See discussions, stats, and author profiles for this publication at: [Link]

net/publication/220816284

SIS: A Shell for Interview Systems.

Conference Paper · January 1987

Source: DBLP

CITATIONS READS

7 17

4 authors, including:

Atsuo Kawaguchi Riichiro Mizoguchi

Ricoh Company, Ltd. Japan Advanced Institute of Science and Technology
23 PUBLICATIONS 496 CITATIONS 580 PUBLICATIONS 8,787 CITATIONS

SEE PROFILE SEE PROFILE

Takahira Yamaguchi
Keio University
183 PUBLICATIONS 1,051 CITATIONS

SEE PROFILE

All content following this page was uploaded by Atsuo Kawaguchi on 27 October 2023.

The user has requested enhancement of the downloaded file.

 SIS:A SHELL FOR INTERVIEW SYSTEMS

A t s u o KAWAGUCHI, R i i c h i r o MIZOGUCHI,
T a k a h i r a YAMAGUCH1 and Osamu KAKUSHO

The I n s t i t u t e o f S c i e n t i f i c and I n d u s t r i a l R e s e a r c h , Osaka

U n i v e r s i t y , 8 - 1 M i h o g a o k a , I b a r a k i , Osaka 567, Japan

Abstract a c q u i s i t i o n s y s t e m , M0RE[2}, r e v e a l s some k n o w l e d g e

