The Basic Block Model of Extended Explanations 
David J. Mooney and Sandra Carberry and Kathleen F. McCoy 
Department of Computer and Information Sciences 
103 Smith Hall 
University of Delaware 
Newark, Delaware 19716 
Abstract 
In this paper we argue that current generation method- 
ologies are inadequate for determining the high-level 
structure characteristic of naturally-occurring extended 
explanations. Our analysis of such explanations indi- 
cates that high-level structure - composed of a unify- 
ing framework and its associated basic blocks - must 
be determined by bottom-up processes that attempt to 
satisfy speaker, listener, and compositional goals, after 
which top-down strategies can be used to organize the 
material about the selected framework. In addition to a 
description of this structure, this paper describes three 
types of repetition whose use is dependent on this high- 
level structure; their use not only contributes to the co- 
hesiveness of extended explanations, but supports our 
thesis of the non-recursive nature of the high-level struc- 
ture. We conclude with an outline of our computational 
strategy for generating this structure. 
INTRODUCTION 
In this paper, we describe the high-level structure of 
naturally-occurring, extended explanations and con- 
tend that the organizational strategies currently em- 
ployed for structuring short explanations are inade- 
quate for generating this characteristic high-level struc- 
ture. Our analysis suggests that text structure is not 
completely recursive as others have claimed (\[Grosz and 
Sidner, 1986\], \[Reichman, 1978\], \[Polanyi, 1986\], \[Mann 
and Thompson, 1988\]), but rather that the high-level 
structure of extended explanations is determined by 
processes separate from those which organize text at 
lower levels. Evidence for the non-recursive nature of 
this structure is provided by repetition, which is fre- 
quently used in extended explanations for a wide variety 
of purposes. This device, whose use has been neglected 
by current explanation research, exhibits a distinct re- 
lationship to the structure of extended explanations, 
implying that it, too, cannot be generated by recursive 
strategies. 
Our analysis is based on several sources, the primary 
of which is the testimony before congressional com- 
mittees regarding the Three Mile Island (TMI) acci- 
dent (\[Representatives, 1979\], \[Representatives, 1980\]). 
These explanations were punctuated with numerous in- 
terruptions and characteristically involved the opera- 
tion of complex physical devices and extended causal 
chains of events. The majority of these explanations 
were planned beforehand and required many pages of 
text - typically five, ten, or more, exclusive of interrup- 
tions. Furthermore, the testimony comes from a variety 
of sources: operators who were on the scene as the acci- 
dent developed; engineers who gave post-accident eval- 
uations; executives from the company that designed the 
plant. This variety allowed examination of explanations 
of the same set of events from many perspectives. 
A secondary source was transcripts of a medical class 
involved in patient diagnosis. While these transcripts 
differed greatly form those of TMI - in particular, the 
explanations provided were relatively short; they were 
unplanned; and they were concerned with diagnostic 
strategy as opposed to causal events - they exhibited 
many of the characteristics found in the TMI explana- 
tions. 
The next section provides a brief overview of current 
models for structuring text, followed by a description 
of the basic block, the unit of discourse on which our 
model of discourse is based. We then identify three 
types of repetition whose use supports our contentions 
of the non-recursive nature of basic block structure. We 
conclude by describing the characteristics of high-level 
structure of extended explanations, followed by a brief 
description of our strategy for computationally gener- 
ating this structure. A more detailed description of our 
strategy and its role in a complete generational system 
is contained in \[Mooney, el al. 1989\]. 
THE BASIC BLOCK MODEL OF 
EXTENDED DISCOURSE 
Given any large body of information to convey, there 
are many possible combinations of that material, some 
more cohesive than others. Frequently, deciding how to 
organize a large body of material is more difficult than 
determining what to include. Our research is concerned 
with the identification of a coherent unifying framework 
112 
about which an extended explanation can be organized 
and the criteria for selecting from among several frame- 
works when more than one viable alternative exists. 
