Interpreting and Generating Indirect 
Answers 
Nancy Green" 
University of North Carolina at 
Greensboro 
Sandra Carberry t 
University of Delaware 
This paper presents an implemented computational model for interpreting and generating indirect 
answers to yes-no questions in English. Interpretation and generation are treated, respectively, 
as recognition of and construction of a responder's discourse plan for a full answer. An indirect 
answer is the result of the responder providing only part of the planned response, but intending for 
his discourse plan to be recognized by the questioner. Discourse plan construction and recognition 
make use of shared knowledge of discourse strategies, represented in the model by discourse plan 
operators. In the operators, coherence relations are used to characterize types of information 
that may accompany each type of answer. Recognizing a mutually plausible coherence relation 
obtaining between the actual response and a possible direct answer plays an important role in 
recognizing the responder's discourse plan. During generation, stimulus conditions model a 
speaker's motivation for selecting a satellite. Also during generation, the speaker uses his own 
interpretation capability to determine what parts of the plan are inferable by the hearer and thus do 
not need to be explicitly given. The model provides wider coverage than previous computational 
models for generating and interpreting indirect answers and extends the plan-based theory of 
implicature in several ways. 
1. Introduction 
In the following example, 1 Q asks a question in (1)i and R provides the requested 
information in (1)iii, although not explicitly giving (1)ii. (In this paper, we use square 
brackets as in (1)ii to indicate information which, in our judgment, the speaker in- 
tended to convey but did not explicitly state. For consistency, we refer to the ques- 
tioner and responder as Q and R, respectively. For readability, we have standardized 
punctuation and capitalization and have omitted prosodic information from sources 
since it is not used in our model.) 
(1) i. Q: Actually you'll probably get a car won't you as soon as you get 
there? 
ii. R: \[No.\] 
iii. I can't drive. 
Interpreting such responses, which we refer to as indirect answers, requires the hearer 
to derive a conversational implicature (Grice 1975). For example, the inference that R 
Department of Mathematical Sciences, Greensboro, NC 27412-5001 
t Department of Computer and Information Sciences, Newark, DE 19716 1 Based on an example on page 220 in Stenstr6m (1984). The reader may assume that any unattributed 
examples in the paper are constructed. 
(~) 1999 Association for Computational Linguistics 
Computational Linguistics Volume 25, Number 3 
will not get a car on arrival, although licensed by R's use of (1)iii in some discourse 
contexts, is not a semantic consequence of the proposition that R cannot drive. 
According to one study of spoken English (Stenstr6m 1984) (described in Sec- 
tion 2), 13% of responses to certain yes-no questions were indirect answers. Thus, a 
robust dialogue system should be able to interpret indirect answers. Furthermore, there 
are good reasons for generating an indirect answer instead of just a yes or no answer. 
First, an indirect answer may be considered more polite than a direct answer (Brown 
and Levinson 1978). For example, in (1)i, Q has indicated (by the manner in which 
Q expressed the question) that Q believes it likely that R will get a car. By avoiding 
explicit disagreement with this belief, the response in (1)iii would be considered more 
polite than a direct answer of (1)ii. Second, an indirect answer may be more efficient 
than a direct answer. For example, even if (1)ii is given, including (1)iii in R's response 
contributes to efficiency by forestalling and answering a possible follow-up of well, why 
not? from Q, which can be anticipated since the form of Q's question suggests that 
Q may be surprised by a negative answer. Third, an indirect answer may be used to 
avoid misleading Q (Hirschberg 1985), as illustrated in (2). 2 
(2) i. Q: Have you gotten the letters yet? 
ii. R: I've gotten the letter from X. 
This example illustrates a case in which, provided that R had gotten some but not all 
of the letters in question, just yes would be untruthful and just no would be misleading 
(since Q might conclude from the latter that R had gotten none of them). 
We have developed a computational model, implemented in Common LISP, for 
interpreting and generating indirect answers to yes-no questions in English (Green 
1994). By a yes-no question we mean one or more utterances used as a request by Q 
that R convey R's evaluation of the truth of a proposition p. Consisting of one or more 
utterances, an indirect answer is used to convey, yet does not semantically entail, R's 
evaluation of the truth of p, i.e., that p is true, that p is false, that p might be true, that 
p might be false, or that p is partially true. In contrast, a direct answer entails R's eval- 
uation of the truth of p. The model presupposes that Q and R mutually believe that 
Q's question has been understood by R as intended by Q, that Q's question is appro- 
priate, and that R can provide one of the above answers. Furthermore, it is assumed 
that Q and R are engaged in a cooperative and polite task-oriented dialogue. 3 The 
model is based upon examples of uses of direct and indirect answers found in tran- 
scripts of two-person telephone conversations between travel agents and their clients 
(SRI 1992), examples given in previous studies (Brown and Levinson 1978; Hirschberg 
1985; Kiefer 1980; Levinson 1983; Stenstr6m 1984) and constructed examples reflecting 
our judgments. 
To give an overview of the model, generation and interpretation are treated, re- 
spectively, as construction of and recognition of the responder's discourse plan spec- 
ification for a full answer. In general, a discourse plan specification (for the sake of 
brevity, hereafter referred to as discourse plan) explicitly relates a speaker's beliefs 
and discourse goals to his program of communicative actions (Pollack 1990). Dis- 
course plan construction and recognition make use of the beliefs that are presumed 
2 (2) is Hirschberg's example (59). 
3 We assume that it is worthwhile to model politeness-motivated language behavior for both generation 
and interpretation. For example in generation, it would seem to be a desirable trait for a software agent 
that interacts with humans. In interpretation, it would contribute to the robustness of the interpreter. 
390 
Green and Carberry Indirect Answers 
to be shared by the participants, as well as shared knowledge of discourse strategies, 
represented in the model by a set of discourse plan operators encoding generic pro- 
grams of communicative actions for conveying full answers. A full answer consists 
of a direct answer, which we refer to as the nucleus, and "extra" appropriate infor- 
mation, which we refer to as the satellite(s). 4 In the operators, coherence relations 
are used to characterize types of satellites that may accompany each type of answer. 
Stimulus conditions are used to characterize the speaker's motivation for including a 
satellite. An indirect answer is the result of the speaker (R) expressing only part of the 
planned response, i.e., omitting the direct answer (and possibly more), but intending 
for his discourse plan to be recognized by the hearer (Q). Furthermore, we argue that 
because of the role of interpretation in generation, Q's belief that R intended for Q to 
recognize the answer is warranted by Q's recognition of the plan. 
The inputs to the interpretation component of the model (a model of Q's inter- 
pretation of an indirect answer) are the semantic representation of the questioned 
proposition, the semantic representation of the utterances given by R during R's turn, 
shared pragmatic knowledge, and Q's beliefs, including those presumed by Q to be 
shared with R. (Beliefs presumed by an agent to be shared by another agent are here- 
after referred to as shared beliefs, and those that are not presumed to be shared as 
nonshared beliefs). 5 The output is a set of alternative discourse plans that might be 
ascribed to R by Q, ranked by plausibility. R's inferred discourse plan provides the 
intended answer and possibly other information about R's beliefs and intentions. The 
inputs to the generation component (a model of R's construction of a response) are the 
semantic representation of the questioned proposition, shared pragmatic knowledge, 
and R's beliefs (both shared and nonshared). The output of generation is R's discourse 
plan for a full answer, including a specification of which parts of the plan do not need 
to be explicitly given by R, i.e., which parts should be inferable by Q from the rest of 
the answer. 6 
This paper describes the knowledge and processes provided in our model for 
interpreting and generating indirect answers. (The model is not intended as a cogni- 
tive model, i.e., we are not claiming that it reflects the participants' cognitive states 
during the time course of comprehension and generation. Rather, its purpose is to 
compute the end products of comprehension and generation, and to contribute to a 
computational theory of conversational implicature.) As background, Section 2 de- 
scribes some relevant generalizations about questions and answers in English. Sec- 
tion 3 describes the reversible knowledge in our model, i.e., knowledge used both in 
interpretation and generation of indirect answers. Sections 4 and 5 describe the inter- 
pretation and generation components, respectively. Section 5 includes a description of 
additional pragmatic knowledge required for generation. Section 6 provides an eval- 
uation of the work. Finally, the last section discusses future research and provides a 
summary. 
4 This terminology was adopted from Rhetorical Structure Theory (Mann and Thompson 1983, 1988), 
discussed in Section 2. 5 Our notion of shared belief is similar to the notion of one-sided mutual belief (Clark and Marshall 
1981). However, following Thomason (1990), a shared belief is merely represented in the conversational 
record as if it were mutually believed, although each participant need not actually believe it. 6 However, our model does not address the interesting question of under what conditions a direct 
answer should be given explicitly even when it is inferable from other parts of the response. For some 
related work on the function of redundant information, see Walker (1993). 
391 
Computational Linguistics Volume 25, Number 3 
2. Background 
This section begins with some results of a corpus-based study of questions and re- 
sponses in English that provide the motivation for the notion of a full answer in 
our model. Next, we describe informally how coherence relations (similar to subject- 
matter relations of Rhetorical Structure Theory \[Mann and Thompson 1983, 1988\]) are 
used to characterize the possible types of indirect answers handled in our model. 
2.1 Descriptive Study of Questions and Responses 
Stenstr6m (1984) describes characteristics of questions and responses in English, based 
on her study of a corpus of 25 conversations (face-to-face and telephone). She found 
that 13% of responses to polar questions (typically expressed as subject-auxilliary in- 
verted questions) were indirect answers, and that 7% of responses to requests for con- 
firmation (expressed as tag-questions and declaratives) were indirect. 7 Furthermore, 
she points out the similarity in function of indirect answers to the extra information, 
referred to as qualify acts in her classification scheme, often accompanying direct an- 
swers (Stenstr/)m 1984). 8 Stenstr6m notes that both are used 
• to answer an implicit wh-question, as in (3), 9 
(3) i. Q: Isn't your country seat there somewhere? 
ii. R: \[Yes/No\]. 
iii. Stoke d'Abernon. 
• for social reasons, as in (4), 
(4) i. Q: Did you go to his lectures? 
ii. R: \[Yes.\] 
iii. Oh he had a really caustic sense of humour actually. 
• to provide an explanation, as in (5), 
(5) i. Q: And also did you find my blue and green striped tie? 
ii. R: \[No.\] 
iii. I haven't looked for it. 
(6) 
or to provide clarification, as in (6). 
i. Q: I don't think you've been upstairs yet. 
ii. R: \[Yes, I have been upstairs.\] 
iii. Um only just to the loo. 
In the above examples, coherence would not be affected by making the associated 
direct answer explicit. She suggests that the main distinction between qualify acts and 
indirect answers is the absence or presence of a direct answer. 
7 Both of these types of requests are classified as yes-no questions in our model. Also, in Stenstr6m's 
scheme, an utterance may be classified as performing more than one function. For example, an 
utterance may be classified as both a polar question and a request for identification (i.e., an implicit 
wh-question). 8 Other types of acts noted by StenstrOm as possibly accompanying direct answers, 
amplify and expand, 
are not relevant to the problem of modeling indirect answers, 9 (3), (4), (5), and (6) are based on StenstrOm's (65), (67), (68), and (142), respectively. In (3) either a 
yes or 
no could be conveyed, depending upon how there is interpreted and shared background knowledge 
about the location of Stoke d'Abernon. 
392 
Green and Carberry Indirect Answers 
Thus, in our model, the notion of a full answer is used to model both indirect 
answers and direct answers accompanied by qualify acts. A full answer consists of a 
direct answer, which we refer to as the nucleus, and possibly extra information of var- 
ious types, which we refer to as satellites} ° Then, an indirect answer can be modeled 
as the result of R giving one or more satellites of the full answer, without giving the 
nucleus explicitly, but intending for the full answer to be recognized. A benefit of this 
approach is that it also can be used to model the generation of qualify acts accom- 
panying direct answers. (That is, a qualify act would be a result of R providing the 
satellite(s) along with an explicit nucleus.) In the next section, we informally describe 
how different types of satellites of full answers (i.e., types of indirect answers) can be 
characterized. 
2.2 Characterizing Types of Indirect Answers 
Consider the constructed responses shown in (1) through (5) of Table 1, which are 
representative of the types of full answers handled in our model, u The (a) sentences 
are yes-no questions and each (b) sentence expresses a possible type of direct answer. 12 
Each of the sentences labeled (c) through (e) could accompany the preceding (b) sen- 
tence in a full answer, ~3 or could be used without (b), i.e., as an indirect answer used 
to convey the answer given in (b). Also, to the right of each of the (c)-(e) sentences is 
a name intended to suggest the type of relation holding between that sentence and the 
associated (b) sentence. For example, (lc) provides a condition for the truth of (lb), 
(ld) elaborates upon (lb), and (le) provides the agent's motivation for (lb). Many of 
these relations are similar to the subject-matter relations of Rhetorical Structure The- 
ory (RST) (Mann and Thompson 1983, 1988), a general theory of discourse coherence. 
Thus, we refer to these as coherence relations. Other sentences providing the same 
type of information, i.e., satisfying the same coherence relation, could be substituted 
for each (c)-(e) sentence without destroying coherence. For example, another plau- 
sible condition could be substituted for (lc). Thus, as this table illustrates, a small 
set of coherence relations characterizes a wide range of possible indirect answers} 4 
Furthermore, as it illustrates, certain coherence relations are characteristic of only one 
or two types of answer, e.g., giving a cause instead of yes, or an obstacle instead 
of no. 
To give a brief overview of Rhetorical Structure Theory as it relates to our model, 
one of the goals of RST is to provide a set of relations for describing the organization of 
coherent text. An RST relation is defined as a relation between two text spans, called 
the nucleus and satellite. The nucleus is the span which is "more essential to the 
writer's purpose \[than the satellite is\]" (Mann and Thompson 1988, 266). A relation 
definition provides a set of constraints on the nucleus and satellite, and an effect 
field. According to RST, implicit relational propositions are conveyed in discourse. 
10 As noted earlier, this terminology is borrowed from Rhetorical Structure Theory, described below. 11 Constructed examples are used here to provide a concise means of demonstrating the classes of 
satellites. 12 Specifically, the possible types of direct answers handled in the model are: (lb) that p is true, (2b) that p 
is false, (3b) that there is some truth to p, (4b) that p may be true, or (5b) that p may be false, where p is 
the questioned proposition. 13 When more than one of the (c)-(e) sentences is used in the same response, coherence may be improved 
by use of discourse connectives. 14 However, we are not claiming that this set is exhaustive, i.e., that it characterizes all possible indirect 
answers. 
393 
Computational Linguistics Volume 25, Number 3 
Table 1 
Examples of coherence relations in full answers. 
1. 
2. 
3. 
4. 
5. 
a. Are you going shopping tonight? 
b. Yes. 
c. If I finish my homework. 
d. I'm going to the mall. 
e. I need new running shoes. 
a. Aren't you going shopping tonight? 
b. No. 
c. I wouldn't have enough time to study. 
d. My car's not running. 
e. I'm going tomorrow night. 
a. Is dinner ready? 
b. To some extent. 
c. The pizza is ready. 
a. Is Lynn here? 
b. I think so. 
c. Her books are here. 
d. She's usually here all day. 
e. I think she has a meeting here at 5. 
a. Is Lynn here? 
b. I don't think so. 
c. Her books are gone. 
d. She's not usually here this late. 
e. I think she has a dentist appointment 
this afternoon. 
Condition 
Elaboration 
Cause 
Otherwise 
Obstacle 
Contrast 
Contrast 
Result 
Usually 
Possible Cause 
Result 
Usually 
Possible Obstacle 
For example, (7) conveys, in addition to the propositional content of (7)i and (7)ii, the 
relational proposition that the 1899 Duryea is in the writer's collection of classic cars. 15 
(7) i. I love to collect classic automobiles. 
ii. My favorite car is my 1899 Duryea. 