An i n t e r v i e w e r has t w o k i n d s of knowledge. One common t o t h e b o t h . This allows us to consider an
i s a b o u t a domain under c o n s i d e r a t i o n and t h e o t h e r interviewer as a n e x p e r t s y s t e m and s u g g e s t s the
is k n o w l e d g e f o r i n t e r v i e w i t s e l f w h i c h makes the possibility o f a g e n e r a l framework for interview
interviewer an expert of interview. The latter s y s t e m s . T h i s paper d e s c r i b e s a S h e l l f o r I n t e r v i e w
seems to be i n d e p e n d e n t of t h e domain because an Systems, SIS for short, based o n t h i s i d e a . Of
experienced interviewer, such as a TV i n t e r v i e w e r , course, t h e r e a r e many o t h e r a s p e c t s o f interview
can c a r r y o n h i s t a s k s i n many f i e l d s . F u r t h e r m o r e , systems and knowledge a c q u i s i t i o n systems such as
w e b e l i e v e t h a t t h e i n t e r v i e w knowledge c o n s i s t s o f R0GET[3], ETS[4], e t c . The a u t h o r s a r e e s p e c i a l l y
several i n t e r v i e w knowledge p r i m i t i v e s . Based on interested in discussion on interviewer's question
this idea, we are developing SIS, a Shell for strategies inherent in interviewing.
I n t e r v i e w Systems w h i c h has seven q u e s t i o n s t r a t e g y V a r i o u s i n t e r v i e w systems can b e c o n s t r u c t e d by
primitives as the knowledge for interviewing. u s i n g SIS w i t h d e s c r i p t i o n o f t h e meta knowledge o f
Generality and effectiveness of SIS are shown the domain and task. SIS has seven q u e s t i o n
through two i m p l e m e n t a t i o n examples o f interview strategy primitives. One can define various
systems, I2s and MORE. The seven p r i m i t i v e s a r e question s t r a t e g i e s (dependent on the task) by
shown to be e f f i c i e n t f o r t h e two system whose s p e c i f y i n g the combination of the p r i m i t i v e s .
domains and t a s k s a r e e n t i r e l y d i f f e r e n t .
2 . SIS: S h e l l f o r I n t e r v i e w Systems
SIS is a shell for constructing interview
1• Introduction systems. A general flow of i n t e r v i e w i n g c o n s i s t s of
Interviewing plays a crucial role i n many three stages. First, an i n t e r v i e w e r requires an
fields t o make ambiguous knowledge clear. In i n t e r v i e w e e o f i n i t i a l i n f o r m a t i o n w h i c h becomes a
building expert systems, for instance, the cue f o r the f o l l o w i n g i n t e r v i e w . The interviewer
h e u r i s t i c s are g e n e r a l l y obtained through i n t e r v i e w constructs an incomplete domain model in this
of a knowledge e n g i n e e r w i t h a domain e x p e r t . In stage. And h e u s u a l l y has many q u e s t i o n s a t this
software s p e c i f i c a t i o n , interview is indispensable time. W e c a l l such q u e s t i o n s ' a t t e n t i o n s ' . I n t h e
t o e x t r a c t i o n o f requirements from customers. second s t a g e , t h e i n t e r v i e w e r asks t h e i n t e r v i e w e e
The a u t h o r s have been i n v e s t i g a t i n g i n t e r v i e w a s additional questions according t o the attentions.
a discourse aiming at transferring (ambiguous) Finally the i n t e r v i e w e r produces an o u t p u t of the
knowledge. The purpose of t h i s research is to interview.
uncover t h e knowledge and mechanism n e c e s s a r y for In order to c a r r y out i n t e r v i e w i n g , an i n t e r v i e w
interview, t o d i s c u s s t h e q u e s t i o n s t r a t e g i e s used system must have the following functions and
in interview and to develop a framework for knowledge f o r t h e m :
i n t e r v i e w systems. The a u t h o r s have d e v e l o p e d I2S (a) to construct a domain model from
(an I n t e l l i g e n t I n t e r v i e w System) whose t a s k i s t o i n f o r m a t i o n e x t r a c t e d from an i n t e r v i e w e e ,
interview a data base customer and to design (b) to make a t t e n t i o n s i n t h e f i r s t and second
logical structure o f t h e d a t a base a s a part of stages,
t h i s study s o f a r [ l l . (c) to i n f e r a b o u t t h e a t t e n t i o n s and t o ask
A n e x p e r i e n c e d knowledge e n g i n e e r can interview t h e i n t e r v i e w e e q u e s t i o n s based o n t h e m ,
domain experts and construct expert systems in ( d ) t o make a n o u t p u t f r o m t h e domain model and
various domains. A TV i n t e r v i e w e r can also ( e ) t o a n a l y z e and g e n e r a t e s e n t e n c e s .
i n t e r v i e w p e r s o n s i n many f i e l d s . These o b s e r v a t i o n On t h e b a s i s of t h e above d i s c u s s i o n s , we have
suggest t h a t t h e y have domain independent skills designed an a r c h i t e c t u r e of SIS. It consists of
(knowledge) f o r i n t e r v i e w i n g . I n t h e development o f f o u r modules such a s p a r s e r , g e n e r a t o r , memory and
I2S, it was a n i m p o r t a n t p o i n t t o d i s t i n g u i s h t h e inference engine. The parser analyzes input
d o m a i n - i n d e p e n d e n t i n t e r v i e w knowledge f o r logical sentences, w h i l e the generator generates sentences
design o f d a t a base f r o m d o m a i n - d e p e n d e n t one and for the interviewee. They employ a demon-based
four domain-independent question strategies were mechanism. The memory contains knowledge for
obtained. interviewing d e s c r i b e d by a system developer and
Moreover, s u c h d o m a i n - i n d e p e n d e n t knowledge f o r information obtained through interview. Seven
i n t e r v i e w i n g dependent on the task consists of modules for question strategy primitives (QPs)
several primitives independent of the task. For built in the memory. They check t h e information
example, a careful c o m p a r i s o n between the four obtained from the interviewee and generate
q u e s t i o n s t r a t e g i e s o f I 2 S and t h o s e o f a knowledge attentions. The i n f e r e n c e e n g i n e i s e s s e n t i a l l y a