CURRENT MODELS OF TEXT 
STRUCTURE 
COMPUTATIONAL MODELS Two gen- 
eral methodologies have been applied to the structur- 
ing of explanations: schemas (\[McKeown, 1985\], \[Mc- 
Coy, 1985\], \[Paris, 1987\]) and plans based on rhetori- 
cal structure theory (RST) (\[Hovy and McCoy, 1989\], 
\[Hovy, 1988a\], \[Moore and Paris, 1988a\], \[Moore and 
Swartout, 1988b\]). Each methodology assumes that 
text is hierarchically structured and that its operators 
can be recursively expanded at any level within the dis- 
course hierarchy. While these methods have proven 
to be effective for organizing short pieces of text, we 
maintain that they are inadequate for generating the 
characteristic high-level structure of eztended explana- 
tions and certain forms of repetition. Rather, we main- 
tain that this high-level structure can only he captured 
by a separate, bottom-up process. Once this has been 
completed, the integration of appropriate repetition can 
proceed in conjunction with, hut independently of, the 
application of recursive strategies to the organization of 
text at lower levels. 
RHETORIC Rhetoric, the formal study of the art 
of good writing, provides general strategies for orga- 
nizing text at a high level that are absent from the 
computational models. Analysis - "the method of ex- 
planation whereby a subject is divided into its separate 
component parts" 1 _ is possibly the most instrumental 
of these strategies. There are no hard-and-fast rules 
for determining what constitutes an appropriate anal- 
ysis of a subject. As \[Wicker and Albrecht, 1960\] ob- 
serves, a subject may be classified in as many ways as 
it has characteristics/parts/stages/etc. However, there 
are three criteria which experts (\[Wicker and Albrecht, 
19601, \[Arena, 1975\], \[Thompson, 19571, \[Daniel, 19671, 
\[Kane and Peters, 1966\]) mutually consider essential for 
a satisfactory organizational strategy: 
1. The scheme should be logical; a single, consistent 
criterion should be used for the analysis (e.g., time, 
steps in a process). 
2. The scheme should exhaust all of the possibilities; 
everything to be conveyed should be encompassed by 
the scheme. 
3. The resultant categories should be mutually exclu- 
sive; nothing should belong to more than one. 
While the type of explanation with which this paper 
is concerned exhibits a high-level organization reflec- 
tive of these criteria, the criteria by themselves do not 
l\[Arena, 1975\] page 107 
Now, what happened at Three Mile Island was that a 
feedwater transient was experienced, and by that I mean, 
simply, we lost feedwater to the plant momentarily. 
Now, with loss of feedwater to the steam generator, the 
plant will experience a momentary pressurization above its 
normal pressure. This pressurization is released by a relief 
valve located at the top of the pressurizer. It simply opens 
and lets out a little bit of steam to take care of the excess 
pressure that is < interruption occurs here > 
Then at 15 seconds into the event-keeping in mind that 
the valves opened maybe 5 seconds into the event-at 15 sec- 
onds the pressure started coming down because the valve 
had opened and cut off the pressure. 
The valve should have reclosed when it got back down 
to about 2,250 pounds; it did not reclose. The pressure pro- 
ceeded to comeon down. At about 30 seconds into the event, 
this water here started disappearing, of course, because now 
you are continuing to remove a very large amount of heat 
here, which is then coming off as secondary side steam gen- 
eration, and this water will proceed to disappear if you do 
not start replacing it. 
And the auxiliary feedwater which normally comes on to 
make sure this does not go dry came on at about 30 seconds 
into the event. And at least the pumps were running. So this 
picture here is just the first 30 seconds and-or thereabouts. 
And this figure here is indicative of the situation from 1 
to 6 minutes into the event. 
Figure 1: A Typical Primary Segment 
provide the specificity necessary for computational gen- 
eration. In addition, these guidelines include no sugges- 
tions for dealing with situations in which no logical, all- 
inclusive framework can be identified, nor do they offer 
suggestions for selecting among several organizational 
schemes which meet the prescribed criteria equally well. 
Furthermore, the guidelines are not sufficient in-and-of 
themselves to account for all of the observed phenom- 
ena discussed in the following sections. 