Such relational propositions are described in RST in a relation definition's effect field. 
The organization of (7) would be described in RST by the relation of Elaboration, 
where (7)i is the nucleus and (7)ii a satellite. To see the usefulness of RST for the 
analysis of full answers to yes-no questions, consider (8). 
(8) i. Q: Do you collect classic automobiles? 
ii. R: Yes. 
iii. I recently purchased an Austin-Healey 3000. 
Although (8)ii is not semantically entailed by (8)iii, R could use (8)iii alone in response 
to (8)i to conversationally implicate (8)ii. Further, just as (7)ii provides an elaboration 
15 This example is from Mann and Thompson (1983), page 81. 
394 
Green and Carberry Indirect Answers 
Table 2 
Similar RST relations. 
Coherence Relation 
Cause 
Condition 
Contrast 
Elaboration 
Obstacle 
Otherwise 
Possible-cause 
Possible-obstacle 
Result 
Usually 
Similar RST Relation Name(s) 
Non-Volitional Cause, Purpose, Volitional Cause 
Condition 
Contrast 
Elaboration 
Otherwise 
Non-Volitional Result, Volitional Result 
of (7)i, (8)iii provides an elaboration of (8)ii, whether (8)ii is given explicitly as an 
answer or not. 16 Also, in giving just (8)iii as a response, R intends Q to recognize not 
only (8)ii but also this relation, i.e., that the car is part of R's collection. 
Table 2 lists, for each of the coherence relations defined in our model (shown in 
the left-hand column), similar RST relations (shown in the right-hand column), if any. 
Although other RST relations can be used to describe other parts of a response (e.g., 
Restatement), only relations that contribute to the interpretation of indirect answers are 
included in our model. The formal representation of the coherence relations provided 
in our model is discussed in Section 3. 
3. Reversible Knowledge 
As shown informally in the previous section, coherence relations can be used to char- 
acterize various types of satellites of full answers. Coherence rules, described in Sec- 
tion 3.1, provide sufficient conditions for the mutual plausibility of a coherence rela- 
tion. During generation, plausibility of a coherence relation is evaluated with respect 
to the beliefs that R presumes to be shared with Q. During interpretation, the same 
rules are evaluated with respect to the beliefs Q presumes to be shared with R. Thus, 
during generation R assumes that a coherence relation that is plausible with respect 
to his shared beliefs would be plausible to Q as well. That is, Q ought to be able to 
recognize the implicit relation between the nucleus and satellite. 
However, the generation and interpretation of indirect answers requires additional 
knowledge. For example, for R's contribution to be recognized as an answer, there 
must be a discourse expectation (Levinson 1983; Reichman 1985) of an answer. Also, 
during interpretation, for a particular answer to be licensed by R, the attribution of 
R's intention to convey that answer must be consistent with Q's beliefs about R's 
intentions. For example, a putative implicature that p holds would not be licensed 
if R provides a disclaimer that it is not R's intention to convey that p holds. This 
and other types of knowledge about full answers is represented as discourse plan 
operators, described in Section 3.2. In our model, a discourse plan operator captures 
shared, domain-independent knowledge that is used, along with coherence rules, by 
16 This may seem to conflict with the idea in RST that the nucleus, being more essential to the writer's 
purpose than a satellite, cannot be omitted. However, at least in the case of the coherence relations 
playing a role in our model, it appears that the nucleus need not be given explicitly when it is inferable in the discourse context. 
395 
Computational Linguistics Volume 25, Number 3 
It is mutually plausible to the agent that (cr-obstacle q p) holds, 
where q is the proposition that a state Sq does not hold during time period tq, 
and p is the proposition that an event e v does not occur during time period t v, 
if the agent believes it to be mutually believed that Sq is a precondition 
of a typical plan for doing ev, 
and that tq is before or includes tv, 
unless it is mutually believed that sq does hold during tq, 
or that ep does occur during tp. 
It is mutually plausible to the agent that (cr-obstacle q p) holds, 
where q is the proposition that a state sq holds during time period tq, 
and p is the proposition that a state sv does not hold during time period tv, 
if the agent believes it to be nmtually believed that 8q typically prevents sp, 
and that tq is before or includes tv, 
unless it is mutually believed that Sq does not hold during lq, 
or that s v does hold during t v. 
Figure 1 
Glosses of two coherence rules for cr-obstacle. 
the generation component to construct a discourse plan for a full answer. Interpreta- 
tion is modeled as inference of R's discourse plan from R's response using the same 
set of discourse plan operators and coherence rules. Inference of R's discourse plan 
can account for how Q derives an implicated answer, since a discourse plan explicitly 
represents the relationship of R's communicative acts to R's beliefs and intentions. 
Together, the coherence rules and discourse plan operators described in this section 
make up the reversible pragmatic knowledge, i.e., pragmatic knowledge used by both 
the generation and interpretation components, of the model. Other pragmatic knowl- 
edge, used only by the generation process to constrain content planning, is presented 
in Section 5. 
3.1 Coherence Rules 
Coherence rules specify sufficient conditions for the plausibility to an agent with re- 
spect to the agent's shared beliefs (which we hereafter refer to as the mutual plausi- 
bility) of a relational proposition (CR q p), where CR is a coherence relation and q and 
p are propositions. (Thus, if the relational proposition is plausible to R with respect to 
the beliefs that R presumes to be shared with Q, R assumes that it would be plausible 
to Q, too.) To give some examples, glosses of some rules for the coherence relation, 
which we refer to as cr-obstacle are given in Figure 1.17 The first rule characterizes a 
subclass of cr-obstacle, illustrated in (9), relating the nonoccurrence of an agent's voli- 
tional action (reported in (9)ii) to the failure of a precondition (reported in (9)iii) of a 
potential plan for doing the action. 
(9) i. Q: Are you going to campus tonight? 
ii. R: No. 
iii. My car's not running. 
17 For readability, we have omitted the prefix cr- in Tables 1 and 2. 
396 
Green and Carberry Indirect Answers 
In other words, it is mutually plausible to an agent that the propositions conveyed in 
(9)iii and (9)ii are related by cr-obstacle, provided that the agent has a shared belief that 
a typical plan for R to go to campus has a precondition that R's car is running. The 
second rule in Figure 1 characterizes another subclass of cr-obstacle, illustrated in (10), 
relating the failure of one condition (reported in (10)i) to the satisfaction of another 
condition (reported in (10)ii). 
(10) i. R: My car's not running. 
ii. The timing belt is broken. 
In other words, it is mutually plausible to an agent that the propositions conveyed in 
(10)ii and (10)i are related by cr-obstacle, provided that the agent has a shared belief 
that having a broken timing belt typically prevents a car from running. 
Coherence rules are evaluated with respect to an agent's shared beliefs. Coherence 
rules and the agent's beliefs are encoded as Horn clauses in the implementation of 
our model. The sources of an agent's shared beliefs include: 
terminological knowledge: e.g., that driving a car is a type of action, 
domain knowledge, including 
-- domain planning knowledge: e.g., that a subaction of a typical 
plan to go to campus is to drive to campus, and that a typical 
plan for driving a car has a precondition that the car is running, 
-- other domain knowledge: e.g., that a broken timing belt 
typically prevents a car from running, and 
• the discourse context: e.g., that R has asserted that R's car is not running. 
3.2 Discourse Plan Operators 
The discourse plan operators provided in the model encode generic programs for 
expressing full answers (and subcomponents of full answers). TM For example, the dis- 
course plan operators for constructing full yes (Answer-yes) and full no (Answer-no) 
answers are shown in Figure 2.19 
The first line of a discourse plan operator, its header, e.g., (Answer-yes s h ?p), gives 
the type of discourse action, the participants (s denotes the speaker and h the hearer), 
and a propositional variable. (Propositional variables are denoted by symbols prefixed 
with "?".) In top-level operators such as Answer-yes and Answer-no, the header vari- 
able would be instantiated with the questioned proposition. Applicability conditions, 
when instantiated, specify necessary conditions for appropriate use of a discourse plan 
operator. 2° For example, the first applicability condition of Answer-yes and Answer-no 
requires the speaker and hearer to share the discourse expectation that the speaker will 
inform the hearer of the speaker's evaluation of the truth of the questioned proposition 
p. Present in each of the five top-level answer operators, this particular applicability 
condition restricts the use of these operators to contexts where an answer is expected, 
18 The particular formalism we adopted to encode the operators was chosen to provide a concise and 
perspicuous organization of the knowledge required for our interpretation and generation components. We make no further claims about the formalism itself. 
19 There are three other "top-level" operators in the model for expressing the remaining types of full 
answers illustrated in Table 1. 20 In general, an applicability condition is a condition that must hold for a plan operator to be invoked, 
but that a planner will not attempt to bring about (Carberry 1990). 
397 
Computational Linguistics Volume 25, Number 3 
(Answer-yes s h ?p): 
Applicability conditions: 
(discourse-expectation 
(informif s h ?p)) 
(bel s ?p) 
Nucleus: 
(inform s h ?p) 
Satellites: 
(Use-condition s h ?p) 
(Use-elaboration s h ?p) 
(Use-cause s h ?p) 
Primary goals: 
(BMB h s ?p) 
Figure 2 
(Answer-no s h ?p) 
Applicability conditions: 
(discourse-expectation 
(informif s h ?p) 
(bel s (not ?p)) 
Nucleus: 
(inform s h (not ?p)) 
Satellites: 
(Use-otherwise s h (not ?p)) 
(Use-obstacle s h (not ?p)) 
(Use-contrast s h (not ?p)) 
Primary goals: 
(BMB h s (not ?p)) 
Discourse plan operators for yes and no answers. 
and is needed to account for the hearer's attempt to interpret a response as an an- 
swer, even when it is not a direct answer, m The second applicability condition of the 
top-level operators requires the speaker to hold the evaluation of p to be conveyed; 
e.g., in Answer-no it requires that the speaker believe that p is false. The primary goals 
of a discourse plan specify the discourse goals that the speaker intends for the hearer 
to recognize, n For example, the primary goal of Answer-yes can be glossed as the goal 
that Q will accept the yes answer, at least for the purposes of the conversation. 23 
The nucleus and satellites of a discourse plan describe primitive or nonprimitive 
acts to be performed to achieve the primary goals of the plan. 24 Inform is a primitive 
act that can be realized directly. The nonprimitive acts are defined by discourse plan 
operators themselves. (Thus, a discourse plan may have a hierarchical structure.) A 
full answer may contain zero, one, or more instances of each type of satellite, and the 
default (but not required) order of nucleus and satellites in a full answer is the order 
given in the corresponding operator. 
Consider the Use-elaboration and Use-obstacle discourse plan operators, shown in 
Figure 3, describing possible satellites of Answer-yes and Answer-no, respectively. All 
satellite operators include a second propositional variable referred to as the existential 
21 Without recourse to the notion of discourse expectation, it is difficult to account for the interpretation 
in (9)iii of My car's not running as The speaker is not going to campus tonight, while blocking interpretations 
such as The speaker will rent a car. Note that the latter interpretation may be licensed when the discourse 
expectation is that R will provide an answer to Are you going to rent a car? In general, discourse 
expectations provide a contextual constraint on what inferences are licensed by the speaker. (Similarly, 
it has been argued that scalar implicatures depend on the existence of a salient partially ordered set in 
the discourse context; see Section 4.3.) For a discussion of the overall role of discourse expectations in 
our model, see Section 4.2. One might argue that this type of applicability condition limits the 
generality of the operators and thus, could lead to a proliferation of context-specific operators, which 
would result in inefficient processing. First, we are not claiming that all discourse operators require this 
type of applicability condition, only those operators characterizing discourse-expectation-motivated 
units of discourse. Second, with an indexing scheme sensitive to discourse expectations, this would not 
necessarily lead to efficiency problems. 
22 We refer to these as primary to distinguish them from other discourse goals the speaker may have but 
that he does not necessarily intend for the hearer to recognize. 
23 During interpretation (see Section 4.1), in order for the implicature to be licensed, the applicability 
conditions and primary goals of any plan ascribed to R must be consistent with Q's beliefs about R's 
beliefs and goals. Thus, applicability conditions and primary goals play an important role in canceling 
spurious putative implicatures. 24 The discourse plan operators in our model are not intended to describe all acts that may accompany a 
direct answer. For example, the model does not address the generation of parts of the response, such as 
repetition or restatement, which entail the answer. 
398 
Green and Carberry Indirect Answers 
(Use-elaboration s h ?p): 
Existential variable: ?q 
Applicability conditions: 
(bel s (cr-elaboration ?q ?p)) 
(Plausible (cr-elaboration ?q ?p)) 
Nucleus: 
(inform s h ?q) 
Satellites: 
(Use-cause s h ?q) 
(Use-elaboration s h ?q) 
Primary goals: 
(BMB h s (cr-elaboration ?q ?p)) 
Figure 3 
Two satellite discourse plan operators. 
(Use-obstacle s h ?p): 
Existential variable: ?q 
Applicability conditions: 
(bel s (cr-obstacle ?q ?p)) 
(Plausible (cr-obstacle ?q ?p)) 
Nucleus: 
(inform s h ?q) 
Satellites: 
(Use-obstacle s h ?q) 
(Use-elaboration s h ?q) 
Primary goals: 
(BMB h s (cr-obstacle ?q ?p)) 
variable. For example, (9)ii-(9)iii could be described by a plan constructed from an 
Answer-no discourse plan operator 
whose header variable is instantiated with the proposition p that R is 
going to campus tonight, and 
which has a satellite constructed from a Use-obstacle discourse plan 
operator whose header variable is instantiated with (not p), the 
proposition that R is not going to campus tonight, and whose existential 
variable q is instantiated with the proposition that R's car is not running. 
In general, each satellite operator in our model has applicability conditions and 
primary goals analogous to those shown in Figure 3. (Each satellite operator has a 
name of the form, Use-CR, where CR is the name of a coherence relation.) The first ap- 
plicability condition of a satellite operator, Use-CR, requires that the speaker believes 
that the relational proposition (CR q p) holds for propositions q and p instantiating the 
existential variable and header variable, respectively. The second applicability condi- 
tion requires that, given the beliefs that the speaker presumes to be shared with the 
hearer, this relational proposition is plausible. (Mutual plausibility is evaluated using 
the coherence rules described in Section 3.1.) The primary goal of a satellite operator 
can be glossed as the goal that the hearer will accept the relational proposition. 
4. Interpretation 
This section describes the interpretation process. In our model, implicated answers 
are derived by an answer recognizer. Algorithms for the answer recognizer are de- 
scribed in Section 4.1. Of course, dialogue consists of more than questions and answers. 
Section 4.2 describes the role of the answer recognizer in a discourse-processing ar- 
chitecture. Finally, Section 4.3 discusses how this model relates to previous models of 
conversational implicature. 
4.1 Answer Recognizer 
4.1.1 Main Algorithm. The structure of the answer recognizer is shown in Figure 4. 
The inputs to the answer recognizer include: 
• the set of discourse plan operators and coherence rules described in 
Section 3, 
399 
Computational Linguistics Volume 25, Number 3 
I DiscoursePlanl 
Operators 
I Q's Shared ~_~ 
Beliefs 
Q's question 
Discourse Expectation 
R's turn 
_l Top-down Plan 
r Recognition 
I I 
Theorem t -\[ Hypothesis 
Prover Generation 
Coherence Rules I 
Figure 4 Structure of the answer recognizer. 