Kawaguchi, Mizoguchi, Yamaguchi, and Kakusho 359

production system and processes the attention. QPs can be used to detect many patterns which
Domain model in SIS is a kind of network suggest the necessity of asking questions for more
composed of domain_constituents (DCs) as nodes and information.
domain_links (DLs) as arcs. DCs are represented in
terms of frames, while knowledge for inference 3• Examples
about attention is represented by rules in SIS, An This section describes implementation of two
interview system developer can describe the interview systems, which are entirely different in
knowledge (a), (d) and (e) discussed above by their tasks and domains. Effectiveness and
defining prototypes of these frames with associated generality of QPs w i l l be shown through these
demons for parsing and generating, and knowledge examples.
(c) by defining rules with some extra information
for controlling the system. 3.1 I 2 S[1]
Whenever any information is given to the memory, I 2 S (Intelligent Interview System for logical
it scans the domain model to check if any part of design of data base) interviews a data base
the model satisfies the conditions for making customer and designs logical structure of data base
attentions. The conditions correspond to the according to the following steps:
knowledge (c). They are defined by specifying the (1) It asks questions of the customer
objects of application of QPs, (interviewee) to get requirements of a target
QPs have their own conditions when to make data base as i n i t i a l information,
attentions. They apply the conditions to a l l (2) It constructs a domain model called plan-
objects defined by the system developer and make structure and refines it by asking the
attentions when the conditions are s a t i s f i e d . The customer some questions,
attentions indicate important issue underlying in (3) It makes conceptual schema of the target data
domain model to be discussed with the interviewee. base from the plan-structure as i t s output.
By defining inference rules for the attentions, the I2S employs the concept of plan to represent a
system developer can control the behavior of the domain. In the domain model called plan-structure,
system to the questions. each noun or verb is represented as a frame. A noun
The detailed description of the QPs' conditions frame contains several attributes which the
are shown below. corresponding entity has, while a verb frame has
(1) planner (see Figure 1) description of i t s a c t i v i t y represented in terms of
For paths and DCs obtained by tracing DLs the concept of plan.
from a DC, Various question strategies are used by I2S to
(a) there are prototypes which could be refine a plan-structure and design conceptual
chained to the paths, schema. We have obtained four kinds of strategies
(b) there are more than one path to reach one as follows:
DC or (1) planning,
(c) there are paths from the same DC to (2) making the detailed level of description
different one. equal,
The chainability in (1) is determined from (3) using scripts for nouns and
intersection of predefined slot sets of each (4) interrupting when an interviewee mentioned
frame. 'time'.
(2) script_tracer While a l l of these strategies can be described in
There is a script for asking questions in terms of QPs, we w i l l show an example concerning
a prototype corresponding to a DC. planning strategy.
(3) unknown_object_detector I2S reasons about a c t i v i t i e s to get unknown
Information which is not defined in a c t i v i t i e s or relations between them based on the
prototypes are entered, current plan-structure and shows the interviewee
(4) discord__detector the results. We c a l l this process planning.
Slot values of the DCs corresponding to Suggestions presented by this planning strategy are
each other are not the same, very effective because it is rather d i f f i c u l t for a
(5) ambiguity__detector customer to prepare a l l data items necessary for
There are more than one value in a s l o t , his data base in advance. For instance, suppose a
(6) certainty_manager customer said a fact shown in Figure 2(a). A plan-
A certainty factor assigned to a DL is out structure corresponding to this fact is shown in
of the range the system developer defined. (b). If a dictionary of I2S contains semantic
(7) constraint__checker description of 'stock' as Figure 2(c), I2S finds it
A DCs slot value violates i t s constraint. and asks an interviewee some questions as shown in
(d) because the precondition of 'construct' can be
satisfied by the goal of 'stock'
In order to implement this strategy by SIS, a
system developer defines prototype frames of verbs
with appropriate demons for translating English
sentences into DCs and vice versa. Then he
implements the planning strategy by using planner's
condition (1) to search candidate verbs.
'Found_candidate' attention w i l l be generated when
one of the following conditions are s a t i s f i e d :
(1) there are some prototypes (DCs)
Fig. 1 Planner's conditions. whose goals are equivalent to precondition of