BASIC BLOCKS 
Our model is based on a discourse unit which we have 
termed a basic block. A basic block consists of two ele- 
ments: 
1. an organizational focus, such as a person or location, 
and 
2. a set of propositions to be conveyed that are related 
to that focus. 2 
The focus is what makes a cohesive unit of the material 
in the block; it is the thread common to all of this 
material, whether directly or indirectly. 
2The term basic blockis also used to refer to the segment 
of text generated from these propositions. 
113 
A basic block will be realized as a primary segment 
of text which occupies the first level of the discourse 
hierarchy. In a coherent discourse, the foci on which 
the basic blocks are based are themselves related, each 
representing a different aspect of some unifying frame- 
work. This framework is analogous to the organiza- 
tional scheme of rhetorical analysis. 
The testimony from which the basic block in Fig- 
ure 1 was extracted 3 illustrates these points. This block 
references a particular time frame: zero to thirty sec- 
onds of the accident at TMI. The remaining blocks of 
that testimony (not shown) are' similarly constructed 
around time frames, e.g., one to six minutes, six min- 
utes to one hour, etc. Observed frameworks demon- 
strate a gamut of types: properties of the concepts (lo- 
cation, time), planning strategies in which events are 
involved (medical diagnosis), and characteristics that 
are not only inherent in the material but also due in 
part to the speaker's perception of them (significant 
factors). There appears to be no limit to what can con- 
stitute an acceptable framework, only that it is derived 
from the material itself and not from an independent 
device solely concerned with text structure. What may 
be a potential framework for one set of material may be 
totally inadequate for another. Note that these features 
are reflective of the guidelines suggested by analysis. 
In addition to forming a cohesive unit, basic block 
structure is explicitly distinguished in the following two 
ways. First, it is often explicitly marked. In Figure 1, 
the speaker closes the block with explicit reference to 
its focus: "So this picture here is just the first 30 sec- 
onds and-or thereabouts." The subsequent block in this 
testimony is also well marked, this time by its initial 
sentence (which is the last sentence of the figure): "And 
this figure here is indicative of the situation from 1 to 
6 minutes." While the explicit marking of basic blocks 
in discourse provides persuasive evidence for their dis- 
tinctive role in the discourse hierarchy, even stronger 
testimony is supplied by the manner in which certain 
forms of repetition are used. 
REPETITION 
Our analysis of naturally-occurring extended explana- 
tions has identified three types of repetition whose use 
reflects a close dependence on the basic block structure 
of extended explanations: Final Summaries, Transition 
Summaries, and Back-on-Track repetitions. 
FINAL SUMMARIES 
The term Final Summary is self-descriptive: it appears 
at the end of a discourse 4, emphasizing the major points 
3 Space limitations prevent inclusion of the complete text. 
4We loosely interpret the term discourse to refer to a 
piece of text that can stand as an independent discourse 
even when extracted from a larger text of which it may be 
a component, as in the case of a chapter from a technical 
book. 
1. They are getting a little out of place here. 
2. So at the end of one hour things look pretty good. 
3. The fact is around this period of time he also got the 
water up in the steam generator, but he had plenty of heat 
removal capability. 
4. But I am sure he was getting a little concerned about the 
reactor coolant pumps because they were - their flow was 
dropping off, and if you read some of the material produced 
by the operators and the utility, it indicates they were get- 
ting vibration, and they were getting ammeter surging, the 
surging effect on the ammeter itself. 
5. So there was something strange about pumping this fluid, 
whatever it was; he did not and has no way to know that 
this is here, and he has no way to know that this condition 
has developed over here, no direct way. \[Indicating\] 
6. He does know that his pressurizer is in good condition, 
and he is trained, I believe, to think that when the pressur- 
izer is full, the system is in good shape, a 
aFrom \[Representatives, 1979\] 
Figure 2: Transition Summary: Events 
of the text. The final summary serves several purposes: 
• It is a means of emphasizing material that an author 
finds most important. 
• It is a vehicle for including information that does not 
fit within the framework of the main body of text. 
