<?xml version="1.0" standalone="yes"?>
<Paper uid="P84-1055">
  <Title>Appendix I: Coherent Transmission Strategies</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
A COMFUTATIONAL THEORY OF THE FUNCTION OF CLUE WORDS
IN ARGUMENT UNDERSTANDING
</SectionTitle>
    <Paragraph position="0"> This paper examines the use of clue words in argument dialogues. These are special words and phrases directly indicating the structure of the argument to the hearer. Two main conclusions are drawn: I) clue words can occur in conjunction with coherent transmissions, to reduce processing of the hearer 2) clue words must occur with more complex forms of transmission, to facilitate recognition of the argument structure. Interpretation rules to process clues are proposed. In addition, a relationship between use of clues and complexity of processing is suggested for the case of exceptional transmission strategies.</Paragraph>
  </Section>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
! Overview
</SectionTitle>
    <Paragraph position="0"> In argt~nent dialogues, one often encounters words which serve to indicate overall structure - phrases that link individual propositions to form one coherent presentation. Other researchers in language understanding have acknowledged the existence of these &amp;quot;clue words&amp;quot;. Birnbat~n \[Birnbaum 823 states that in order to recognize argument structures it would be useful to identify typical signals of each form.</Paragraph>
    <Paragraph position="1"> In \[Cohen 83\] we develop a computational model for argument analysis. The setting is a dialogue where the speaker tries to convince the hearer of a particular point of view; as a first step, the hearer tries to construct a representation for the structure of the arg~ent, indicating the underlying claim and evidence relations between propositions. Within this framework, a theory of linguistic clues is developed whlch categorizes the function of different phrases, presenting interpretation rules.</Paragraph>
    <Paragraph position="2"> What we have done is develop a model for argument analysis which is sufficiently well-defined in terms of algorithms, with measurable complexity, to allow convenient study of the effect of clue words on processing. Two important observations are made: (I) clue words cut processing of the hearer in recognizing coherent transmissions (2) clue words are used to allow the recognition of transmissions which would be incoherent (too complex to reconstruct) in the absence of clues.</Paragraph>
    <Paragraph position="3"> Considering arguments as goal-oriented dialogues, the use of clue words by the speaker can be construed as attempts to facilitate the heater's plan reconstruction process. Thus, there exist words and even entire statements with the sole function of indicating structure (vs. content) in the argument. The importance of structure to argument understanding is first of all a by-product of our imposed pragmatic approach to analysis. To understand the argument intended by the speaker, the hearer must determine, for each proposition uttered, both where it fits with respect to the dialogue so far and how, in particular, it relates to some prior statement. In addition, it is precisely the expected form of arguments which can be used to control the analysis (since content can't be stereotyped as in the case of stories).</Paragraph>
    <Paragraph position="4"> It is this importance of form which necessitates clue words and presents the research problem of specifiying their function precisely.</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="252" type="metho">
    <SectionTitle>
II Background
</SectionTitle>
    <Paragraph position="0"> To understand the role of clue words in facilitating analysis, some detail on the overall argument understanding model is required. (For further reference, see \[Cohen 80\], \[Cohen 81\], \[Cohen 83\]). Each proposition of the argument is analyzed, in turn, with respect to the argument so far. A proposition is interpreted by determining the claim and evidence relations it shares with the rest of the argument's propositions. Leaving the verification of evidence to an oracle, the main analysis task is determining where a current proposition fits.</Paragraph>
    <Paragraph position="1"> To understand the examples introduced in this paper, it is useful to present the starting definition of evidence, as used in the model. A proposition P is evidence for a proposition O if there is some rule of inference such that P is premise to Q's conclusion. The rule most often observed is modus ponens, with missing major premise - i.e. P, Q are given and one must fill P --&gt; Q to recognize the support intended from P to Q. More detail on the definition of evidence is presented in \[Cohen 83\].</Paragraph>
    <Paragraph position="2"> Determining an interpretation for a proposition is restricted to a computationally reasonable task by characterizing possible coherent transmission  strategies on the part of the speaker and reducing analysis to a recognition of these forms. These algorithms are outlined in detail in \[Cohen 83\]. The basic restrictions yield e limited set of propositions to search. The representation is a tree of claim and evidence relations where evidence are sons to the father claim. Essentially, the last proposition eligible to relate to the current is tracked (called LAST). LAST and its ancestors in the tree are all eligible relatives and each is tested in turn, to set the interpretation of the current p~oposition. The analysis algorithm is termed &amp;quot;hybrid reception&amp;quot; because it is designed to recognize transmission strategies where each constituent sub-argument is presented either claim first or claim last. Complexity analysis of this algorithm indicates that it works in linear time (i.e. it takes a linear factor of the number of nodes of the tree to locate all propositions in tile representation).</Paragraph>