~~I Ranked Set of 
--~ Plan Ranking Candidate 
Discourse Plans 
• Q's beliefs (including the discourse expectation that R will provide an 
answer to the questioned proposition p), 
• the semantic representation of p, and 
• for each utterance performed by R during R's turn, the type of 
communicative act signaled by its form (e.g., to inform), and the 
semantic representation of its content. 25 
Answer recognition is performed in two phases. The goal of the first phase is to derive 
a set of candidate discourse plans plausibly underlying R's response. The first phase 
makes use of two subcomponents: one that we refer to as the hypothesis generation 
component, and a theorem prover. The output of the first phase of answer recognition 
is a set of candidate discourse plans since there may be alternate interpretations of R's 
response. The goal of the second phase of answer recognition is to evaluate the relative 
plausibility of each candidate discourse plan. The final output of answer recognition 
consists of a partially ordered set of the candidates ranked by plausibility. 
Plan recognition is primarily top-down, i.e., expectation-driven. More specifically, 
Q26 attempts to interpret the response as having been generated from a discourse 
plan constructed from the discourse plan operators for full answers. The problem of 
reconstructing R's discourse plan has several aspects (to be described in more detail 
shortly): 
• Instantiating discourse plan operators with the questioned proposition 
and appropriate propositions from R's response. 
25 The turn in question need not be the turn immediately following Q's asking of the question, as 
discussed in Section 4.2. Also, we make the simplifying assumption that R's answer is given within a 
single turn. 
26 For convenience, we refer to the answer recognizer component as Q, and to the answer generator as R. 
400 
Green and Carberry Indirect Answers 
• Consistency checking: determining whether the beliefs and goals that 
would be attributed to R by virtue of ascribing a particular discourse 
plan to R are consistent with Q's beliefs about R's beliefs and goals. 
• Coherence evaluation: determining whether a putative satellite of a 
candidate plan is plausibly coherent, i.e., given a candidate plan's (or 
subplan's) nucleus proposition p, putative satellite proposition q, and the 
putative satellite's coherence relation CR, determining whether Q 
believes that (CR q p) is mutually plausible. Coherence evaluation makes 
use of the coherence rules described in Section 3.1. 
• Hypothesis generation: hypothesizing any "missing parts" of the 
response that are required in order to assimilate acts in R's response into 
a coherent candidate plan. Hypothesis generation also makes use of the 
coherence rules. 
Initially, the header variable of each "top-level" answer discourse plan operator 27 
is instantiated with the questioned proposition p, i.e., all occurrences of the header 
variable are replaced with p. Next, consistency checking is performed to eliminate 
any candidates whose applicability conditions or primary goals are not consistent 
with Q's beliefs about R's beliefs and goals. For all remaining candidates, the answer 
recognizer next attempts to recognize an act from R's turn as the nucleus of the plan, 
i.e., to check whether R gave a direct answer. If no acts in R's turn match the nucleus, 
then the nucleus is marked as hypothesized. For all remaining acts in R's turn, the 
answer recognizer attempts to recognize all possible satellites, as specified in each 
remaining candidate plan. In the model the discourse plan operators do not specify a 
required ordering of satellites. 2s The subprocedure of satellite recognition is described 
in more detail in Section 4.1.2. 
4.1.2 Satellite Recognition. Satellite recognition is the (recursive) process of recogniz- 
ing an instance of a satellite of a candidate plan. The inputs consist of: 
• sat-op, a discourse plan operator for a possible satellite, 
• the proposition p conveyed by the nucleus of the higher-level plan (i.e., 
the plan whose satellites are currently being recognized), 
• act-list, a list of acts in R's turn that have not yet been assimilated into 
the candidate plan, 
• cur-act, the current act (inform s h q) in act-list, where s is the speaker, h is 
the hearer, and q is the propositional content of the act. 
The output is a set (possibly empty) of candidate instances of sat-op. To give a sim- 
plified, preliminary version of the algorithm, first, the header variable and existential 
variable of sat-op are instantiated with p and q, respectively. Then, coherence evalu- 
ation and consistency checking are performed. If successful, cur-act is recognized to 
27 Five of these are defined in our model, corresponding to the five types of answers illustrated in Table 1. 28 The operators do specify a preferred order, however, which is used in generation. Also, our process 
model includes a structural constraint on satellite ordering. During interpretation, only instances 
satisfying this constraint are considered. That is, the constraint eliminates interpretations which, in our 
judgment, are not plausible due to incoherence. For a description of the constraint, see Green (1994). We expect that other such constraints may be incorporated into the process model. 
401 
Computational Linguistics Volume 25, Number 3 
Answer-no 
\[ii\] Use-obstacle 
\[iii\] Use-obstacle 
1v 
Figure 5 
Candidate discourse plan with hypotheses. 
be the nucleus of sat-op, and for each remaining act in act-list, satellite recognition is 
performed for each satellite of sat-op. 
However, the satellite recognition algorithm as described so far would not be able 
to handle R's response in (11), since there is no plausible coherence relation in the 
model directly relating (11)iv to (11)ii (or to any other direct answer that could be 
recognized in the model). 
(11) i. Q: Are you driving to campus tonight? 
ii. R: \[No.\] 
iii. \[My car's not running.\] 
iv. My car has a broken timing belt. 
Whenever the answer recognizer is unable to recognize cur-act as the nucleus of sat-op, 
a subprocedure we refer to as hypothesis generation is invoked. Hypothesis genera- 
tion will be described in detail in the following section. It returns a set of alternative 
hypothesized propositions, each of which represents the content of a possible implicit 
inform act to be inserted at the current point of expanding the candidate plan. 29 In 
this example, the proposition conveyed in (11)iii would be returned as a hypothesized 
proposition, which is used to instantiate the existential variable of a Use-obstacle satel- 
lite, thereby enabling satellite recognition to proceed. Then, (11)iv can be recognized 
(without hypothesis generation being required) as a satellite of (11)iii. Ultimately, the 
plan shown in Figure 5 would be inferred. (Only the hierarchical structure and com- 
municative acts are shown. By convention, the left-most child of a node is the nucleus 
and its siblings are the satellites. Labels of sentences in (11) that could realize a leaf 
node are used to label the node. Hypothesized nodes are indicated by square brack- 
ets.) The complete satellite recognition algorithm, employing hypothesis generation, 
is given in Figure 6. 
29 Thus, hypothesis generation may provide additional inferences, i.e., more than just the implicated 
answer. Hinkelman (1989) refers to such implicatures, licensed by attributing a plan to an agent, as 
plan-based implicatures. 
402 
Green and Carberry Indirect Answers 
INPUT 
OUTPUT 
p: proposition from nucleus of higher-level plan 
cur-act: current act, (inform s h q), to be recognized 
act-list: list of remaining acts in R's turn 
op: discourse plan operator (Use-CR s h ?p) 
sat-cand-set: set of candidate instances of op 
underlying part of R's response 
1. Instantiate header variable ?p of op with p. 
2. Instantiate existential variable ?q of op with q of cur-act. 
a. Prove that it is plausible that q and p are related by CR. 
If not, go to step 2c. 
b. Check consistency. If not consistent, then go to step 2c; 
else go to step 3a. 
c. Try substituting each q returned by hypothesis generation for ?q: 
Check consistency and coherence as in steps 2a and 2b. 
For each q passing both checks, proceed with step 3b. 
If none pass, then fail. 
3. a. Mark cur-act as used. Go to step 4. 
b. Mark nucleus as hypothesized. 
4. For each unused act in act-list, attempt to recognize each satellite 
of op. 
Figure 6 
Satellite recognition algorithm. 
4.1.3 Hypothesis Generation. Based upon the assumption that the response is co- 
herent, the goal of hypothesis generation is to fill in missing parts of a candidate 
plan in such a way that an utterance in R's turn can be recognized as part of the 
plan. The use of hypothesis generation broadens the coverage of our model to cases 
where more is missing from a full answer than just the nucleus of a top-level op- 
erator. (From the point of view of generation, it enables the construction of a more 
concise, though no less informative, response.) The hypothesis generation algorithm 
constructs chains of mutually plausible propositions, each beginning with the propo- 
sition (e.g., the proposition conveyed in (11)iv) to be related to a goal proposition in 
a candidate plan (e.g., the proposition conveyed in (11)ii), and ending with the goal 
proposition, where each pair of adjacent propositions in the chain is linked by a plau- 
sible coherence relation. The algorithm returns the proposition (e.g., the proposition 
conveyed in (11)iii) immediately preceding the goal proposition in each chain. Thus, 
when top-down recognition has reached an impasse, hypothesis generation (a type of 
bottom-up data-driven reasoning) provides a hypothesis that enables top-down recog- 
nition to continue another level of growth. An example of hypothesis generation is 
given in Section 4.1.5. 
The algorithm for hypothesis generation, which is given in Figure 7, performs a 
breadth-first search subject to a processing constraint on the maximum depth of the 
search tree. Note that a chain may have a length greater than three, e.g., the chain 
may consist of propositions (p0, pl, P2, P3), where P0 is the proposition to be related 
to the candidate plan, p3 is the goal, and P2 would be returned as a hypothesized 
proposition. In such a case, after p2 has been assimilated into the candidate plan, if pl 
is not present in R's turn, then hypothesis generation is invoked again and pl would 
403 
Computational Linguistics Volume 25, Number 3 
INPUTS 
OUTPUT 
P0: initial proposition 
pg: goal proposition 
GCR: goal coherence relation, i.e., coherence relation 
that must hold between hypothesized proposition and Pa 
S: set of coherence relations 
N: maximum search depth 
hypoth-list: list of alternative hypothesized propositions 
1. Initialize root of search tree with p0. 
2. Expand nodes of tree ill breadth-first order until either 
no more expansion is possible or maximum tree depth of N is reached, 
whichever happens first. To expand a node pi: 
a. Find all nodes pi+l such that for some relation CR in S, 
(Plausible (CR Pi pi+l)) is provable. 
b. Make each such Pi+l a child of Pi, linked by CR. 
c. A goal state is reached whenever Pi+l is pg and 
CR is identical to GCR. 
d. Whenever a goal state is reached, add the parent of pg 
to hypoth-list. 
Figure 7 
Hypothesis generation algorithm. 
be hypothesized also. 3° Finally, the search for a proposition Pi+l in step 2a is performed 
in our implementation using a theorem prover. 
4.1.4 Ranking Candidate Plans. Two heuristics are used to rank the relative plausi- 
bility of the set of candidate plans output by the first phase of answer recognition. 
First, plausibility decreases as the number of hypotheses in a candidate increases. 
(Assuming that all else is equal, it is safer to favor interpretations requiring fewer 
hypotheses.) Second, plausibility increases as the number of utterances in R's turn 
that are accounted for by the plan increases. (The more of R's turn accounted for, the 
more coherent the turn is likely to be, although not all of the utterances in R's turn are 
necessarily part of the full answer.) To give an example, consider the two candidate 
plans shown in Figure 8, corresponding to alternative interpretations of R's response 
in (12). 31 
(12) i. Q: Are you going to campus tonight? 
ii. R: \[No/Yes\] 
iii. \[My car's not running.\] 
iv. My car has a broken timing belt, 
v. \[so\] I'm going to take the bus. 
vi. Do you know how much the fare is? 
30 The implementation saves the chains to avoid the expense of recomputing intermediate hypotheses. 
31 This constructed example was designed to illustrate multiple aspects of the model. In our judgement, 
normally it would sound more coherent to give (12)v before (12)iv if (12)vi were not included. 
However, when (12)vi is included, (12)v not only elaborates upon the yes, but also serves as 
background for (12)vi. Another possible motivation for giving (12)v after (12)iv might be to delay 
giving dispreferred information (Levinson 1983), e.g., if the speaker believed that a yes was an 
unexpected or unwanted answer to (12)i. 
404 
Green and Carberry Indirect Answers 
Answer-no Answer-yes 
\[no\] Use-obstacle \[yes\] Use-elaboration 
\[iii\] Use-obstacle v Use-cause 
iv \[iiil Use-cause I 
iv 
Figure 8 
Ranking candidate plans. 
By these heuristics, a yes answer would be the preferred interpretation, since the candi- 
date Answer-yes plan uses the same number of hypotheses as the candidate Answer-no 
plan, and accounts for more of R's response. ((12)vi is not recognized as part of ei- 
ther answer.) The preference heuristics are intended to capture local coherence only. 
Since global information may play a role in selecting the correct interpretation, the 
higher-level discourse processor (described in section 4.2) must decide which plan to 
attribute to the speaker. 
4.1.5 Answer Recognition Example. In this section, we illustrate the interpretation of 
indirect answers in the model by describing how the two candidate plans shown in 
Figure 8 would be derived from R's response of (12)iv through (12)vi. First, each of the 
five top-level answer discourse plan operators would be instantiated with the ques- 
tioned proposition p, the proposition that R is going to campus tonight. Assuming that 
Q has no beliefs about R's beliefs and goals that are inconsistent with the applicability 
conditions and primary goals of these candidates, none of the candidates would be 
eliminated yet. Second, for each candidate the recognizer would check whether the 
communicative act specified in the nucleus was present in R's turn. In this example, 
since a direct answer was not explicitly provided by R, the recognizer would mark the 
nucleus of each candidate as hypothesized. The hypothesized nucleus of the candidate 
Answer-no and Answer-yes plans would be (inform s h (not p)) and (inform s h p), respec- 
tively. Next, the recognizer would try to recognize the acts expressed as (12)iv through 
(12)vi as satellites of each candidate plan. Assume that these acts are represented as 
(inform s h piv), (inform s h pv), and (inform s h pvi), respectively. 
To recognize an instance of a satellite, first, a satellite discourse plan operator 
would be instantiated. The header variable would be instantiated by unifying the 
satellite plan header with the corresponding act in the higher-level plan. For example, 
the header variable of a Use-obstacle satellite of an Answer-no candidate would be in- 
stantiated with (not p) in this example. The existential variable would be instantiated 
with the proposition conveyed in some utterance to be recognized as a satellite, e.g., 
Ply. However, before a candidate satellite may be attached to the higher-level candi- 
date plan, the answer recognizer must verify that the candidate satellite passes the 
following two tests: First, the candidate satellite's applicability conditions and pri- 
mary goals must be consistent with Q's beliefs about R's beliefs and goals. Second, 
the specified coherence relation must be plausible with respect to the beliefs that Q 
presumes to be shared with R, i.e., the satellite's instantiated applicability condition 
405 
Computational Linguistics Volume 25, Number 3 
of the form (Plausible (CR q p)) must be provable using the coherence rules described 
in Section 3. For example, given the beliefs that Q presumes to be shared with R and 
the coherence rules provided in the model, the act underlying (12)iv could not be the 
nucleus of a candidate Use-obstacle satellite of the Answer-no candidate, because the 
recognizer would not be able to prove that cr-obstacle is a plausible coherence relation 
holding between Piv and (not p). On the other hand, the act underlying (12)v would 
be interpreted as the nucleus of a candidate Use-elaboration satellite of the Answer-yes 
candidate, since the above tests are satisified, e.g., the recognizer could prove that 
cr-elaboration is a plausible coherence relation holding between pv and p. 
To return to consideration of the recognition of the Answer-no candidate, upon 
finding that the act underlying (12)iv cannot serve as a satellite, hypothesis genera- 
tion would be attempted. Recall that the goal of hypothesis generation is to supply 
a hypothesized missing act of the plan so that top-down recognition can continue. 