• It is a "traditional" closing device. 
In the final summaries which we have examined, there 
is a strong correlation between the points included in 
the summary and the basic block structure of the text. 
Although a final summary is not constrained to address 
each basic block, there is usually a one-to-one relation 
between the foci of the blocks and the items stressed in 
a summary. These features are demonstrated by the 
rather extensive final summary contained in the ap- 
pendix 5 
TRANSITION SUMMARIES 
Unlike the situation for final summaries, both the 
conditions calling for the use of a transition summary 
and the determination of its contents are relatively 
clear-cut. Transition summaries are similar to final 
summaries in that they summarize important informa- 
tion, but the similarities end there. While final sum- 
maries are used to conclude a discourse, transition sum- 
maries are used to signal a shift in circumstances that 
5The summary has been condensed from the original, 
which was punctuated with numerous questions. Although 
the speaker indicates six major points, only five are dis- 
cussed, the extended questioning preventing him from con- 
cluding his summary. 
114 
are not implicated by the preceding material and for 
which the listener is probably unprepared; where a fi- 
nal summary stresses the major points of a complete 
discourse, a transition summary emphasizes that ma- 
terial instrumental to a listener's understanding of this 
change in direction and consequently prevents poten- 
tial confusion. Transition summaries consist of a sum- 
mary of the material which implicates an expected, but 
unrealized, continuance of the discourse. Additionally, 
reference to the material which is responsible for pre- 
venting this expected outcome and an indication that 
everything is not as it seems may also be included. Note 
that this information, unlike that in a final summary, 
may play a very minor role in the overall text. Transi- 
tion summaries also exhibit a relation to basic blocks. 
While final summaries are found only at the end of a 
text, transition summaries have only been observed at 
the end of blocks within the body of the text. 
Figure 2 provides a typical example of a transition 
summary. Lines 3 and 6 describe the facts, presented 
in the testimony preceding this summary, that imply 
to the operator that everything is under control; lines 
2 and 6 summarize this condition. However, the facts 
contained in lines 4 and 5 (also presented in the earlier 
testimony), indicate that all is not as well as the op- 
erator may have thought. In fact, the conditions they 
describe lead to the next sequence of events, in which 
the serious problems become evident. Thus, this sum- 
mary provides a transition from a situation in which 
things seem to be under control to one in which they 
are not. 
BACK-ON-TRACK REPETITION 
Back-on-track repetition is a recovery device, usually 
brief, which occurs after an interruption. Our tran- 
script analysis suggests that its primary use is to refo- 
cus the dialogue on the interrupted segment. Its form 
is the same regardless of the cause of the interruption, 
consisting of one of the following: 
• a repetition of the last concept presented (or par- 
tially presented) before the interruption. This often 
consists of an (almost) exact reproduction of the orig- 
inal phrase; 
• one or two lines summarizing the main point(s) im- 
mediately preceding the interruption; 
• a sentence that refers generally to the interrupted 
context. 
Figure 3 provides an example of the first type. 
Research (e.g., \[Reichman, 1978\], \[Grosz and Sidner, 
1986\]) indicates that an interrupting segment is explic- 
itly closed before returning to the one interrupted. Al- 
though repetition cannot serve as a closing device it- 
self, it frequently accompanies those that do. While 
the conditions for use of back-on-track repetition are 
vague, we have observed that back-on-track repetitions 
very frequently accompany interruptions which occur 
At this point the operator was obliged to shut off the reactor 
coolant pumps, the reason being as follows: we know the 
fact that there were significant vibrations as determined by 
vibrational logs, and there were ammeter problems. As the 
operator observed the ammeter, it was oscillating. 
< interruption > 
OK. He was obliged to trip the reactor coolant pumps, a 
aFrom \[Representatives, 1979\]. Italics mine. 
Figure 3: Back-On-Track: Literal Repeat 
within a basic block, but not between them. This can be 
explained by hypothesizing that if a speaker has com- 
pleted a block, his mental stack is essentially empty 
(except for the highest level goal); consequently there 
is nothing to which to return. However, when the block 
has not been completed, the speaker needs to return 
attention to the open focus space on top of the stack. 