360 KNOWLEDGE ACQUISITION

 a DC (prototype).
(2) there are some DCs whose goals are equivalent
to precondition of a DC,
And f i n a l l y , he describes the knowledge of how to
process the attention for planning strategy in rule
form. After defining a l l of these strategies, SIS
w i l l automatically parse i n i t i a l information into
plan-structure, apply QPs, make attention, infer
about them, and make question and answer with the
interviewee.
The system works in an interview as shown in
Figure 3. Suppose an interviewee mentioned the fact
shown in Figure 2(a) as a requirement. F i r s t , the
parser makes constituents of plan-structure from
interviewee's input (1). Then, they are stored in
the memory. QPs are applied to the constituents and
in this example, the 'construct' DC and 'stock*
prototype satisfy the condition for making
found_candidate attention. The attention is stored
in the attention l i s t . Fig.3 Processing example of planning stragety.
After obtaining the i n i t i a l information, the
system begins to process the attentions. Attentions 'Differentiation' strategy can be implemented in
are taken out from the attention l i s t one by one SIS by using planner's condition (c) to detect not-
(4) and sent to the inference engine. In this differentiated hypotheses. The planner is directed
example, the system asks the interviewee questions to start tracing from a symptom and to find paths
as shown in Figure 2(a). from a symptom to different hypotheses. A domain-
model w i l l be automatically constructed by defining
3.2 M0RE[2] prototype frames of symptoms and hypotheses with
MORE is a knowledge acquisition system for demons for translating input information into
constructing diagnostic expert systems. It f i r s t frames linked each other. A system developer also
asks a domain expert about hypothesis, symptoms and describes rules for checking if there are really
other conditions (as i n i t i a l information), then not differentiating symptoms when 'same_source'
constructs a domain model. A domain model attention is made, and asking an expert new
constructed by MORE is a network in which symptoms.
hypotheses, symptoms and conditions are represented In an interview, when the pair of hypotheses to
as nodes and linked together by ' l i n k s ' or 'paths'. which there are same paths from a symptom are
Next, MORE applies eight question strategies to the found, the attention 'same_source• will be
model, asks the expert and refines i t . Finally, it generated. Then the system w i l l infer how to treat
makes a rule set for the expert system from the the attention using the rules described by the
model. system developer and ask an expert for a new
Among the eight question strategies, additional symptoms.
' d i f f e r e n t i a t i o n ' is one of the most basic ones. A
symptom S is said to differentiate hypothesis HI 4.. Conclusion
from H2 when there is a path from H1 to S, and no We have described SIS, a shell for interview
path from H2 to S. When there is a pair of systems. SIS has seven question strategy primitives
hypotheses for which there is no differentiating for representing domain-dependent knowledge for
symptom, MORE asks a domain expert about i t . interviewing. They are essentially detectors of
problems in the network representing domain model
Some parts are constructed from other parts and can be used as drivers for interviewing. The
which are supplied from other companyies. generality and effectiveness of the primitives have
(a) Fact. been evaluated through the implementation examples
of two interview systems, I2S and MORE.
SIS is under implementation on Hewlett-Packard's
HP-9000 model 320 using C_Prolog.

References:
[1] Kawaguchi, A., et al., "An Intelligent
Interview System for Conceptual Design of
Database," Proc. ECAI'86, 1986.
[2] Kahn, G., Nowlan, S. and McDermott, J . ,
"Strategies for Knowledge Acquisition," IEEE
PAMI, [Link]-7, No. 5, 1985.
[3] Bennett, J. S., "ROGET: A Knowledge-Based
System for Acquiring the Conceptual Structure
Of a Diagnostic Expert System," Journal of
Automated Reasoning, 1, 1985.
[4] Boose, J. H., Expertise Transfer for Expert
System Design. Elsevier, 1986.

Kawaguchi, Mizoguchi, Yamaguchi, and Kakusho 361

View publication stats