Pohfroni, J., S. Seneff, and V. W. Zue, "Collection of 
Spontaneous Speech for the ATIS Domain and Com- 
paxative Analyses of Data Collected at MIT and TI," 
Proc. Fourth DARPA Speech and Language Workshop, 
P. Price (ed.), Morgan Kaufmann, February 1991. 
3. Seneff, S., L. Hirschman, and V. Zue, "Interactive 
Problem Solving and Dialogue in the ATIS Domain," 
Proc. Fourth DARPA Speech and Language Workshop, 
P. Price (ed.), Morgan Kaufmann, February 1991. 
48 