This serves not only as a clue to the listener but may 
also allow the speaker to collect his thoughts. 
The relationship between basic blocks and repeti- 
tion is strong evidence for the distinctive role played 
by basic blocks in the discourse hierarchy; it also ar- 
gues the inadequacy of recursive strategies to generate 
these rhetorical devices. Completely recursive organi- 
zational strategies assume that their operators can be 
expanded at any level within the discourse hierarchy. 
Consequently, we could expect them to generate tran- 
sition summaries at the end of any segment of text, not 
just at the end of a basic block; we could expect them 
to generate final summaries at the end of any segment 
which contained several constituent subsegments, not 
just at the end of an entire discourse; and we could 
expect them to generate back-on-track repetition after 
any interruption, not only those occurring within a ba- 
sic block. However, the above types of repetition ex- 
hibit a distinct relationship to the basic block structure 
and to no other level of segmentation smaller than the 
basic block within the hierarchy. Since the application 
of a recursive operator is independent of the level of 
recursion, such organizational strategies cannot guar- 
antee that these types of repetition will be generated 
only in relation to the basic block structure. 
CHARACTERISTIC HIGH-LEVEL 
STRUCTURE 
Given the existence of a high-level structure as evi- 
denced by explicit markings and repetition, we must 
consider how such a framework is chosen. If the only 
consideration were the identification of a set of related 
115 
foci which can partition the material to be conveyed, 
then any set of related concepts about which the ma- 
terial can be cohesively organized would suffice. Con- 
sider the motivation behind the block of Figure 1. On 
the surface, the events in this block all occurred within 
the first thirty seconds of the accident. However, it is 
doubtful whether the driving force behind the construc- 
tion of this block was to communicate the time frame 
in which these events occurred; rather, what is of im- 
portance is their relative sequence in the total series 
of events, their cause-effect relations, and their impact 
on the resulting accident. One could argue that the 
individual events in this block represent a cause-effect 
chain, and hence their mutual grouping; but this chain 
is continued in the next block of the testimony. Ap- 
parently, other factors beyond the ability to cohesively 
juxtapose clauses contribute to the segmentation and 
the high-level framework about which it is constructed. 
In most of the dialogues examined, basic blocks from 
a given discourse were of approximately the same size. 
Furthermore, the size of basic blocks was approximately 
the same from dialogue to dialogue; the "ideal" size of 
a basic block seems to be three or four paragraphs in 
length for an explanation of three pages. We contend 
that the high-level structure of extended explanations 
reflects the characteristics of an "ideal" framework in 
which: 
• The basic block foci about which the material is or- 
ganized reflect various aspects of the unifying frame- 
work. 
• Basic blocks are the same size. 
• The length of each basic block approximates the ideal 
size. 
Our basic block model of discourse posits that the at- 
tainment of each of these features, termed composilional 
goals, is instrumental in the selection of a high-level 
framework. 
We postulate that a speaker attempts to identify an 
organizational framework that is capable of coherently 
expressing all of the material he wishes to convey while 
satisfying the compositional goals. Of course, a speaker 
is rarely blessed with the ideal situation. As a result, 
certain anomalies do occur; e.g., we have observed ex- 
planations that exhibit a well-defined high-level struc- 
ture that accounts for all of the basic blocks except for 
the final one, which is at best loosely related to the oth- 
ers. In addition, some explanations contain final sum- 
maries that include material not found in the text (see 
FINAL SUMMARIES), while others are composed 
of blocks that are not balanced. The problem appears 
to be one of finding a satisfactory, rather than an opti- 
mal, unifying framework. It seems that the satisfaction 
of some goals will be sacrificed in favor of others so that 
a framework that provides the best overall solution will 
be selected. 