    <Paragraph position="3"> A sample tree and the processing required for the current proposition is illustrated below:</Paragraph>
    <Paragraph position="5"> With the initial argument above, a new proposition (8) would be checked to be evidence for 7, 6, 5 and I in turn. If these tests fail, it is then attached as a son to the dummy root (expecting a father in upcoming propositions). The final tree above, for example, may result if the next proposition (9) is processed and succeeds as father to 8. Note that in processing 8 initially, 4, 3, and 2 were not eligible relatives. This is because an earlier brother to a subsequent proposition is closed off from consideration according to the specifications of the hybrid algorithm. See Appendix I for a detailed description of possible coherent transmission strategies and their &amp;quot;reception&amp;quot; algorithms.</Paragraph>
    <Paragraph position="6"> III Clues to reduce processing (Helpfulness) With coherent transmissions characterized, the role of clue words can be investigated more closely. Note first that the restrictions of the analysis algorithms are such that the proposition to which the current one relates is not always the immedimte prior proposition. In fact, sometimes the claim is located far back in the dialogue.</Paragraph>
    <Paragraph position="7"> Consider the following example: EXI: 1)The city is a mess  Here, the representation for the following tree:  argument is the The last proposition, b, is evidence for I, one of the claims higher up in the tree. Many arguments which re-address earlier claims assist the hearer by specifically including a clue of re-direction as in EX2 below.</Paragraph>
    <Paragraph position="8"> EX2: 1)The city is a mess  system needs revamping Here, the search up the right border of the tree (from 5, 3, 2 to I) for a possible claim to the current proposition b is cut short and the correct father (I) indicated directly. One can hypothesize a general reduction on processing complexity from linear to real-time, if clues are consistently used by the speaker to re-direct the hearer with chains that are sufficiently long.</Paragraph>
    <Paragraph position="9"> Connectives are another type of clue word, used extensively. Hobbs (\[Hobbs 76J) attempts a characterization with respect to his coherence relations for a couple of words. Reichman (\[Reichman 81\]) associates certain expressions with particular conversational moves, but there is no unified attempt at classification. We develop a taxonomy so that clues of the same semantic function are grouped to assign one interpretation rule for the dominated proposition within the claim and evidence framework. Consider the following example:</Paragraph>
    <Paragraph position="11"> The connective in 4, &amp;quot;as a result&amp;quot;, suggests that some prior proposition connects to 4 and that this proposition acts as evidence for 4. 'lhe relation of the prior proposition is set out b.elow according the the interpretation rule for the category that &amp;quot;as a result&amp;quot; belongs to in the taxonomy. The particular evidence connection advocated here is of the form: &amp;quot;If our city needs help, then we will ask for federal aid&amp;quot;. \[Note: Whether I is evidence for 4 is tested by trying a modus ponens major premise of the form: &amp;quot;(For all cities) if a  city needs help, then it can ask for federal aid&amp;quot;, and then using &amp;quot;our city&amp;quot; as the specific case\]. The taxonomy (drawn from \[Quirk 72\]) is intended to cover the class of connectives and presents default interpretation rules.</Paragraph>
    <Paragraph position="12"> (P indicates prior proposition; S has the clue)</Paragraph>
  </Section>
  <Section position="4" start_page="252" end_page="253" type="metho">
    <SectionTitle>
CATEGORY RELATION:P to S EXAMPLE
</SectionTitle>
    <Paragraph position="0"> parallel brother in addition detail father in particular inference son as a result summary multiple sons in SL~n reformulation father and son in other words contrast father or brother conversely Note that the classification of connectives provides a reduction in processing for the hearer. For example, in EX3 with a casual connective, the analysis for the proposition 4 is restricted to a search for a son. In short, connective interpretation rules help specify the type of relation between propositions; re-direction clues help determine which prior proposition is related to the current one. All together, clue words function to reduce overall processing operations. See Appendix II for more examples of relations of the taxonomy.</Paragraph>