Hypothesis generation would search for a chain of plausibly related propositions, be- 
ginning with the proposition (i.e., Piv) to be related to the candidate Answer-no plan, 
and ending with the goal proposition (i.e., (not p)). As mentioned in Section 4.1.3, each 
pair of adjacent propositions in the chain must be linked by a plausible coherence re- 
lation. In this example, hypothesis generation would construct the chain (piv, piii, (not 
p)), where both pairs of adjacent propositions would be related by cr-obstacle and Piii 
is the hypothesis that R's car is not running. Hypothesis generation would return the 
proposition immediately preceding the goal proposition in this chain, i.e., pill. Thus, 
piii would be used to instantiate the existential variable of a Use-obstacle satellite of the 
Answer-no candidate plan, and satellite recognition would proceed. (The nucleus of this 
satellite would be marked as hypothesized.) Then, the recognizer would recognize piv 
(without requiring hypothesis generation) as a Use-obstacle satellite of this Use-obstacle 
satellite. No remaining utterances in R's turn can be related to the candidate Answer-no 
plan, resulting in the candidate shown on the left in Figure 8. 
Finally, to finish consideration of the recognition of the Answer-yes candidate, since 
neither the act underlying (12)iv nor the act underlying (12)vi can serve as a satellite of 
the Answer-yes candidate or its Use-elaboration satellite, hypothesis generation would 
again be invoked. Hypothesis generation would provide Piii, the hypothesis that R's car 
is not rtmning, as a plausible explanation for why R is going to take the bus. Thus, piii 
would be used to instantiate the existential variable of a Use-cause satellite of the Use- 
elaboration satellite of the Answer-yes candidate plan, and satellite recognition would 
proceed. (The nucleus of the Use-cause satellite would be marked as hypothesized.) 
Then, the recognizer would recognize piv (without requiring hypothesis generation) as 
a Use-cause satellite of this Use-cause satellite. No remaining utterances in R's turn can 
be related to the candidate Answer-yes plan, resulting in the candidate shown on the 
right in Figure 8. 
Given the shared beliefs and the coherence rules provided in the model, none of 
the utterances in R's turn would be recognized as satellites of the other three top- 
level candidate answer plans. Candidates that do not account for any actual parts 
of the response are eliminated at the end of phase one. Thus the output of phase 
one of interpretation would be just the two candidates shown in Figure 8. Phase 
two would evaluate the Answer-yes candidate as more preferred than the Answer-no 
candidate, since the former interpretation requires the same number of hypotheses 
and also accounts for more of R's response. 
4.2 Role of the Answer Recognizer in Discourse Processing 
As discourse researchers have pointed out (e.g., Reichman 1985; Levinson 1983)) the 
asking of a yes-no question creates the expectation that R will provide the answer 
406 
Green and Carberry Indirect Answers 
(directly or indirectly), if possible. Other acceptable, though less preferred, responses 
include I don't know and replies that provide other helpful information. Furthermore, 
an answer need not be given in the turn immediately following the turn in which the 
question was asked. For example, in (13) the yes-no question in (13)i is not answered 
until (13)v, separated by a request for clarification in (13)ii and its answer in (13)iii. 
(13) i. Q: Is Dr. Smith teaching CS360 next semester? 
ii. R: Do you mean Dr. Smithson? 
iii. Q: Yes. 
iv. R: \[No.\] 
v. He will be on sabbatical next semester. 
In Carberry's discourse-processing model for ellipsis interpretation (Carberry 1990), 
a mechanism is provided for updating the shared discourse expectations of dialogue 
participants throughout a conversation. Our answer recognizer would have the follow- 
ing role in such an architecture: The answer recognizer would be invoked whenever 
the current discourse expectation is that R will provide an answer. (If answer recog- 
nition were unsuccessful, then the discourse processor would invoke other types of 
recognizers for other types of responses.) The answer recognizer returns a partially 
ordered set (possibly empty) of answer discourse plans that it is plausible to ascribe 
to R as underlying (part or all of) the turn. The final choice of which discourse plan 
to ascribe to R should be made by the higher-level discourse processor, since it must 
select an interpretation consistent with the rest of the discourse. 
4.3 Comparison to Previous Approaches to Conversational Implicature 
Grice (1975) has proposed a theory of conversational implicature to account for certain 
types of conversational inferences. According to Grice, a speaker may convey more 
than the conventional meaning of an utterance by making use of the hearer's expec- 
tation that the speaker is adhering to general principles of cooperative conversation. 
Two necessary (but not sufficient) properties of conversational implicatures involve 
cancelability and speaker intention (Grice 1975; Hirschberg 1985). First, potential con- 
versational implicatures may be canceled explicitly, i.e., disavowed by the speaker in 
the preceding or subsequent discourse context, or even canceled implicitly given a 
particular set of shared beliefs. In fact, potential implicatures may undergo a change 
in status from cancelable to noncancelable in the subsequent discourse (Gunji 1981). 
Second, conversational implicatures are part of the intended meaning of an utter- 
ance. Grice proposes several maxims of cooperative conversation that a hearer uses as 
justification for inferring conversational implicatures. However, Grice's theory is inad- 
equate as the basis for a computational model of how conversational implicatures are 
derived. As frequently noted, Grice's maxims may support spurious or contradictory 
inferences. To date, few computational models have addressed the interpretation of 
conversational implicatures. 
Hirschberg's model (Hirschberg 1985) addresses a class of conversational impli- 
catures, scalar implicatures, which overlaps with the class of implicated answers ad- 
dressed in our model. (That is, scalar implicatures arise in question-answer exchanges 
as well as in other contexts, and, not all types of implicated answers are scalar im- 
plicatures.) According to Hirschberg, a scalar implicature depends upon the existence 
of a partially ordered set of values that is salient in the discourse context. Her model 
provides licensing rules that specify, given such a set, which scalar implicatures are 
407 
Computational Linguistics Volume 25, Number 3 
It is mutually plausible to the agent that (cr-contrast q p*) holds, 
where q is a proposition 
and p* is the proposition that p is partly true, 
if the agent believes it to be mutually believed that q is less than p in a salient partial order, 
unless it is mutually believed that p is true or that q is not true. 
It is mutually plausible to the agent that (cr-contrast q (not p)) holds, 
where q and p are propositions, 
if the agent believes it to be mutually believed that q is an alternate to p in a salient partial order, 
unless it is mutually believed that p is true or that q is not true. 
Figure 9 
Glosses of two coherence rules for cr-contrast. 
licensed in terms of values in the set that are lower than, alternate to, or higher than 
the value referred to in an utterance. For example, given a salient partially ordered 
set such that the value for the letter from X is lower than the value for all of the letters in 
question, in saying (2)ii (repeated below in (14)ii) R licenses the implicature that R has 
not gotten all of the letters in question. 
(14) i. Q: Have you gotten the letters yet? 
ii. R: I've gotten the letter from X. 
In our model, the response in (14)ii would be analyzed as generated from an Answer- 
hedge discourse plan whose nucleus has not been explicitly given and which has a 
single Use-contrast satellite whose nucleus is expressed in (14)ii. 3a The coherence rules 
for cr-contrast, which are based upon the notions elucidated by Hirschberg, are glossed 
in Figure 9. 33 However, the discourse plan operators in our model also characterize a 
variety of indirect answers that are not scalar implicatures, i.e., indirect answers based 
on the other coherence relations shown in Table 2. 
A model such as Hirschberg's, which does not take the full response into account, 
faces certain problems in handling cancellation by the subsequent discourse context 
("backwards" cancellation). For example, given a salient partially ordered set such that 
going to campus is ranked as an alternate to going shopping, Hirschberg's model would 
predict, correctly in the case of (15) and incorrectly in the case of (16), that R intended 
to convey a no. 
(15) i. Q: Are you going shopping? 
ii. R: \[no\] 
iii. I'm going to campus. 
iv. I have a night class. 
(16) i. Q: Are you going shopping? 
ii. R: \[yes\] 
32 The nucleus of such a plan conveys that the questioned proposition is partly but not completely true. 
33 The uppermost rule in the figure is the one applying to this example. The other rule applies to Use-contrast 
satellites of Answer-no plans. 
408 
Green and Carberry Indirect Answers 
iii. I'm going to campus. 
iv. The bookstore is having a sale. 
In our model, (16) would be interpreted by recognizing an Answer-yes plan (with a 
Use-elaboration and a Use-cause satellite underlying (16)iii and (16)iv, respectively) as 
more plausible than an Answer-no plan, rather than by use of backwards cancellation. 34 
In other words, in our model subsequent context can provide evidence for or against 
a particular interpretation, since a discourse plan may be expressed by multiple utter- 
ances. 
Also, a model such as Hirschberg's provides no explanation for why potential im- 
plicatures may become noncancelable. Our model predicts that a potential implicature 
of an utterance becomes noncancelable after the point in the conversation when the 
full discourse plan accounting for that utterance has been attributed to the speaker. 
For example, imagine a situation in which Q and R mutually intend to discuss two job 
candidates, A and B. Also, suppose that they mutually believe that they should not 
discuss any candidate until two letters of recommendation have been received for the 
candidate, and further, that both letters for B have been received. Our model predicts 
that the scalar implicature potentially licensed in (17)ii (i.e., that R has not gotten both 
letters for A yet) is no longer cancelable after R's turn in (17)iv, since by that point, the 
participants apparently would share the belief that Q had succeeded in recognizing 
R's discourse plan underlying (17)iiY 
(17) i. Q: Have you gotten the letters for A yet? 
ii. R: I've gotten the letter from X. 
iii. Q: Then let's discuss B now. 
iv. R: O.K. I think we should interview B, don't you? 
Inference of coherence relations has been used in modeling temporal (Lascarides 
and Asher 1991; Lascarides, Asher, and Oberlander 1992) and other defeasible dis- 
course inferences (Hobbs 1978; Dahlgren 1989). Inference of plausible coherence re- 
lations is necessary but not sufficient for interpreting indirect answers. For example, 
Q also must believe that there is a shared discourse expectation of an answer to a 
particular question. In other words, in our model, discourse plans provide additional 
constraints on the beliefs and intentions of the speaker that a hearer uses in interpret- 
ing a response. Another limitation of the above approaches is that they provide no 
explanation for the phenomenon of loss of cancelability described above. 
Plan recognition has been used to model the interpretation of indirect speech acts 
(Perrault and Allen 1980; Hinkelman 1989) and ellipsis (Carberry 1990; Litman 1986), 
discourse phenomena that share with conversational implicature the two necessary 
conditions described above, cancelablity and speaker intention. However, these models 
are inadequate for interpreting indirect answers, i.e., for deriving an implicated answer 
p from an indirect answer q. In these models, for p to be derivable from q, it is necessary 
34 of course, in the case where R provides only I'm going to campus, both yes and no interpretations would 
be inferred as equally plausible in our model. Although prosodic information is not used in our model, 
it is an interesting question for future research whether it can help in recognizing the speaker's 
intentions in such cases. 35 In other words, it would sound as if R had changed his mind or was contradicting himself if he said 
In 
fact I've gotten both letters for A after saying (17)iv. 
409 
Computational Linguistics Volume 25, Number 3 
for the hearer to infer that the speaker is performing or at least constructing a domain 
plan relating p and q. However, q need not play such a role in the speaker's inferred or 
actual domain plans, as shown in (18). 36 That is, it is not necessary to infer that R has 
a domain plan involving the renting of a car by X in order to recognize R's intention 
to convey no.) 
(18) i. Q: X will be renting a car, won't he? 
ii. R: \[No.\] 
iii. He can't drive. 
In other words, these models lack requisite knowledge encoded in our model in terms 
of possible satellites (based on coherence relations) of top-level discourse plan opera- 
tors. Also, the above plan-based models face the same problems as Hirschberg's since 
they do not address multiutterance responses. 
Philosophers (Thomason 1990; McCafferty 1987) have argued for a plan-based 
theory of implicature as an alternative to Grice's theory. Thomason proposes that im- 
plicatures are comprehended by a process of accommodation of the conversational 
record to fit the inferred plans of the speaker. According to McCafferty, "implicatures 
are things that the speaker plans that the hearer believe (and that the hearer can realize 
that the speaker plans that the hearer believe)" (p. 18). He claims that a theory based 
upon inferring the speaker's plan avoids the problem of predicting spurious impli- 
catures, since the spurious implicature would not be part of the speaker's plan. Our 
model is consistent with this view of conversational implicature. McCafferty sketches 
a possible plan-based model to account for the implicated yes answer in (19). 37 
(19) i. Q: Has Smith been dating anyone? 
ii. R: \[Yes.\] 
iii. He's been flying to New York every weekend. 
Although it was not McCafferty's intention to provide a computational model, but 
rather to show the plausibility of a plan-based theory of conversational implicature, 
some limitations of his suggestions for developing a computational model should be 
noted. First, his proposed rules cannot be used to derive an alternate, plausible inter- 
pretation of (19)iii, in which R scalar implicates a no. 38 Our model can account for both 
interpretations. The first interpretation would be accounted for by an inferred Answer- 
yes plan with a Use-elaboration satellite underlying (19)iii, while the latter would be 
accounted for by an inferred Answer-no plan with a Use-contrast satellite underlying 
(19)iii. More generally, his proposed rules cannot account for types of indirect an- 
swers described in our model by coherence relations whose definitions do not involve 
planning knowledge. Second, even if rules could be added to McCafferty's model to 
account for a speaker's plan to convey a no by use of (19)iii, his model does not pro- 
vide a way of using information from other parts of the response, e.g., (20)iv, to help 
recognize the intended answer. As noted earlier, in our model such information can 
36 (18) is based upon (1), modified for expository purposes. 37 (19) is from McCafferty (1987), page 67, and is similar to an example of Grice's. In a Gricean account, 
this implicature would be justified in terms of the Maxim of Relevance. 38 That is, an interpretation in 
which flying to New York is mutually believed to be an alternate to dating 
someone in a salient partially ordered set. 
410 
Green and Carberry Indirect Answers 
be used to provide evidence favoring one candidate discourse plan over another. (For 
example, (20)iv would be accounted for by the addition of a Use-obstacle satellite to 
the Answer-no candidate described above.) 
(20) i. Q: Has Smith been dating anyone? 
ii. R: \[No.\] 
iii. He's been flying to New York every weekend. 
iv. Besides, he's married. 
4.4 Summary 
This section closes with a summary of the argument for the adequacy of our model 
as a model of conversational implicature. As discussed earlier, two necessary condi- 
tions for conversational implicature are cancelability and speaker intention. We have 
demonstrated that our model can handle forward and backward cancellation, and pro- 
vides an explanation for the "loss of cancelability" phenomenon. Regarding speaker 
intention, in our model a conversationally implicated answer is an answer that R 
planned that Q recognize (and that Q recognizes that R planned that Q recognize). 39 
We have demonstrated how Q's recognition of R's discourse plan (in particular, the 
goal to provide an answer to the question) can be performed using the knowledge and 
algorithms in our model. Furthermore, we argue that Q's recognition of R's intention 
that Q recognize R's plan follows from the role of interpretation in generation, namely, 
Q and R mutually believe that R will not say what he does unless R believes that Q 
will be able to interpret the response as intended. In our model, during generation 
(to be described in Section 5), R constructs a model of Q's beliefs (using R's shared 
beliefs), and then simulates Q's interpretation of a trial pruned response. R's decision 
to use the pruned response depends upon whether R believes that Q would still be 
able to recognize the answer after the plan has been pruned. During interpretation, 
given the shared discourse expectation that R will provide an answer to Q's yes-no 
question, Q's use of (Q's) shared beliefs to interpret the response, and Q's belief that R 
expects that Q will be able to recognize the answer, Q's recognition of a discourse plan 
for an answer warrants Q's belief that R intended for Q to recognize this intention. 
5. Generation 
This section describes our approach to the generation of indirect answers. Genera- 
tion is modeled as a two-phase process of discourse plan construction. First, in the 
content planning phase, a discourse plan for a full answer is constructed. Second, 
the plan pruning phase uses the model's own interpretation capability to determine 
what information in the full response does not need to be stated explicitly. In ap- 
propriate discourse contexts, i.e., in contexts where the direct answer can be inferred 
by Q from other parts of the full answer, a plan for an indirect answer is thereby 
generated. When the direct answer must be given explicitly, the result is a plan for a 
direct answer accompanied by appropriate extra information. (According to the study 
mentioned in Section 2 \[Stenstrbm 1984\], 85% of direct answers are accompanied by 
such information. Thus, it is important to model this type of response as well.) 