A MODEL FOR GENERATION 
We are currently developing a computational system 
for generating extended explanations that captures 
the structural characteristics observed in naturally- 
occurring, descriptive explanations. Our strategy, mo- 
tivated by the basic block model of discourse, is based 
on the hypothesis that the high-level structure of a 
discourse can be determined by bottom-up processes 
that attempt to satisfy speaker, listener, and compo- 
sitional goals. Once this organization has been estab- 
lished, top-down processes (e.g., RST or schemas. See 
\[Hovy, 1988b\] or \[McKeown, 1985\], for example.) can 
be used to organize the information within the resulting 
basic blocks and to supplement that information based 
on the choice of framework. This section describes our 
general strategy for identifying an organizational frame- 
work and the resulting basic blocks given an initial set 
of concepts to be conveyed. 
IDENTIFICATION OF CANDIDATE 
FRAMEWORKS 
We suggest that a speaker, when organizing an ex- 
tended explanation, will be faced with one of the fol- 
lowing general situations: 
• He already has a well-defined organizational structure 
in which the material to be conveyed has already been 
organized. 
• He has a set of goals about which he wants to struc- 
ture the discourse. In this situation, the goals serve 
as basic block foci about which he must now attempt 
to partition his material. 
• The speaker is starting "from scratch"; he needs to 
find a unifying framework about which he can struc- 
ture what he wants to say. It is this situation which 
our strategy intends to capture. 
Our strategy assumes that the generation process 
starts with some initial state of affairs and a goal to 
be achieved. Based on this, a rough set of information 
to be conveyed will be identified. The selection of this 
material is not our concern; rather, we are concerned 
with determining a reasonable high-level structure for 
this information. We assume that the material to be 
conveyed has been categorized into at least two levels 
of importance: that which must be included at all costs 
(the kernel), and that which would be nice to include, 
but due to time, style, or coherence may be left out. 
The process of identifying candidate frameworks begins 
with the kernel. 
The implication of the discussion describing the char- 
acteristics of basic blocks suggests that a unifying 
framework will be some feature to which all of the ker- 
nel concepts can be related. In our model, we assume 
that the domain knowledge is maintained as a hierar- 
chical network. Such a representation suggests that a 
node to which all of these concepts converge could serve 
116 
as the unifying framework. The basic block foci, which 
represent various aspects of this framework, would then 
correspond to children of this node. 
Given a set of propositions to be conveyed, one is 
randomly selected from the kernel and a trace is per- 
formed upwards from each of its arguments through 
the hierarchy. The traversal of the hierarchy is per- 
formed using generalization links, e.g. ISA, ISPART, 
and SUBGOAL, incrementing counters associated with 
each node that is traversed. Consider, for example, the 
two propositions (agent open1 Fred1) and (object open1 
valve1). Using ISA links, we might expect a traver- 
sal starting from the action open1 to pass through the 
nodes opening-event, general-event, etc. Furthermore, 
openl may have been a substep of a higher-level plan 
which would also be encountered while performing a 
traversal using SUBGOAL links. By the same token, 
we could expect a traversal starting at Fred1 which uses 
ISA links to encounter person, animal, etc., while one 
originating from valve1 might pass through the nodes 
valve, mechanical-device, etc. Since valve1 is also part 
of secondary.system1, a traversal through secondary- 
system1 is initiated in the same manner. 
Once the traversal has been completed for each ele- 
ment of the kernel, the nodes are ranked according to 
their counters; this will give some idea of the number 
of concepts that converge on each. Only those nodes 
on which a majority of the kernel concepts converge 
will be further considered as possible candidates for the 
unifying framework. To continue our example, suppose 
that, given a kernel of sixty propositions, our traversal 
resulted in a convergence of fifty of those propositions 
on the person node, ten on mechanical-device, fifteen on 
event, and even fewer on the remaining nodes. Given 
that we are only interested in nodes at which the major- 
ity of propositions converge, only person will be given 
further consideration. 