    <Paragraph position="1"> IV Clues to support complex transmissions (Necessity) C%ue words also exist in conjunction with transmissions which violate the constraints of the hybrid model of expected coherent structure. The claim is that clues provide a necessary reduction in complexity, to enable the hearer to recognize the intended structure. Consider the following examples: EX4: 1)The city is a mess  the sandboxes are a mess 6)As for the highways, the gravel is shot 7)And as for the buildings, the bricks are rotting The initial tree for the argument is as follows: In EX4, the last proposition cannot be interpreted as desired; the probable intended father proposition (2) is not an eligible candidate to relate to the current proposition (5) according to .he hybrid specifications. In EX5, however, a parallel construction is specifically indicated through clue words, so that the connections can be recognized by the hearer and the appropriate representation constructed as below: 11C--.. 5.2 6/3 7/4 It now becomes important to provide a framework for accommodating &amp;quot;extended&amp;quot; transmission strategies in the model. First, the complexity of processing without clues is a good measure for determining whether a strategy should be considered exceptional. Then, to be acceptable in the model the proposed transmission must have some characterizable algorithm - i.e. still reflect a coherent plan of the speaker. Further, exceptional tranmsission strategies must be clearly marked by the speaker, using clues, in cases where the transmission can be assigned an alternate reading according to the basic processing strategy. The hearer should be expected to expend the minimum computational effort, so that the onus is on the speaker to make exceptional readings explicit. In brief, we propose developing a &amp;quot;clue interpetation module&amp;quot; for the analysis model, which would be called by the basic proposition analyzer to handle extended transmissions in the presence of clues. Then, complexity of processing should be used as s guide for determining the preferred analysis.</Paragraph>
    <Paragraph position="2"> To illustrate, consider another acceptable extended transmission strategy - mixed-mode sub-arguments, where evidence both precedes and follows a claim.</Paragraph>
    <Paragraph position="3"> EXd: l)The grass is rotting</Paragraph>
    <Paragraph position="5"> Here, it is preferable to keep I and 2 as evidence for 3, because this requires less computational effort than the re-attachment of sons which takes place to construct the other possible representation. In other words, computational effort is a good guide for the specification of processing strategies.</Paragraph>
    <Paragraph position="6"> Finally, it is worth noting that the specific clue word used may influence the processing for these extended transmissions. In EXd, if the last proposition (4) was introduced by the clue word &amp;quot;in addition&amp;quot;, then the alternate tree would not be an eligible reading. This is because &amp;quot;in addition&amp;quot; forces 4 to find a brother among the earlier propositions, according to the interpretation rule for the &amp;quot;parallel&amp;quot; class of the taxonomy of  connectives.</Paragraph>
    <Paragraph position="7"> In sum, we propose particular extended transmission strategies for the model, including (i) parallel (ii) mixed-mode (iii) multiple relations. \[Note: More discussion of (iii) is in \[Cohen 33\]. We consider as an acceptable exceptional strategy the case where one proposition acts as evidence for an entire set of claims following it immediately in the stream. Other configurations of multiple relations seem to present additional processing problems\]. We demand clue words to facilitate the analysis and we begin to suggest how to accommodate uses of these exceptional cases in the overall analysis model.</Paragraph>
  </Section>
  <Section position="5" start_page="253" end_page="254" type="metho">
    <SectionTitle>
V Related Topics
</SectionTitle>
    <Paragraph position="0"> A. Nature of clues The exact specification of a clue is a topic for further research. Since it is hypothesized that clues are necessary to admit exceptional transmissions, what constitutes a clue is a key issue. Within Quirk's classification of connectives (\[Quirk 72\]) both special words and connecting phrases (&amp;quot;integrated markers&amp;quot;) are possible. For example, one may say &amp;quot;in conclusion&amp;quot; or &amp;quot;I will conc\].ude by saying&amp;quot;.</Paragraph>
    <Paragraph position="1"> Quirk also discusses several mechanisms for indicating connectives which need to be examined more closely as candidates for clue words. These comstructions are all &amp;quot;indirect&amp;quot; indications. a) lexical equivalence: This includes the case where synonyms are used to suggest a connection to a previous clause. For example: &amp;quot;The monkey learned to use a tractor. By age 9, he could work solo on the vehicle.&amp;quot; In searching for evidence relations, the hearer may faciltate his analysis by recognizing this type of connective device. But it unclear that the construction should be considered an additional &amp;quot;clue&amp;quot;.</Paragraph>
    <Paragraph position="2"> b) substitution, reference, comparison, ellipsis: Here, the &amp;quot;abbreviated&amp;quot; nature of the constructions may be significant enough to provide an extra signal to the hearer. For now, we do not consider these devices as clues, but examine the relations between the use of anaphors and clues in the next section.</Paragraph>
    <Paragraph position="3"> Even w!thin the classification of connectives, there is a question of level of explicitness of the clues. Consider the example: EX7: 1)The city is dangerous  c) have a phrase indicating a causal connection, but c) requires a kind of referent resolution as well; d) requires recognizing &amp;quot;the problem&amp;quot; as an indication of cause. The problem addressed in this example is similar to the one faced by Allen (\[Allen 79\]): handling a variety of surface forms which all convey the same intention. In our case, the &amp;quot;intention&amp;quot; is that one proposition act as evidence for another.</Paragraph>