While the pragmatic knowledge described in Section 3 is sufficient for interpreta- 
tion, it is not sufficient for the problem of content planning during generation. Applica- 
39 Applying McCafferty's description of conversational implicature to indirect answers. 
411 
Computational Linguistics Volume 25, Number 3 
(Use-elaboration s h ?p): 
Existential variable: ?q 
Applicability conditions: 
(bel s (cr-elaboration ?q ?p)) 
(Plausible (cr-elaboration ?q ?p)) 
Stimulus conditions: 
(answer-ref-indicated s h ?p ?q) 
(clarify-concept-indicated s h ?p ?q) 
Nucleus: 
(inform s h ?q) 
Satellites: 
(Use-cause s h ?q) 
(Use-elaboration s h ?q) 
Primary goals: 
(BMB h s (or-elaboration ?q ?p)) 
Figure 10 
(Use-obstacle s h ?p): 
Existential variable: ?q 
Applicability conditions:- 
(bel s (cr-obstacle ?q ?p)) 
(Plausible (cr-obstacle ?q ?p)) 
Stimulus conditions: 
(explanation-indicated s h ?p ?q) 
(excuse-indicated s h ?p ?q) 
Nucleus: 
(inform s h ?q) 
Satellites: 
(Use-obstacle s h ?q) 
(Use-elaboration s h ?q) 
Primary goals: 
(BMB h s (or-obstacle ?q ?p)) 
Two satellite discourse plan operators with stimulus conditions. 
bility conditions prevent inappropriate use of a discourse plan. However, they do not 
model a speaker's motivation for choosing to provide extra information. Consider (21). 
(21) i. Q: I need a ride to the mall. 
ii. Are you going shopping tonight? 
iii. R: \[No.\] 
iv. My car's not running. 
v. The timing belt is broken. 
R's reason for providing the information in (21)iv might have been to give an excuse for 
not being able to offer Q a ride, and R's reason for providing the information in (21)v 
might have been to provide an explanation for news in (21)iv that may surprise Q. 
Furthermore, a full answer might be too verbose if every satellite whose applicability 
conditions held were included in the full answer. On the other hand, at the time when 
he is asked a question, R may not hold the primary goals of a potential satellite. (In our 
model the only goal R is assumed to have initially is the goal to provide the answer.) 
Thus, an approach to selecting satellites driven only by these satellite goals would fail. 
To overcome these problems, we have augmented the satellite discourse plan oper- 
ators, as described in Section 3, with one or more stimulus conditions. Two examples 
are shown in Figure 10. Stimulus conditions describe general types of situations in 
which a speaker is motivated to include a satellite during plan construction. They can 
be thought of as situational triggers, which give rise to new speaker goals (i.e., the 
primary goals of the satellite operator), and which are the compiled result of deeper 
planning based upon principles of cooperativity (Grice 1975) or politeness (Brown 
and Levinson 1978). 4o In order for a satellite to be included, all of its applicability 
conditions and at least one of its stimulus conditions must be true. 
40 It was beyond the scope of our research to model recognition of stimulus conditions. We argue in 
Section 5.3, however, that this does not compromise our approach as a model of conversational 
implicature. 
412 
Green and Carberry Indirect Answers 
Our methodology for identifying stimulus conditions was to survey linguistic 
studies, described in Section 5.1, as well as to analyze the possible motivation of 
the speaker in the examples in our corpus. The rules used in our model to evaluate 
stimulus conditions are given in Section 5.2. Section 5.3 presents our implemented 
generation algorithm, and Section 5.4 illustrates the algorithm with an example. 
5.1 Linguistic Studies 
In linguistic studies, the reasons given for including extra information 41 in a response 
to a yes-no question can be categorized as: 
• to provide implicitly requested information, 
• to provide an explanation for an unexpected answer, 
• to qualify a direct answer, or 
• politeness-related. 
5.1.1 Implicitly Requested Information. As mentioned in Section 2, Stenstr6m claims 
that the typical reason for providing extra information is to answer an implicit wh- 
question. Kiefer (1980) observes that several types of yes-no questions, when used 
to perform indirect speech acts, have the property that one or both of the "binary" 
answers (i.e., yes or no) used alone is an inappropriate response to them. For example, 
in response to (22)i, 42 when interpreted as (22)ii, an answer of (22)iii or (22)v 43 would 
be appropriate, but not (22)iv alone. 
(22) i. Q: Is John leaving for Stockholm TOMORROW? 
ii. Q: When is John leaving for Stockholm? 
iii. R: Yes. 
iv. R: No. 
v. R: No, John is going to leave the day after tomorrow. 
Kiefer also provides examples of cases where the other binary answer alone is inap- 
propriate, or where either alone is inappropriate. 
Clark (1979) studied how different factors may influence the responder's con- 
fidence that the literal meaning of a question was intended and confidence that a 
particular indirect meaning was intended. In one experiment, in which subjects re- 
sponded to the question, Do you accept credit cards?, about half of the subjects provided 
information answering an indirect request of What credit cards do you accept? Clark spec- 
ulates that the half who included information addressing the indirect request in their 
response had some, but not necessarily total, confidence that it was intended. 
According to Levinson (1983), a yes-no question often may be interpreted as a 
prerequest for another request, i.e., it may be used in the first position of the sequence 
41 We are reporting only cases where the extra information may be used as an indirect answer. 42 In this example, Kiefer's (11b), we follow Kiefer's use of capitalization to indicate that tomorrow would 
be stressed in spoken English. 43 Kiefer's (21b). 
413 
Computational Linguistics Volume 25, Number 3 
T1-T4, where the occurrence of T3 and T4 are conditional upon R's answer in T2: 
• TI: Q makes a prerequest to determine if a precondition of an action to 
be requested by Q in T3 holds. 
• T2: R gives an answer indicating whether the precondition holds 
• T3: Q makes the request 
• T4: R responds to the request in T3. 
Levinson claims that prerequests are used to check whether the planned request (in T3) 
is likely to succeed so that a dispreferred response to it can be avoided by Q. Another 
reason is that, since receiving an offer is preferred to making a request (Schegloff 1979), 
by making a prerequest, Q gives R the opportunity to offer whatever Q would request 
in T3, i.e., the sequence would then consist of just T1 and T4. In analyses based on 
speech act theory, in a sequence consisting of just T1 and T4, the prerequest would be 
referred to as an indirect speech act. 
5.1.2 Explanation for Unexpected Answer. Stenstr6m notes that a reason for pro- 
viding extra information is to provide an explanation justifying a negative answer. 44 
According to Levinson (1983), the presence of an explanation is a distinguishing fea- 
ture of dispreferred responses to questions and other second parts of adjacency pairs 
(Schegloff 1972). In an adjacency pair, each member of the pair is produced by a differ- 
ent speaker, and the occurrence of the first part creates the expectation that the second 
part will appear, although not necessarily immediately following the first member. 
Levinson claims that dispreferred responses to first parts of adjacency pairs can be 
identified by structural features such as: 
• use of pauses or displacement, 
• prefacing with markers (e.g. uh or well), appreciations, apologies, or 
refusals, 
• providing explanations, and 
• declinations given in an indirect or mitigated manner. 
For example in (23), 45 the marker well is used and an explanation is given. 
(23) i. Q: What about coming here on the way or doesn't that give you 
enough time? 
ii. R: Well no I'm supervising here 
Although Levinson defines preference in terms of structural features, he notes that 
there is a correlation between preference and content. For example, unexpected an- 
swers to questions, refusals of requests and offers, and admissions of blame are typi- 
cally marked with features from the above list. 
44 She found that 61% of negative direct answers but only 24% of positive direct answers were 
accompanied by qualify acts. 
45 Levinson's example (55). 
414 
Green and Carberry Indirect Answers 
5.1.3 Avoid Misunderstanding. Stenstr6m notes that extra information may be given 
to qualify an answer. Hirschberg (1985) claims that speakers may give indirect answers 
to block potential unintended scalar implicatures of a yes or no alone. For example in 
(2), repeated below as (24), R's response is preferable to just no, since that would license 
the incorrect scalar implicature that R had not received any of the letters in question. 
However, by use of (24)ii in an appropriate discourse context, R is able to convey 
explicitly which letter has been received as well as to conversationally implicate that 
R has not gotten the other letters in question. 
(24) i. Q: Have you gotten the letters yet? 
ii. R: I've gotten the letter from X. 
5.1.4 Politeness. StrenstrOm claims that extra information may be given for social rea- 
sons. Kiefer notes that extra information may be given as an excuse when the answer 
indicates that the speaker has failed to fulfill a social obligation. Brown and Levinson 
(1978) claim that politeness strategies, which may at times conflict with Gricean max- 
ims, account for many uses of language. According to Brown and Levinson, certain 
communicative acts are intrinsically face-threatening acts (FTAs). That is, doing an 
FTA is likely to injure some conversational participant's face, or public self-image. 
For example, orders and requests threaten the recipient's negative face, "the want ... 
that his actions be unimpeded by others" (p. 67). On the other hand, disagreement or 
bearing "bad news" threatens the speaker's positive face, the want to be looked upon 
favorably by others. Further, they claim that politeness strategies can be ranked, and 
that the greater the threat associated with a face-threatening act, the more motivated 
a speaker is to use a higher-numbered strategy. 
Brown and Levinson propose the following ranked set of strategies (listed in order 
from lower to higher rank): 
1. 
. 
perform the FTA. (Brown and Levinson claim that this amounts to 
following Gricean maxims.) 
perform the FTA with redressive action, i.e., in a manner that indicates 
that no face threat is intended, using positive politeness strategies 
(strategies that increase the hearer's positive face). Such strategies 
include: 
. 
. 
Strategy 1: attending to the hearer's interests or needs 
Strategy 6: avoiding disagreement, e.g., by displacing an answer 
perform the FTA with redressive action, using negative politeness 
strategies (strategies for increasing negative face). These include: 
Strategy 6: giving an excuse or an apology 
perform the FTA off-record, i.e., by use of conversational implicature. 
In the next section, we provide several stimulus conditions that reflect positive po- 
liteness strategy 1 and negative politeness strategy 6. However, although politeness 
considerations may motivate a speaker to convey an answer indirectly, it is beyond 
the scope of our generation model to choose between a direct and an indirect answer 
on this basis. 
415 
Computational Linguistics Volume 25, Number 3 
Table 3 
Stimulus conditions of discourse plan operators. 
Operator 
Use-cause 
Use-condition 
Use-contrast 
Use-elaboration 
Use-obstacle 
Use-otherwise 
Use-possible-cause 
Use-possible-obstacle 
Use-result 
Use-usually 
Stimulus Conditions 
explanation-indicated 
excuse-indicated 
clarify-condition-indicated 
appeasement-indicated 
answer-ref-indicated 
clarify-extent-indicated 
substitute-indicated 
answer-ref-indicated 
clarify-concept-indicated 
explanation-indicated 
excuse-indicated 
explanation-indicated 
excuse-indicated 
explanation-indicated 
excuse-indicated 
explanation-indicated 
excuse-indicated 
explanation-indicated 
explanation-indicated 
5.2 Stimulus Conditions 
In this section we provide glosses of rules giving sufficient conditions for the stim- 
ulus conditions used in our model. (The rules are encoded as Horn clauses in our 
implementation of the model.) Table 3 summarizes which stimulus conditions appear 
in which discourse plan operators. As mentioned above, for an instance of a satellite 
operator to be selected during generation, all of its applicability conditions and at least 
one of its stimulus conditions must hold. 
5.2.1 Explanation-indicated. This stimulus condition appears in all of the operators 
for providing causal explanations. For example in (1), repeated below as (25), R gives 
an explanation of why R won't get a car. 
(25) i. Q: Actually you'll probably get a car won't you as soon as you get 
there? 
ii. R: \[No.\] 
iii. I can't drive. 
R's response may contribute to greater dialogue efficiency by anticipating a follow-up 
request for an explanation. The rule for this stimulus condition may be glossed as: (a 
speaker) s is motivated to give (a hearer) h an explanation for (not p), if s suspects that 
h suspects that (a proposition) p is true, unless it is the case that s has reason to believe 
that h will accept (not p) on s's authority, s may acquire the suspicion that h suspects 
that p is true by means of syntactic clues from the yes-no question, e.g., from the form 
of the question in (25)i. 
5.2.2 Excuse-indicated. Although this stimulus condition appears in some of the same 
causal operators as explanation-indicated, it represents a different kind of motivation. 
416 
Green and Carberry Indirect Answers 
A yes-no question may be interpreted as a prerequest. Thus, a negative answer to a 
yes-no question used as a prerequest may be interpreted as a refusal. To soften the 
refusal, i.e., in accordance with negative politeness strategy 6, the speaker may give 
an explanation of the negative answer, as illustrated in (21), repeated below in (26). 
(26) i. Q: I need a ride to the mall. 
ii. Are you going shopping tonight? 
iii. R: \[No.\] 
iv. My car's not running. 
v. The timing belt is broken. 
The rule for this stimulus condition may be glossed as: s is motivated to give h an 
excuse for (not p), if s suspects that h's request, (informifs h p), is a prerequest. Techniques 
for interpreting indirect speech acts (Perrault and Allen 1980; Hinkelman 1989) can be 
used to determine whether the rule's antecedent holds. 
5.2.3 Answer-ref-indicated. This condition appears in Use-elaboration, illustrated by 
(27), 46 and in Use-contrast, illustrated by (28). 47 
(27) i. Q: Did you have a hotel in mind? 
ii. \[What hotel did you have in mind?\] 
iii. R: \[Yes.\] 
iv. There's a Holiday Inn right near where I'm working. 
(28) i. Q: You're on that? 
ii. \[Who's on that?\] 
iii. R: No no no. 
iv. Dave is. 
In (27), R has interpreted the question in (27)i as a prerequest for the wh-question 
shown in (27)ii. Thus, (27)iv not only answers the question in (27)i but also the an- 
ticipated wh-question in (27)ii. Similarly in (28), R may interpret the question in (28)i 
as a prerequest for the wh-question in (28)ii, and so gives (28)iv to provide an answer 
to both (28)i and (28)ii. The rule for this stimulus condition may be glossed as: s is 
motivated to provide h with q, if s suspects that h wants to know the referent of a 
term t in q. As in excuse-indicated, techniques for interpreting indirect speech acts can 
be used to determine if the rule's antecedent holds. 48 
46 From SRI Tapes (1992), tape 1. 
47 Stenstr6m's (102). A no answer may be conversationally implicated by use of (28)iv alone. 48 However, following Clark (1979), the rule does not require that R be certain that Q was making an 
indirect request. 
417 
Computational Linguistics Volume 25, Number 3 
5.2.4 Substitute-indicated. This condition appears in Use-contrast, illustrated by (29). 
(29) i. Q: Do you have Verdi's Otello or Aida? 
ii. R: \[No.\] 
iii. We have Rigoletto. 
Although Q may not have intended to use (29)i as a prerequest for the question What 
Verdi operas do you have?, R suspects that the answer to this wh-question might be 
helpful to Q, and so provides it (in accordance with positive politeness strategy 1). 
The rule for this stimulus condition may be glossed as: s is motivated to provide h 
with q, if s suspects that it would be helpful for h to know the referent of a term t in 
q. The rule's antecedent would hold whenever obstacle detection techniques (Allen 
and Perrault 1980) determine that h's not knowing the referent of t is an obstacle to 
an inferred plan of h's. However, not all helpful responses, in the sense described in 
Allen and Perrault (1980), can be used as indirect answers. For example, even if the 
clerk (R) at the music store believes that Q's not knowing the closing time could be 
an obstacle to Q's buying a recording, a response of (30) alone would not convey no 
since it cannot be coherently related to an Answer-no plan. 