In all likelihood, none of the selected candidates will 
account for all of the kernel concepts. However, it may 
be possible to link a candidate to concepts not directly 
associated with it by finding ties between these concepts 
and ones for which the candidate already accounts, pos- 
sibly via a property they have in common. With respect 
to our example, assume that (object open1 valve1) does 
not converge on person. However, because (agent open1 
Fred1) does, and each of these propositions concerns the 
event open1, the first proposition can be effectively in- 
corporated into this framework. Additionally, it may be 
possible to include such material in a final summary or 
in a "catchall" block. That speakers do introduce such 
material in a final summary is evidenced by the sample 
final summary contained in the appendix, in which all 
of the "significant factors" addressed in the summary 
had been discussed in the main body of the text except 
for the fourth. We claim that inclusion in a final sum- 
mary is warranted if the speaker intended to include a 
final summary initially, and if the amount of material 
is of the same order as that comprising the individual 
segments of the summary. The transcripts also supply 
evidence for speakers' use of catchall basic blocks at the 
end of a discourse. We claim that for such an option 
to be viable, there must be sufficient material to con- 
struct an adequate basic block and that the material 
must form a cohesive unit. s 
SELECTION OF THE UNIFYING 
FRAMEWORK 
Once potential frameworks have been identified, each 
must be evaluated according to how well it meets the 
criteria described above: 
• How thoroughly does a candidate account for the se- 
lected material? 
• How uniformly does a candidate distribute the con- 
cepts among the resulting basic blocks? 
• How closely do the generated blocks conform to the 
ideal size? 
Additionally, a candidate may be evaluated as to how 
well it meets the needs of a user model. 
The actual blocks are constructed around a candi- 
date's immediate descendents, not the candidate itself. 
So, for example, if a candidate were the node person, 
its children - actual people involved in the events - will 
become the basic block foci. The ideal case is one in 
which each child of the candidate accounts for approx- 
imately the same number of concepts. The balance 
each candidate achieves can be determined by compar- 
ing the counters of its children. At the same time, 
the candidate will be rated according to how closely 
its blocks conform to the ideal size. Lastly, we can 
rate the candidates on how well they meet the de- 
mands, if any, imposed by a user model. For exam- 
ple, if we know the user is familiar with the location in 
which the events occurred, basic blocks based on loca- 
tion should be given higher ratings; organizing an ex- 
planation around a framework with which the listener 
is familiar will facilitate his assimilation of the informa- 
tion. 
Once this process has been accomplished, the candi- 
date with the highest rating will be selected as the best 
overall framework. At this point, the material has been 
effectively partitioned about the resulting basic blocks; 
the material within each individual block can then be 
organized using a top-down methodology (e.g., RST or 
schemas), and then realized into text. 
CONCLUSIONS 
This paper has identified several characteristics of 
naturally-occurring explanations that cannot be ac- 
counted for by recursive text organization strategies. 
While recursive methods such as RST and schemas are 
SSee \[Mooney, et al. 1989\] for a more complete 
discussion. 
117 
adequate for local organization of text, they are insuffi- 
cient for the determination of high-level structure, pro- 
viding neither criteria which describe what constitutes 
an acceptable framework for a discourse nor any clues as 
to how this framework should be constructed. We have 
presented a computational strategy for generating the 
high-level structure for extended explanations that ad- 
dresses both issues. Our strategy is based on the basic 
block, a discourse unit consisting of an organizational 
focus and text structured about that focus. We con- 
tend that basic block structure cannot be adequately 
generated by recursive structur{ng methods, but must 
be identified by bottom-up strategies driven by the in- 
formation to be conveyed. Support for our contentions 
is provided by repetition, a device frequently used in 
extended explanations for a variety of purposes. 
Our anMysis of high-level structure suggests a strat- 
egy for generating extended explanations which we are 
currently implementing as part of a larger system de- 
signed to generate extended explanations in an interac- 
tive environment. The system is intended as a testbed 
for identifying viable organizational strategies and re- 
covery techniques due to user interruptions within the 
basic block paradigm. We believe that our analysis 
of basic block structure provides criteria necessary for 
bridging the gap between the generation of short ex- 
planations and the successful generation of extended 
explanations by explicating how the framework of an 
extended explanation can be constructed and how rep- 
etition can be cohesively incorporated into it. 