    <Paragraph position="4"> Finally, there are different kinds of special phrases used to influence the credibilty of the hearer: I) attitudinal expressions reflecting the speaker's beliefs and 2) expressions of emphasis. Since our model focuses on the first step in processing of recognizing structural connections, these clues have not be examined more closely.</Paragraph>
    <Paragraph position="5"> However, examples of these expressions are listed in Appendix III, along with phrases indicating structure.</Paragraph>
    <Paragraph position="6"> B. Relation to reference resolution and focus There are some important similarities between our approach to reconstructing argument structure and the problem of representing focus for referent resolution addressed in \[Sidner 79\] and \[Grosz 77J. For both tasks, a particular kind of semantic re\]ation between parts of a dialogue must be found and verified. In both cases, a hierarchical representation is constructed to hold structural information and is searched in some restricted fashion.</Paragraph>
    <Paragraph position="7"> Orosz's hierarchical model of focus spaces, with visibility constraints imposed by the task domain, is maintained in a fashion similar to our tree model. Information on which of the focus spaces is &amp;quot;active&amp;quot; and which are &amp;quot;open&amp;quot; (possible to shift to) is kept; open spaces are determined by the active space and the visibilty constraints.</Paragraph>
    <Paragraph position="8"> Analysis for a problem such as resolving definite noun phrase referents can be limited by choosing only those items &amp;quot;in focus&amp;quot;.</Paragraph>
    <Paragraph position="9"> In \[Sidner 79\] focus is introduced to determine eligible candidates for a co-specification. But the ultimate choice rests with verification by the hearer, using inferencing, that the focus element relates to the anaphor. This is parallel to our approach of narrowing the search for a proposition's intepretation, but requiring testing of possible relations in order to establish the desired link. To set the focus, Sidner suggests either: I) using special words to signal the hearer or 2) relying on shared knowledge to establish an unstated connection. This is analogous to our cases of processing with and without clues.</Paragraph>
    <Paragraph position="10"> In Sidner's theory there is also a clear distinction between returning to an element previously in focus (one from the focus stack) or choosing a completely &amp;quot;new topic&amp;quot; from prior elements (using the alternate focus list). We distinguish returning to some ancestor of the last proposition (a choice of eligible proposition) from the case of re-addressing a &amp;quot;closed&amp;quot; proposition.  In this latter case, we require a clue word to re-direct. What we have tried to do is clearly separate eligible relatives from exceptional cases and connect the required use of clues to the exceptional category. Grosz and Sidner both allow &amp;quot;focus shifts&amp;quot; and Sidner explicitly discusses uses of &amp;quot;special phrases&amp;quot;, but we have tried to study the connections between clues and exceptions more closely.</Paragraph>
    <Paragraph position="11"> Finally, it is worth noting that the problem of reference resolution is similar to that of evidence determination, but still distinct. In the example below, constraints suggested by referent resolution theories should not be violated by our restricted processing suggestions: Exa: 1)The city is a mess  In 4, &amp;quot;it&amp;quot; is resolved as referring to &amp;quot;the highway&amp;quot; in 3; this proposition is eligible and the closer connection is preferred.</Paragraph>
    <Paragraph position="12"> But clue interpretation is not equivalent to referent resolution. The clue &amp;quot;for example&amp;quot; may be expressed as &amp;quot;one example for this is&amp;quot; but could also be presented as &amp;quot;one example for this problem is&amp;quot;. Since the search for a referent may differ according to the surface form (\[Sidner 79\]) there is no clear mapping from processing propositions with clues to those with referents. For our model, surface form may vary widely, but the search is restricted according to interpretation rules for a taxonomy - according to the semantics of the clue and the solution is dictated by the structure of the argument so far.</Paragraph>
    <Paragraph position="13"> C. Necessity in the base case The main points raised in this paper are that clues can be used with a basic transmission strategy to cut processing and must be used in more complex transmissions. The question of whether certain basic transmissions still require clues is worth investigating further. In particular, it has been suggested (personal communication with psychologists) that deep stacks require clues to remind the hearer, due to &amp;quot;space&amp;quot; limitations. It may be productive to examine the computational properties of this situation more closely.</Paragraph>
    <Paragraph position="14"> Further, clues are often used to delineate sub-arguments when shifting topics. Again, some memory limitations for the hearer may be in effect here.</Paragraph>
  </Section>
class="xml-element"></Paper>