(30) i. R: We close at 5:30 tonight. 
5.2.5 Clarify-concept-indicated. This stimulus condition appears in Use-elaboration, as 
illustrated in (31). 49 
(31) i. Q: Do you have a pet? 
ii. R: We have a turtle. 
In (31), R was motivated to elaborate on the type of pet R has since turtles are not 
prototypical pets. The rule for this stimulus condition may be glossed as: s is motivated 
to clarify p to h with q, if p contains a concept c, and q provides an atypical instance 
of c. Stereotypical knowledge would be used to evaluate the rule's antecedent. 
5.2.6 Clarify-condition-indicated. This stimulus condition appears in the operator 
Use-condition, as illustrated by (32). 5o 
(32) i. Q: Um let me can I make the reservation and change it by tomorrow? 
ii. R: \[Yes.\] 
iii. If it's still available. 
In (32), a truthful yes answer depends on the truth of (32)iii. The rules for this stimulus 
condition may be glossed as: s is motivated to clarify a condition q for p to h if 1) s 
doesn't know if q holds, or 2) s suspects that q does not hold. 
49 Example (177) from Hirschberg (1985). 50 From SRI Tapes (1992), tape 10ab. 
418 
Green and Carberry Indirect Answers 
5.2.7 Clarify-extent-indicated. This stimulus condition appears in Use-contrast, as il- 
lustrated by (2), repeated below as (33). 
(33) i. Q: Have you gotten the letters yet? 
ii. R: I've gotten the letter from X. 
On the strict interpretation of (33)i, Q is asking whether R has gotten all of the letters, 
but on a looser interpretation, Q is asking if R has gotten any of the letters. Then, if 
R has gotten some but not all of the letters, just yes would be untruthful. However, if 
Q is speaking loosely, then just no might lead Q to erroneously conclude that R has 
not gotten any of the letters. R's answer circumvents this problem, by conveying the 
extent to which the questioned proposition (on the strict interpretation) is true. 
The rules for this stimulus condition may be glossed as: s is motivated to clarify to 
h the extent q to which p is true, or the alternative q to p which is true, if s suspects that 
h does not know if q holds, and s believes that q is the highest expression alternative to 
p that does hold. According to Hirschberg (1985) (following Gazdar), sentences pi and 
pj (representing the propositional content of two utterances) are expression alternatives 
if they are the same except for having comparable components ei and ej, respectively. 
As mentioned earlier, Hirschberg claims that in a discourse context there may be a 
partial ordering of values that the discourse participants mutually believe to be salient. 
She claims that the ranking of ei and ej in this ordering can be used to describe the 
ranking of pi and pj. In the above example, (33)ii is a realization of the highest true 
expression alternative to the questioned proposition, p, i.e., the proposition that R has 
gotten all the letters. 51 
5.2.8 Appeasement-indicated. This stimulus condition appears in Use-contrast, as il- 
lustrated by (34). 52 
(34) i. Q: Did you manage to read that section I gave you? 
ii. R: I've read the first couple of pages. 
In (34), R conveys that there is some (though not much) truth to the questioned propo- 
sition in an effort to soften his answer (in accordance with positive politeness strategy 
1). More than one stimulus condition may motivate R to include the same satellite. 
For example, in (34), R may have been motivated also by clarify-extent-indicated, which 
was described above. However, it is possible to provide a context for (35) where 
appeasement-indicated holds but clarify-extent-indicated does not, or a context for (34) 
where the converse is true. 
(35) i. Q: Did you wash the dishes? 
ii. R: I brought you some flowers. 
The rules for this stimulus condition may be glossed as: s is motivated to appease h 
with q for p not holding or only being partly true, if s suspects that (not p) is undesirable 
51 Recall that additional constraints on p and q arise from the applicability conditions of operators 
containing this stimulus condition, namely Use-contrast in this case. Thus, another constraint is that it is 
plausible that cr-contrast holds. The coherence rule for cr-contrast was described in Section 4.3. 52 Example (56) from Hirchberg (1985). 
419 
Computational Linguistics Volume 25, Number 3 
Table 4 
General principles underlying stimulus conditions. 
1. Efficiency: explanation-indicated, answer-ref-indicated 
2. Accuracy: clarify-concept-indicated, clarify-extent-indicated,clarify-condition-indicated 
3. Politeness: excuse-indicated, appeasement-indicated, substitute-indicated 
to h but that q is desirable to h. The antecedents to this rule would be evaluated using 
heuristic rules and stereotypical and specific knowledge about h's desires. For example, 
two heuristics of rational agency that might lead to beliefs about h's desires are 1) if 
an agent wants you to perform an action A, then your failure to perform A may be 
undesirable to the agent, and 2) if an agent wants you to do A, then it is desirable to 
the agent that you perform a part of A. 
5.2.9 Summary. In summary, the stimulus conditions in our model can be classified 
according to three general principles, as shown in Table 4. The first category, effi- 
ciency, includes the motivation to provide implicitly requested information as well 
as to provide an explanation for unexpected information. In other words, giving this 
type of extra information contributes to the efficiency of the conversation by elimi- 
nating the need for follow-up wh-questions or follow-up why? or why not? questions, 
respectively. In the category of accuracy, in addition to the reason cited by Hirschberg 
(which is represented in our model as clarify-extent-indicated), we have identified two 
other reasons for giving extra information, which contribute to accuracy. The category 
of politeness includes reasons for redressing face-threatening acts using positive and 
negative politeness strategies. 
5.3 Generation Algorithm 
The inputs to generation consist of: 
• the set of discourse plan operators (described in Section 3) augmented 
with stimulus conditions, 
• the set of coherence rules (also described in Section 3), 
• the set of stimulus condition rules (described in Section 5.2), 
• R's beliefs (including the discourse expectation that R will provide an 
answer to some questioned proposition p), and 
• the semantic representation of p. 
The model presupposes that when answer generation begins, the speaker's (R's) only 
goal is to satisfy the above discourse expectation. R's nonshared beliefs (including 
beliefs whose strength is not necessarily certainty) about Q's beliefs, intentions, and 
preferences are used in generation to evaluate whether a stimulus condition holds. 
The output of the generation algorithm is a discourse plan that can be realized by a 
tactical generation component (McKeown 1985). 53 
53 The plan that is output specifies an ordering of discourse acts based upon the ordering of coherence 
relations specified in the discourse plan operators. However, reordering may be required, e.g., to model 
a speaker who has multiple goals. 
420 
Green and Carberry Indirect Answers 
The answer generation algorithm has two phases. In the first phase, content plan- 
ning, the generator creates a discourse plan for a full answer, i.e., a direct answer 
and extra appropriate information. In the second phase, plan pruning, the generator 
determines which propositions of the planned full answer do not need to be explic- 
itly stated. For example, given an appropriate model of R's beliefs, the system would 
generate a plan for asserting only the proposition conveyed in (36)v and (36)vi as an 
answer to (36)i. 54 
(36) i. Q: Is Mark here \[at the office\]? 
ii. R: \[No.\] 
iii. \[He's at home.\] 
iv. \[He is caring for his daughter.\] 
v. His daughter has the measles, 
vi. but he's logged on. 
An advantage of this approach is that, even when it is not possible to omit the direct 
answer, a full answer is generated. 
5.3.1 Content Planning. Content planning is performed by top-down expansion of an 
answer discourse plan operator. First, each top-level answer discourse plan operator 
is instantiated with the questioned proposition until one is found such that its applica- 
bility conditions hold. s5 Next, the satellites of this operator are expanded (recursively). 
The algorithm for expanding a satellite adds each instance of a satellite such that all 
of its applicability conditions and at least one of its stimulus conditions hold. Thus, 
different instantiations of the same type of satellite may be included in a plan for 
different reasons. For example, (36)iii and (36)vi both realize Use-contrast satellites, the 
former included due to the answer-ref-indicated stimulus condition, and the latter due 
to the substitute-indicated stimulus condition. 
For each stimulus condition of a satellite, our implementation of the algorithm 
uses a theorem prover to search the set of R's beliefs (encoded as Horn clauses) 
for a proposition satisfying a formula consisting of a conjunction of the applicability 
conditions and that stimulus condition. A proposition satisfying each such formula 
is used to instantiate the existential variable of the satellite operator. For example, 
to generate the response in (36), the following formula would be constructed from 
the Use-contrast operator's applicability conditions and one of its stimulus conditions, 
(answer-ref-indicated), where p is the proposition that Mark is at the office, and ?q is the 
existential variable to be instantiated. 
((and (bel s (cr-contrast ?q (not p))) 
(Plausible (cr-contrast ?q (not p))) 
(answer-ref-indicated s h ?q))) 
The result of the search is to instantiate ?q with the proposition that Mark is at home, 
due to the speaker's belief that the hearer might have been using (36)i as a prerequest 
54 Parts (36)i-(36)v were overheard by one of the authors in a naturally occurring dialogue, and (36)vi 
was added for expository purposes. 
55 It is assumed that exactly one is appropriate. 
421 
Computational Linguistics Volume 25, Number 3 
for the question, Where is Mark? For a more complete description of how R's response 
in (36) is generated, see Section 5.4. 
We have employed a simple approach to planning because the focus of our re- 
search was on the use of the response as an indirect answer, i.e., on aspects of response 
generation that play a role in interpreting the implicature. In a more sophisticated dis- 
course planning formalism, such as argued for in Young, Moore, and Pollack (1994), it 
would be possible to represent and reason about other intended effects of the response. 
(In our model, the effects or primary goals are used in interpretation but their only 
role in generation is in simulated interpretation. However, their role in interpretation is 
important; they constrain what discourse plans can be ascribed to the speaker.) While 
we believe that use of more sophisticated planning formalisms is well motivated for 
discourse generation in general, we leave the problem of generating indirect answers 
in such formalisms for future research. The use of stimulus conditions to motivate the 
selection of optional satellite operators is sufficient for our current goals. 
5.3.2 Plan Pruning. The output of the content planning phase, an expanded discourse 
plan representing a full answer, is the input to the plan pruning phase of generation. 
The expanded plan is represented as a tree of discourse acts. The goal of this phase is 
to make the response more concise, i.e., to determine which of the planned acts can be 
omitted while still allowing Q to infer the full discourse plan. 56 To do this, the generator 
considers each of the acts in the frontier of the tree from right to left. (This ensures that 
a satellite is considered before its related nucleus.) The generator creates a trial plan 
consisting of the original plan minus the nodes pruned so far and minus the current 
node. Then, using the answer recognizer, the generator simulates Q's interpretation of 
a response containing the information that would be given explicitly according to the 
trial plan. In the simulation, R's shared beliefs are used to model Q's shared beliefs. 
If Q could infer the full plan (as the most preferred interpretation), then the current 
node can be pruned. Otherwise, it is left in the plan and the next node is considered. 
For example, consider Figure 11 as we illustrate the possible effect of pruning 
on a full discourse plan. The leaf nodes, representing discourse acts, are numbered 1 
through 8. Arcs labeled N and S lead to a nucleus or satellite, respectively. Node 8 
corresponds to the direct answer. Plan pruning would process the nodes in order from 
1 to 8. The maximal set of nodes that could be pruned in Figure 11 is the set containing 
2, 3, 4, 7, and 8. That is, nodes 2 through 4 might be inferable from 1, node 7 from 5 
or 6, and node 8 from 4 or 7, but nodes 1, 5, and 6 cannot be pruned since they are 
not inferable from other nodesY In the event that it is determined that no node can 
be pruned, the full plan would be output. The interpretation algorithm (described in 
Section 4) would use hypothesis generation to recognize missing propositions other 
than the direct answer, i.e., the propositions at nodes 2, 3, 4, and 7. 
To comment on the role of interpretation in generation, it is a key component of our 
claim to have provided an adequate model of conversational implicature. Given the 
shared discourse expectation that R will provide an answer to Q's yes-no question, Q's 
use of (Q's) shared beliefs to interpret the response, and Q's belief that R expects that Q 
will be able to recognize the answer, Q's recognition of a discourse plan for an answer 
warrants Q's belief that R intended for Q to recognize this intention. In particular, R 
would not have pruned the direct answer unless, given the beliefs that R presumes 
56 Conciseness is not the only possible motive for omitting the direct answer. As mentioned earlier, an 
indirect answer may be used to avoid performing a face-threatening act. However, it is beyond the 
scope of our model to determine whether to omit the direct answer on grounds of politeness. 
57 In fact, leaves that have no satellites of their own cannot be pruned. 
422 
Green and Carberry Indirect Answers 
7 6 5 4 
N 
Figure 11 
Example of full discourse plan before pruning. 
Q: 
R: 
a. 
b. 
C. 
d. 
Figure 12 
Is Mark here? 
\[No.\] 
\[He's at home.\] 
\[He is caring for his daughter.\] 
His daughter has the measles. 
but he is logged on. 
Generation example: exchange. 
i 
2 1 
to be shared with Q, R believes that Q will be able to recognize a chain of mutually 
plausible coherence relations from the actual response to the intended answer, and 
thus be able to recognize R's plan. Note that although stimulus conditions are not 
recognized during interpretation in our approach, the model does account for the 
recognition of those parts of the plan concerning the answer. For example, although 
Q may not know whether R was motivated by excuse-indicated or explanation-indicated 
to provide (21)iv in response to (21)ii, Q can recognize R's intention to convey a no 
by Q's recognition of (21)iv as the nucleus of a Use-Obstacle satellite of Answer-No. 
Thus, Q can thereby attribute to R the primary goal of the Answer-No plan to convey 
a no. 
5.4 Generation Example 
This example models R's generation of the response in the exchange shown in Fig- 
ure 12, which repeats (36). The discourse plan constructed by the algorithm is depicted 
in Figure 13, where (a) through (d) refer to communicative acts that could be performed 
by saying the sentences with corresponding labels in Figure 12. Square brackets in the 
plan indicate acts that have been pruned, i.e., that are not explicitly included in the 
response. 
First, each top-level answer operator is instantiated with the questioned propo- 
sition, p, the proposition that Mark is at the office. An (Answer-no s h p) plan would 
be selected for expansion since its applicability conditions can be proven. To expand 
this plan, the algorithm attempts to expand each of its satellites as described in Sec- 
tion 5.3.1. The generation algorithm searches for (at most) one instance of a satellite for 
each possible motivation of a satellite. In this example, two satellites of (Use-contrast s 
h (not p)) are added to the plan. In one, motivated by the Answer-ref stimulus condi- 
423 
Computational Linguistics Volume 25, Number 3 
\[no\] 
(Answer-no s h p) 
(Use-contrast s h (not p)) 
j~xexplanation-indicated 
\[a\] (Use-cause s h pa ) ~ 
ation-indicated 
\[b\] (Use-cause s h Pb ) 
(Use-contrast s h (not p)) 
Figure 13 
Final plan. 
tion, the existential proposition is instantiated with pa, the proposition that Mark is at 
home. In other words, proposition, pa was found to satisfy ?q in formula 1 below. 
1. ((and (bel s (cr-contrast ?q (not p))) 
(Plausible (cr-contrast ?q (not p))) 
(answer-ref-indicated s h ?q))) 
In the other (Use-contrast s h (not p)) satellite, motivated by the substitute-indicated stim- 
ulus condition, the existential proposition is instantiated with pd, the proposition that 
Mark is logged on. That is, Pa was found to satisfy ?q in formula 2 below. 