References 
\[Arena, 1975\] L. A. Arena. Linguistics and Composi- 
tion. Georgetown University Press, 1975. 
\[Daniel, 1967\] R. W. Daniel. A Contemporary 
Rhetoric. Little, Brown and Company, 1967. 
\[Grosz and Sidner, 1986\] B. J. Grosz and C. L. Sidner. 
Attention, Intentions, and the Structure of Discourse. 
Computational Linguistics, 12, 1986. 
\[Hovy, 1988a\] E. Hovy. Approaches to the Planning of 
Coherent Text. Fourth International Workshop on 
Natural Language Generation, 1988. 
\[Hovy, 1988b\] E. H. Hovy. Planning Coherent Multi- 
sentential Text. In: Twentysixth Annual Meeting of 
the Association for Computational Linguistics, ACL, 
1988. 
\[Hovy and McCoy, 1989\] E. Hovy and K. McCoy. Fo- 
cusing Your RST: A Step Toward Generating Co- 
herent Multisentential Text. In: Proceedings of the 
Eleventh Annual Meeting of the Cognitive Science 
Society, Cognitive Science Society, 1989. 
\[Kane and Peters, 1966\] T. S. Kane and L. J. Peters. 
A Practical Rhetoric of Expository Prose. Oxford 
University Press, 1966. 
\[Mann and Thompson, 1988\] W. C. Mann and S. A. 
Thompson. Rhetorical Structure Theory: Toward 
a Functional Theory of Text Organization. Text, 8, 
1988. 
\[McCoy, 1985\] K. F. McCoy. Correcting Object-Related 
Misconceptions. PhD thesis, University of Pennsyl- 
vania, 1985. 
\[McKeown, 1985\] K. McKeown. Text Generation. 
Cambridge University Press, 1985. 
\[Mooney, et al. 1989\] D. J. Mooney, S. Carberry, and 
K. F. McCoy. The Identification of a Unifying Frame- 
work for the Organization of Extended, Interactive 
Explanations. Technical Report 90-1, University of 
Delaware, 1989. 
\[Moore and Paris, 1988a\] J. Moore and C. Paris. Con- 
structing Coherent Text using Rhetorical Relations. 
In: Proceedings of the Tenth Annual Meeting of the 
Cognitive Science Society, Cognitive Science Society, 
1988. 
\[Moore and Swartout, 1988b\] J. D. Moore and W. R. 
Swartout. A Reactive Approach to Explanation. 
Fourth International Workshop on Natural Language 
Generation, 1988. 
\[Paris, 1987\] C. Paris. The Use of Explicit User Models 
in Text Generation: Tailoring to a User's Level of 
Expertise. PhD thesis, Columbia University, 1987. 
\[Polanyi, 1986\] L. Polanyi. The Linguistic Discourse 
Model: Towards a Formal Theory of Discourse Struc- 
ture. Technical Report 6409, BBN, 1986. 
\[Reichman, 1978\] R. Reichman. Conversational Co- 
herency. Cognitive Science, 2, 1978. 
\[Representatives, 1979\] U.S. House of Representatives, 
Ninetysixth Congress: first session. Accident at the 
Three Mile Island Nuclear Powerplant: Oversight 
Hearings before a Task Force of the Subcommittee 
on Energy and the Environment of the Committee on 
Interior and Insular Affairs. U.S. Government Print- 
ing Office, 1979. 
\[Representatives, 1980\] U.S. House of Representatives, 
Ninetysixth Congress: second session. Nuclear Pow- 
erplant Safety after TMI: Oversight Hearings before 
a Task Force of the Subcommittee on Energy and the 
Environment of the Committee on Interior and Insu- 
lar Affairs. U.S. Government Printing Office, 1980. 
\[Thompson, 1957\] W. N. Thompson. Fundamentals of 
Communication. McGraw Hill, 1957. 
\[Wicker and Albrecht, 1960\] C. V. Wicker and W. P. 
Albrecht. The American Technical Writer. American 
Book Company, 1960. 