2. ((and (bel s (cr-contrast ?q (not p))) 
(Plausible (cr-contrast ?q (not p))) 
(substitute-indicated s h ?q))) 
The former (Use-contrast s h (not p)) satellite (i.e., the one constructed using pa) can 
be expanded by adding a (Use-cause s h pa) satellite to it. This satellite's existential 
variable is instantiated with Pb, the proposition that Mark is caring for his daughter, 
which was found to satisfy ?q in formula 3 below. 
3. ((and (bel s (cr-cause ?q pa)) 
(Plausible (cr-cause ?q pa)) 
(explanation-indicated s h pa))) 
Finally, this satellite is expanded using pc, the proposition that Mark's daughter has 
the measles, which was found to satisfy ?q in formula 4 below. 
4. ((and (bel s (cr-cause ?q pb)) 
(Plausible (cr-cause ?q Pb)) 
(explanation-indicated s h Pb))) 
The output of phase one is a discourse plan for a full answer, as shown in Figure 14. 
The second phase of generation, plan pruning, will walk the tree bottom-up. The root 
424 
Green and Carberry Indirect Answers 
5 
(Answer-no s h p) 
no (Use-contrast s h (not p)) (Use-contrast s h (not p)) 
a (Use-cause s h p ) d 
b (Use-cause s h Pb ) 
C 
Figure 14 
Plan for full answer (before pruning). 
of each subtree has been annotated with a sequence number to show the order in 
which a subtree is visited in the bottom-up traversal of the tree, i.e., 1 through 5. 
Since subtree 1 has no satellites (only a nucleus), the traversal moves to subtree 2. 
For the same reason, the traversal moves to subtree 3. Next, the nucleus of subtree 
3 is tentatively pruned, i.e., a trial response consisting of the direct answer plus (a), 
(c), and (d) is created. Simulated interpretation of this trial response results in the 
inference of a discourse plan identical to the full plan as the most preferred (in fact, 
the only) interpretation of the trial response. Thus, (b) can be pruned, and subtree 4 is 
considered next. By a similar process, (a) is also pruned. Last, the tree with root labeled 
5 is examined, and it is determined that the direct answer (no) can also be pruned. 
The final result of the traversal is that the direct answer, (a), and (b) are marked as 
pruned, and a response consisting of just acts (c) and (d) is returned by the generator. 
5.5 Related Work in Generation 
This work differs from most previous work in cooperative response generation in that 
the information given in an indirect answer conversationally implicates the direct an- 
swer. Hirschberg (1985) implemented a system that determines whether a yes or no 
alone licenses any unwanted scalar implicatures, and if so, proposes alternative true 
scalar responses that do not. In our model, that type of response is generated by con- 
structing a response from an Answer-no or Answer-hedge operator having a single Use- 
contrast satellite, motivated by clarify-extent-indicated, as illustrated in Section 5.2.7. 58 
However, Hirschberg's model does not account for other types of indirect answers, 
which can be constructed using the other operators (or other combinations of the 
above operators) in our model, nor for other motives for selecting Use-contrast such 
as answer-ref-indicated and appeasement-indicated. 
Rhetorical or coherence relations (Grimes 1975; Halliday 1976; Mann and Thomp- 
son 1988) have been used in several text-generation systems to aid in ordering parts of 
a text (e.g., Hovy 1988) as well as in content planning (e.g., McKeown 1985; Moore and 
Paris 1993). The discourse plan operators based on coherence relations in our model 
58 As mentioned earlier, the coherence rules for cr-contrast as well as the rules for clarify-extent-indicated make use of notions elucidated by Hirschberg (1985). 
425 
Computational Linguistics Volume 25, Number 3 
(i.e., the operators used as satellites of top-level operators) play a similar role in con- 
tent planning. However, none of the above approaches model the speaker's motivation 
for selecting optional satellites. Stimulus conditions provide principled discourse-level 
knowledge (based upon principles of efficiency, accuracy, and politeness) for choice of 
an appropriate discourse strategy. Also, stimulus conditions enable content selection 
to be sensitive not only to the current discourse context, but also to the anticipated 
effect of a part of the planned response. Finally, none of the above systems incorpo- 
rate a model of discourse plan recognition into the generation process, which enables 
indirect answers to be generated in our model. 
Moore and Pollack (1992) show the need to distinguish the intentional and in- 
formational structure of discourse, where the latter is characterized by the sort of 
relations classified as subject-matter relations in RST. In our model, the operators used 
as satellites of top-level answer discourse plan operators are based on relations simi- 
lar to RST's subject-matter relations. The primary goals of these operators are similar 
to the effect fields of the corresponding RST relation definitions. However, our model 
does distinguish the two types of knowledge. In our model stimulus conditions reflect, 
though they do not directly encode, communicative subgoals leading to the adoption 
of informational subgoals. For example, the explanation-indicated stimulus condition 
may be triggered in situations when the responder's communicative subgoal would 
lead R to select a Use-cause satellite of Answer-yes, rather than a Use-elaboration satellite. 
Moore and Paris (1993) argue that it is necessary for generation systems to repre- 
sent not only the speaker's top-level goal, but also the communicative subgoals that 
a speaker hoped to achieve by use of an informational relation so that, if that subgoal 
is not achieved, then an alternative rhetorical means can be tried. Although stimulus 
conditions do reflect the speaker's motivation for including satellites in a plan, it was 
beyond the scope of our work to address the problem of failure to achieve a subgoal 
of the original response. Therefore, our system does not record which stimulus condi- 
tion motivated a satellite; if the stimulus condition was recorded in the final plan then 
our system would have access to information about the speaker's motivation for the 
satellite. In our current approach, if a follow-up question is asked then a response to 
the follow-up question is planned independently of the previous response. However, 
if R's beliefs have changed since the original question was asked by Q (e.g., as a result 
of information about Q's beliefs obtained from Q's follow-up question), then it is pos- 
sible in our approach for R's response to contain different information. Furthermore, 
in our approach the original response may provide the information that a questioner 
would have to elicit by follow-up questions in a system that can provide only direct 
answers. 
Finally, our use of interpretation during plan pruning has precursors in previous 
work. In Horacek's,approach to generating concise explanations (Horacek 1991), a set 
of propositions representing the full explanation is pruned by eliminating propositions 
that can be derived from the remaining ones by a set of contextual rules. Jameson 
and Wahlster (1982) use an anticipation feedback loop algorithm to generate elliptical 
utterances. 
6. Implementation and Evaluation 
We have implemented a prototype of the model in Common LISP. The implemented 
system can interpret and generate the types of examples discussed in Sections 4 and 
5 and the specific examples tested in the experiments described below. The overall 
coverage of the implemented system can be defined as all (direct and indirect) re- 
sponses that can be composed from the 5 top-level operators and 10 satellite operators 
426 
Green and Carberry Indirect Answers 
blue block_____~ \] \[ green block purple cone 
. ~/" live mouse 
red block 
Figure 15 
Blocks world picture used in Experiment 1. 
yellow ball 
O 
(for 8 stimulus conditions and 24 coherence relation rules) provided in the model. 
The performance of the system running on a UNIX workstation depends mainly on 
the amount of hypothesis generation performed (which can be controlled by setting a 
parameter limiting the depth of the breadth-first search during hypothesis generation). 
We have evaluated the system with two experiments. The purpose of the first 
experiment was to determine whether users' interpretations of indirect answers would 
agree with the system's interpretations. The purpose of the second experiment was 
to see how users would evaluate the unrequested information selected for an indirect 
answer by the system. The system was run to verify that it could actually interpret 
or generate the responses that were evaluated in the first and second experiments, 
respectively. Each experiment was conducted by means of a questionnaire given to 
10 adult subjects who were not familiar with this research work. At the beginning 
of each questionnaire, subjects were given a brief textual and pictorial description of 
the setting in which the questions and responses supposedly had occurred. (A black- 
and-white version of the picture shown to subjects for the first experiment is given in 
Figure 15. Since the picture used in the experiment was in color, we have annotated 
the objects in the figure to indicate their color. A similar picture was used for the 
second experiment.) The fictional setting was described as a laboratory inhabited by 
a talking robot and a mouse; outside of the laboratory is a manager who cannot see 
inside the laboratory, but the robot and manager can communicate with each other. 
In both experiments, the questionnaire consisted of questions supposedly posed by 
the manager to the robot, and the robot's possible or actual responses. This setting 
was selected because it could easily be presented to subjects with a minimum of 
description about the beliefs that might motivate the robot's responses. Also, a new 
domain was used so that our experience with the other domains would not influence 
the evaluation. 
427 
Computational Linguistics Volume 25, Number 3 
6.1 Experiment 1 
6.1.1 Experiment. The first experiment addressed whether the subjects' interpretations 
of indirect answers would agree with the system's interpretations. The subjects were 
given 19 yes-no question-response exchanges. Each response consisted of from 1 to 3 
sentences without an explicit yes or no, e.g., as in (37) (item 3 in the questionnaire for 
Experiment 1). 
(37) i. Q: Is the yellow ball on the table? 
ii. R: The yellow ball is on the floor. 
(For more examples, see the appendix.) Fourteen of the responses were indirect, i.e., 
our system would interpret them as generated from Answer-No or Answer-Yes. 59 (For 
example, (37)ii was interpreted as a no generated by Answer-No.) These 14 responses 
made use of all of the possible satellites of Answer-No and Answer-Yes in the model. 
Several responses made use of multiple satellites. For example, the response in item 
19 of the questionnaire was similar to the response shown on the right-hand side of 
Figure 8. The other 5 responses we characterize as bogus, i.e., would not be interpreted 
as answers by our system, e.g., (38) (item 2 in the questionnaire). 
(38) i. Q: Can you pick up the ball? 
ii. R: A red block is on the table. 
The purpose of the so-called bogus responses was to make certain that the subjects 
were not just interpreting every response as saying yes or no. For each response, the 
subjects were asked to select one of the following interpretations: 
1. Yes (glossed as I would interpret this as yes) 
2. Yes-? (glossed as I could interpret this as yes but I am uncertain), 
3. No (glossed as I would interpret this as no), 
4. No-? (glossed as I could interpret this as no but I am uncertain), or 
5. Other (glossed as I would not interpret this as yes or no). 
6.1.2 Results. The results are shown in Table 5. The rows of the table present the results 
for each question-response pair. The second column gives the system's interpretation 
of the response for cases where the system would interpret the response as an indirect 
yes or no, or indicates bogus for a bogus response. The third column gives the number 
of subjects who selected yes or no in agreement with the system's interpretation. The 
fourth column gives the number of subjects who selected yes or no in agreement with 
the system's interpretation but with some uncertainty (i.e., a Yes-? or No-?). The last 
column gives the number of subjects who judged the response as Other. 
In the overwhelming majority of cases the subjects' interpretations agreed with 
the systems' interpretation as a yes or no, though occasionally with some uncertainty. 
None of the subjects selected the opposite interpretation, i.e., a Yes~Yes-? for a no, or 
a No~No-? for a yes. Of the 14 questions interpreted as a yes or no by the system, 
59 We have not yet evaluated indirect answers that would be generated by the other three top-level 
operators in our model. 
428 
Green and Carberry Indirect Answers 
Table 5 
Results of experiment on interpretation of indirect answers. 
Example Indirect Indirect Indirect 
Number Answer Interpretation Interpretation (?) Other 
1 Yes 6 3 1 
2 (bogus) 0 1 9 
3 No 9 0 1 
4 No 5 5 0 
5 No 8 2 0 
6 Yes 7 3 0 
7 (bogus) 0 1 9 
8 No 9 1 0 
9 Yes 9 1 0 
10 No 8 2 0 
11 Yes 3 7 0 
12 (bogus) 0 0 10 
13 Yes 8 2 0 
14 No 8 2 0 
15 No 7 3 0 
16 Yes 4 6 0 
17 (bogus) 0 0 10 
18 (bogus) 0 0 10 
19 Yes 10 0 0 
only 2 items were interpreted by subjects as Other (each by a different subject). In 
28% of the instances where the subjects interpreted a response as saying yes or no, 
they noted some degree of uncertainty. During debriefing, the subjects who tended 
to express uncertainty said that while they might interpret the response as yes or no, 
one generally had some uncertainty when the direct answer was omitted. Only one 
subject interpreted a bogus question as answering yes or no. 
To test the statistical significance of the pattern of responses shown in Table 5, we 
took a very conservative approach. We grouped Indirect Interpretation (?) with Other in 
Table 5 for question-response instances where the system interpreted the response as 
an indirect answer, and we grouped Indirect Interpretation (?) with Indirect Interpretation 
for instances where the system did not interpret the response as an indirect answer. 
Thus Indirect Interpretation (?) responses by the subjects were treated as disagreeing 
with the system's interpretation of the example. We then applied Cochran's Q test 
(Cochran 1950) to the resulting two columns of data. The result shows that the pattern 
of responses is statistically significant (not the result of random chance) at better than 
the level p < .005. To determine whether the subjects differentiated between responses 
that the system interpreted as indirect answers and those that it did not, we applied 
the Mann Whitney U statistic (Siegel 1956), which showed no score overlap at the 
level p < .005. 
6.2 Experiment 2 
6.2.1 Experiment. Although the linguistic studies discussed in Section 5 show that in 
human-human dialogue people often include additional unrequested information in 
their responses to yes-no questions, we conducted a second experiment to determine 
how users would evaluate responses consisting of the kinds of extra unrequested 
information produced by our system. The subjects were given 11 yes-no questions 
(some preceded by 1 or 2 sentences to establish some additional context), each with a 
429 
Computational Linguistics Volume 25, Number 3 
set of 4 possible responses. The subjects were told to suppose that all of the responses 
in a set were true, and were asked to select the best response in each set. For each 
question, the response choices included: 
• a direct response of Yes or No (depending on the correct answer to the 
question), 
• the direct response with further emphasis (such as No, I can't), 
• 2 extended responses, containing the direct answer with extra 
unrequested information, 
e.g., (39)iii-(39)vi, respectively. ((39) is item 1 in the questionnaire for Experiment 2.) 
(39) i. M: I am looking for the blue ball. 
ii. Is it on the table? 
iii. R: No. 
iv. R: No. It's not. 
v. R: No. It's on the floor. 
vi. R: No. It cost $5. 
In 9 of the 11 sets, 1 of the extended responses was motivated by our stimulus condi- 
tions (e.g., (39)v), and 1 was not (e.g., (39)vi). In the other 2 sets (questionnaire items 
3 and 7), neither of the extended responses was motivated by any stimulus condition. 
The purpose of these 2 so-called bogus examples was to make certain that the subjects 
were not inclined to always select responses with extra information. 
6.2.2 Results. The results are shown in Table 6. The rows of the table present the 
results for each question. The second column lists the stimulus condition, if any, that 
our system used to trigger one of the extended responses to the question. Items 3 and 
7 contained bogus responses, i.e., none of the responses was motivated by a stimulus 
condition. The next three columns indicate respectively the number of subjects who 
selected the response motivated by the listed stimulus condition, the number who 
selected the direct answer alone or the direct answer with emphasis but no additional 
information, and the number who selected an extended response not motivated by 
a stimulus condition. Note that none of the subjects selected a response with extra 
information for the two bogus questions, indicating that they were not merely inclined 
to select responses with extra information. 
Items 8 and 10 warrant some discussion. Question 8 was problematic. The origi- 
nal question given to the first four subjects asked whether the robot could tell the lab 
manager the time. The response "No. There is no clock in here." was motivated by the 
stimulus condition excuse-indicated. However, two of the four subjects selected just No 
as the best response, and explained during debriefing that if the robot could tell time, 
then certainly he had an internal clock that he could use (since all computers have 
internal clocks) and thus the absence of a clock in the room was not relevant. Since 
the prior beliefs of these subjects conflicted with the beliefs that were intended as the 
context for interpreting the robot's response, we altered the question for the remainder 
of the study to circumvent this problem. In item 10, the extra information in the sys- 
tem's response was motivated by the appeasement-indicated stimulus condition. In that 
430 
Green and Carberry Indirect Answers 
Table 6 
Results of experiment on including extra information. 
Direct Other 
Example Answer Extended 
Number Stimulus Condition SC Response SC Only Response 
1 answer-re f-indicated 10 0 0 
2 substitute-indicated 10 0 0 
3 none 0 10 0 
4 explanation-indicated 9 1 0 
5 clarify-extent-indicated 10 0 0 
6 clarify-condition-indicated 10 0 0 
7 none 0 10 0 
8 excuse-indicated 8 2 0 
9 clarify-concept-indicated 10 0 0 
10 appeasement-indicated 4 5 1 
11 explanation-indicated 9 1 0 
response, the robot answers No (that he has not yet done the requested task) and then 
attempts to appease the questioner by describing another task that he has completed. 
Since only 4 of the 10 subjects selected this response, it is possible that the subjects 
did not view appeasement as an appropriate stimulus condition in human-machine 
dialogue, despite the fact that it does occur in human-human dialogue. Alternatively, 
the subjects did not have enough information to recognize the response as attempted 
appeasement. 
To test the statistical significance of the pattern of responses in Table 6, we again 
took a conservative approach and grouped Other Extended Response (which was selected 
only once by a subject) with Direct Answer Only so that it was treated as disagreeing 
with the system response. Once again we applied Cochran's Q test (Cochran 1950) and 
the Mann Whitney U statistic (Siegel 1956). Cochran's Q test showed that the pattern 
of responses in Table 6 is statistically significant at the level p < .005, and the Mann 
Whitney U statistic showed that there is no score overlap at the level p ~ .0253. 
6.3 Summary 
The first experiment suggests that our system's interpretations of indirect answers 
agree with the judgments of human interpreters. The second experiment suggests that 
our stimulus conditions result in the construction of responses containing extra infor- 
mation that users will view favorably. However, we have not addressed the question 
of when other stylistic considerations might limit the amount of extra information that 
is included, nor the question of choosing between a direct and an indirect response 
when both are possible. These issues will be addressed in future research. 
7. Conclusions 
In summary, we have proposed and implemented a computational model for inter- 
preting and generating indirect answers to yes-no questions in English. This paper 
describes the knowledge and processes provided by the model. Generation and inter- 
pretation are treated, respectively, as construction of and recognition of a responder's 
discourse plan for a full answer. A discourse plan explicitly relates a speaker's beliefs 
and discourse goals to his program of communicative actions. An indirect answer is 
the result of the responder providing only part of the planned response, but intending 
431 
Computational Linguistics Volume 25, Number 3 
for his discourse plan to be recognized by the questioner. Discourse plan construction 
and recognition make use of the beliefs that are presumed to be shared by the partic- 
ipants, as well as shared knowledge of discourse strategies, represented in the model 
by shared discourse plan operators. In the operators, coherence relations are used to 
characterize types of satellites that may accompany each type of answer. Recognizing 
a mutually plausible coherence relation obtaining between the actual response and 
a possible direct answer plays an important role in recognizing the responder's dis- 
course plan. The use of hypothesis generation in interpretation broadens the coverage 
of the model to cases where more is missing from a full answer than just the nucleus of 
a top-level operator. (From the point of view of generation, it enables the construction 
of a more concise, though no less informative, response.) Stimulus conditions model a 
speaker's motivation for selecting a satellite. During generation, the speaker uses his 
own interpretation capability to determine what parts of the plan are inferable by the 
hearer in the current discourse context and thus do not need to be explicitly given. 
We argue that because of the role of interpretation in generation, Q's belief that R 
intended for Q to recognize the answer is warranted by Q's successful recognition of 
the plan. 
Although it was not our goal to develop a cognitive model of how implicatures are 
produced and comprehended, certain aspects of the model might be incorporated into 
a cognitive model. To a large extent the model is recognitional rather than inferential. Of 
course, we make no claims about the cognitive plausibility of the particular coherence 
relations and discourse plan operators used in our model, which were encoded solely 
on the basis of their descriptive and computational utility. We await further cognitive 
studies on coherence relations as begun in Sanders, Spooren, and Noordman (1992), 
Knott and Dale (1994), and Knott (1995). Since the work reported in this paper was 
performed, the first author has investigated the automatic compilation of discourse 
plan operators in a computational cognitive architecture (SOAR) (Green and Lehman 
1998). 
In conclusion, our model provides wider coverage than previous computational 
models for generating and interpreting answers. Specifically, it covers both direct and 
indirect answers, multiple-sentence responses, a variety of types of indirect answer 
(i.e., characterized in terms of multiple coherence relations), and multiple types of 
speaker motivation for deciding to provide extra information (i.e. characterized in 
terms of different stimulus conditions). In addition, it appears that this approach could 
be extended to other discourse-expectation-based types of conversational implicature. 
As a computational model of conversational implicature, it extends current plan-based 
theories of implicature in several ways. First, it demonstrates the role of shared dis- 
course expectations and pragmatic knowledge. Second, it makes predictions about 
cancelability in terms of intentional structure of discourse. Lastly, it treats generation 
as a process drawing upon the speaker's own interpretation mechanism. 
Appendix: Questionnaire for Experiment 1 
1. Q: Can you make a stack of 3 blocks? 
R: I can put the green block on the blue block. 
2. Q: Can you pick up the ball? 
R: A red block is on the table. 
3. Q: Is the yellow ball on the table? 
R: The yellow ball is on the floor. 
432 
Green and Carberry Indirect Answers 
. 
. 
. 
. 
. 
. 
10. 
11. 
Q: Can you pick up the green block? 
R: The green block is glued to the table. 
I glued it yesterday. 
Q: Can you build a stack with the green block on top of the cone? 
R: The cone does not have a flat top. 
Q: Are you going to move the green block? 
R: The green block is very heavy. 
I'm going to use the forklift. 
Q: Can the mouse climb up the red block? 
R: The yellow ball is on the floor. 
Q: Can you build a stack with the green block on top of the cone? 
R: The green block would fall off the cone. 
Q: Can you pick up the red block? 
R: If I move the blue block off of it. 
Q: Can you pick up the blue block? 
R: My arms won't move. 
The human forgot to oil me. 
Q: Is there a yellow ball on the floor? 
R: The mouse pushed the yellow ball off the table. 
12. Q: Are there any blocks on the table? 
R: The table is about three feet tall. 
13. Q: Is something on the red block? 
R: I put the blue block on the red block. 
14. Q: Can you put the mouse on the green block? 
R: He runs too fast. 
15. 
16. 
17. 
18. 
Q: Is the blue block on the table surface? 
R: I put the blue block on the red block. 
Q: Did you move the cone off the green block? 
R: I wanted to pick up the green block. 
Q: Is the blue block on the red block? 
R: The mouse squeaks a lot. 
Q: Did you pick up the cone? 
R: There are three blocks, one cone, a yellow ball, and a mouse. 
19. Q: 
R: 
Are you going to pick up the blue block? 
The blue block is sticky. 
The mouse poured honey on it. 
I am going to use a pair of tongs. 
433 
Computational Linguistics Volume 25, Number 3 
Acknowledgments 
This describes the first author's dissertation 
research at the University of Delaware. We 
would like to thank Dan Chester of the 
University of Delaware for providing us 
with an implementation of a Horn clause 
theorem prover (Chester 1980) to use in this 
work, and Fred Masterson also of the 
University of Delaware for help with the 
statistical analysis of the experiments. Also, 
we wish to thank the journal referees for 
their helpful comments. 
References 
Allen, James E and C. Raymond Perrault. 
1980. Analyzing intention in utterances. 
Artificial Intelligence 15:143-178. 
Brown, Penelope and Stephen C. Levinson. 
1978. Universals in language usage: 
Politeness phenomena. In Esther N. 
Goody, editor, Questions and Politeness: 
Strategies in Social Interaction. Cambridge 
University Press, pages 56-289. 
Carberry, Sandra. 1990. Plan Recognition in 
Natural Language Dialogue. MIT Press, 
Cambridge, MA. 
Chester, Daniel. 1980. HCPRVR: An 
interpreter for logic programs. In 
Proceedings of the First Annual National 
Conference on Artificial Intelligence, 
pages 93-95. 
Clark, Herbert H. 1979. Responding to 
indirect speech acts. Cognitive Psychology 
11:430-477. 
Clark, Herbert and C. Marshall. 1981. 
Definite reference and mutual knowledge. 
In Aravind K. Joshi, Bonnie Webber, and 
Ivan Sag, editors, Elements of Discourse 
Understanding. Cambridge University 
Press. 
Cochran, W. G. 1950. The comparison of 
percentages in matched samples. 
Biometrika 37:256-266. 
Dahlgren, Kathleen. 1989. Coherence 
relation assignment. In Proceedings of the 
Eleventh Annual Meeting of the Cognitive 
Science Society, pages 588-596. 
Green, Nancy L. 1994. A Computational 
Model for Generating and Interpreting Indirect 
Answers. Ph.D. thesis, University of 
Delaware. 
Green, Nancy and Jill F. Lehman. 1998. An 
application of explanation-based learning 
to discourse generation and 
interpretation. In Papers from the 1998 
AAAI Spring Symposium on Applying 
Machine Learning to Discourse Processing, 
pages 33-39. 
Grice, H. Paul. 1975. Logic and 
conversation. In Peter Cole and Jerry L. 
Morgan, editors, Syntax and Semantics III: 
Speech Acts. Academic Press, pages 41-58. 
Grimes, Joseph E. 1975. The Thread of 
Discourse. Mouton, The Hague. 
Gunji, Takao. 1981. Toward a Computational 
Theory of Pragmatics--Discourse, 
Presupposition, and Implicature. Ph.D. 
thesis, Ohio State University. 
Halliday, M. A. K. and Ruqaiya Hasan. 1976. 
Cohesion in English. Longman, New York. 
Hinkelman, Elizabeth A. 1989. Linguistic and 
Pragmatic Constraints on Utterance 
Interpretation. Ph.D. thesis, University of 
Rochester. 
Hirschberg, Julia B. 1985. A Theory of Scalar 
Implicature. Ph.D. thesis, University of 
Pennsylvania. 
Hobbs, Jerry R. 1978. Resolving pronoun 
references. Lingua, 44:311-338. 
Horacek, Helmut. 1991. Exploiting 
conversational implicature for generating 
concise explanations. In Proceedings of the 
European Association for Computational 
Linguistics, pages 191-193. 
Hovy, Eduard H. 1988. Planning coherent 
multisentential text. In Proceedings of the 
26th Annual Meeting, pages 163-169. 
Association for Computational 
Linguistics. 
Jameson, Anthony and Wolfgang Wahlster. 
1982. User modelling in anaphora 
generation: Ellipsis and definite 
description. In Proceedings of the European 
Conference on Artificial Intelligence. 
Kiefer, Ferenc. 1980. Yes-no questions as 
wh-questions. In John Searle, Ferenc 
Kiefer, and Manfred Bierwisch, editors, 
Speech Act Theory and Pragmatics. Reidel, 
Dordrecht, Holland, pages 48-68. 
Knott, Alistair. 1995. A Data-Driven 
Methodology for Motivating a Set of Coherence 
Relations. Ph.D. thesis, University of 
Edinburgh. 
Knott, Alistair and Robert Dale. 1994. Using 
linguistic phenomena to motivate a set of 
coherence relations. Discourse Processes, 
35-62. 
Lascarides, Alex and Nicholas Asher. 1991. 
Discourse relations and defeasible 
knowledge. In Proceedings of the 29th 
Annual Meeting, pages 55-62. Association 
for Computational Linguistics. 
Lascarides, Alex, Nicholas Asher, and Jon 
Oberlander. 1992. Inferring discourse 
relations in context. In Proceedings of the 
30th Annual Meeting, pages 1-8. 
Association for Computational 
Linguistics. 
434 
Green and Carberry Indirect Answers 
Levinson, Stephen C. 1983. Pragmatics. 
Cambridge University Press, Cambridge. 
Litman, Diane J. 1986. Understanding plan 
ellipsis. In Proceedings of the Fifth National 
Conference on Artificial Intelligence, 
pages 619-624. 
Mann, William C. and Sandra A. 
Thompson. 1983. Relational propositions 
in discourse. Technical Report 
ISI/RR-83-115, Information Sciences 
Institute, University of Southern 
California, Marina del Rey, CA. 
Mann, William C. and Sandra A. 
Thompson. 1988. Rhetorical Structure 
Theory: Toward a functional theory of 
text organization. Text 8(3):167-182. 
McCafferty, Andrew S. 1987. Reasoning about 
Implicature: A Plan-Based Approach. Ph.D. 
thesis, University of Pittsburgh, 
Pittsburgh. 
McKeown, Kathleen R. 1985. Text Generation. 
Cambridge University Press, Cambridge. 
Moore, Johanna D. and Cecile Paris. 1993. 
Planning text for advisory dialogues: 
Capturing intentional and rhetorical 
information. Computational Linguistics 
19(4):651-694. 
Moore, Johanna D. and Martha E. Pollack. 
1992. A problem for RST: The need for 
multi-level discourse analysis. 
Computational Linguistics 18(4):537-544. 
Perrault, Raymond and James Allen. 1980. 
A plan-based analysis of indirect speech 
acts. American Journal of Computational 
Linguistics 6(3-4):167-182. 
Pollack, Martha. 1990. Plans as complex 
mental attitudes. In Philip R. Cohen, Jerry 
Morgan, and Martha Pollack, editors, 
Intentions in Communication. MIT Press, 
Cambridge, MA. 
Reichman, Rachel. 1985. Getting Computers 
To Talk Like You And Me. MIT Press, 
Cambridge, MA. 
Sanders, Ted J., Wilbert P. Spooren, and Leo 
G. Noordman. 1992. Toward a taxonomy 
of coherence relations. Discourse Processes 
15:1-35. 
Schegloff, Emanuel A. 1972. Sequencing in 
conversational openings. In J. J. Gumperz 
and Dell H. Hymes, editors, Directions in 
Sociolinguistics. Holt, Rinehart and 
Winston, New York, pages 346-80. 
Schegloff, Emanuel A. 1979. Identification 
and recognition in telephone conversation 
openings. In G. Psathas, editor, Everyday 
Language: Studies in Ethnomethodology. 
Irvington, New York, pages 23-78. 
Siegel, Sidney. 1956. Nonparametric Statistics 
for the Behavioral Sciences. McGraw-Hill, 
New York. 
SRI Tapes. 1992. Transcripts of audiotape 
conversations. Prepared by Jacqueline 
Kowto under the Direction of Patti Price 
at SRI International, Menlo Park, CA. 
StenstrOm, Anna-Brita. 1984. Questions and 
responses in English conversation. In 
Claes Schaar and Jan Svartvik, editors, 
Lund Studies in English 68. CWK Gleerup, 
Malm6, Sweden. 
Thomason, Richmond H. 1990. 
Accommodation, meaning, and 
implicature: Interdisciplinary foundations 
for pragmatics. In Philip R. Cohen, Jerry 
Morgan, and Martha Pollack, editors, 
Intentions in Communication. MIT Press, 
Cambridge, MA, pages 325-363. 
Walker, Marilyn A. 1993. Informational 
Redundancy and Resource Bounds in 
Dialogue. Ph.D. thesis, University of 
Pennsylvania. 
Young, R. Michael, Johanna D. Moore, and 
Martha E. Pollack. 1994. Towards a 
principled representation of discourse 
plans. In Proceedings of the Sixteenth Annual 
Meeting of the Cognitive Science Society, 
pages 946-951. 
435 
