<?xml version="1.0" standalone="yes"?>
<Paper uid="E91-1021">
  <Title>USING PLAUSIBLE INFERENCE RULES IN DESCRIPTION PLANNING</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
OUTLINE OF
'PLANNER'
EXPLANATION
</SectionTitle>
    <Paragraph position="0"> The system described below 1 aims to show how plausible inference rules may be used to guide explanation planning given different communicative goals. The basic approach is to find some set of possible utterances, and select the one which - assuming that the user makes certain plausible inferences - contributes most to the stated communicative goal. This process is repeated until some terminating condition is met, such as the communicative goal being satisfied.</Paragraph>
    <Paragraph position="1"> This explanation 'planning' strategy is a kind of heuristic search, using a modified best-first search strategy. The search space consists of the space of all possible utterance sequences, and the heuristic scoring function assesses how far each utterance would contribute to the communicative goal.</Paragraph>
    <Paragraph position="2"> Because this gives a potentially very large search space, only certain utterances are considered at each point. Currently these are constrained to be those which appear to make some contribution to the communicative goal - for example, the system might consider describing an object as an instance of some class if that class had some attributes which contributed to the target state. These possible utterances are then scored by using the plausible inference rules to predict what might reasonably be inferred by the user from this statement, given his current knowledge, and comparing that with the communicative goal.</Paragraph>
    <Paragraph position="3"> For example, if the communicative goal is for the user to have a positive impression of the object, and the system knows of some feature which the user believes is desirable in an object, then the system may select utterances which allow the user to plausibly infer this feature given their current assumed knowledge about this and other objects.</Paragraph>
    <Paragraph position="4"> The search space is defined by the range of possible utterance types. Currently the following types (and associated plausible inference procedures) are allowed, where there may be many possible statements about a given object of each type: IReferred to from now on as the GIBBER system - Gen- null Identification, as an instance (or sub-class) of some class.</Paragraph>
    <Paragraph position="5"> Similarity, given some related object with many shared attributes 2.</Paragraph>
    <Paragraph position="6"> Examples, of instances or sub-classes, Attributes of that object.</Paragraph>
    <Paragraph position="7"> selection of possible utterances, and their scoring \[given the probable inferences which might be made) depends on the communicative goal set. In the current system, given some object to describe, two different types of communicative goal may be set. The system may either be given an explicit set of attribute values which should be inferrable from the generated description, or it can be given a 'property' that the inferrable attributes should have. This property can be, for example, that the user believes the attribute value to be a !desirable one, where an 'evaluation form' similar to Jameson's (1983) is used to rate different values. Where a set of attribute values are given these Can be either specific values, or value ranges.</Paragraph>
    <Paragraph position="8"> This approach uses a set of rules which may be used to propose a possible move/statement (given the target/communicative goal), a set of rules which may be used to guess what would be inferred or learned from that statement, given the assumed current state of the user's knowledge, and a scoring function which assesses how far the 'guessed at' inferences would contribute to the target. Statements are generated one at a time, with currently 3 the only relation between the utterances being enforced by the common overall communicative goal and by the fact that the statements are selected to incrementally update the user's model of the object described.</Paragraph>
    <Paragraph position="9"> Using plausible inference rules in this way is undoubtedly error-prone, as assumptions about the user may be wrong and not all hearers will make the expected inferences. However, it is certainly better than ignoring these inferences entirely. So long as the user can ask follow-up questions in an explanatory dialogue (e.g., Cawsey, 1989; Moore, 1990) any such errors are not crucial.</Paragraph>
    <Paragraph position="10"> ~Note that full analogies, where a complex mapping is required between two conceptually distinct objects, are currently not possible in the system.</Paragraph>
    <Paragraph position="11"> SAdding further coherences relations and global strategies may be the subject of further work.</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
INFERENCE RULES AND
KNOWLEDGE
REPRESENTATION
</SectionTitle>
    <Paragraph position="0"> For this approach to text planning to be effective, the rules used for guessing what the reader might infer should correspond as far as possible to human plausible inference rules. There are a relatively small number of AI systems which attempt to model human plausible inferences {compared with those attempting to model efficient learning strategies in artificial situations). Zuckerman (1990) uses some simple plausible inference rules in her explanation system, in order to attempt to block incorrect plausible inferences, while a more comprehensive model of human plausible reasoning is provided by Collins and Michalski (1989). This latter theory is concerned with how people make plausible inferences given generalisation, specia|isation, similarity and dissimilarity relations between objects, using a large number of certainty parameters to influence the inferences. The theory assumes a representation of human memory based on dynamic hierarchies, where, for example, given the statement colour(eyes(John))fblue then colour, eyes, John and blue would all be objects in some hierarchy. The theory is used to account for the plausible inferences made when people guess the answer to questions given uncertain knowledge.</Paragraph>
    <Paragraph position="1"> The GIBBER system uses inference rules somewhat differently to Collins' and Michalski's.</Paragraph>
    <Paragraph position="2"> Whereas they are concerned with the competing inferences which may be made from existing knowledge to answer a single question, the GIBBER system is concerned with mutually supporting inferences from multiple given relationships in order to build up a picture of an object. So, although the basic knowledge representation and relationship types (apart from dissimilarity) are borrowed from their work, the actual inference rules used are slightly different.</Paragraph>
    <Paragraph position="3"> It should be possible to use the inference rules to incrementally update a representation of what is currently known about an attribute, where generalisation, similarity and specialisation relationships may all contribute to the final 'conclusion'. In order to allow such incremental updates, the representation used in Mitchell's version space learning algorithm is adopted (1977), where each attribute has a pointer to the most specific value that attribute could take, and to the most gen- null eral value, given current evidence. Positive examples (or Oeneralisation relationships) are used to generallse the specific value (as in Mitchell's algorithm) 4 while class identification (specialisation) is used to update the general value using the inherited attributes. Similarity transforms are done by first finding a common context for the transform (a common parent object), and then transferring those attributes which belong to that * context which are not ruled out by current evidence. Explicit statement of attribute values fix the attribute value, but further evidence may be used to increase the certainty of any value.</Paragraph>
    <Paragraph position="4"> The system also allows for other kinds of domain specific inference rules to be defined - for example, if a user has just been told that a bike has derailleur gears, a rule may be used to show that the user could probably guess that the bike had between 5 and 21 gears. The different kinds of inference rules are used to incrementally update the representation of the user's assumed knowledge of the object and the scoring function, discussed in the previous section, will compare that assumed knowledge of the object with the target.</Paragraph>
    <Paragraph position="5"> The knowledge representation is based on a frame hierarchy describing the objects in the domain, where the slot values may point to other objects, also in some hierarchy. In figure 4 a small section of a knowledge base of different kinds of bicycle is illustrated, along with some simple hierarchies of attribute values. In the GIBBER system separate hierarchies are defined for the system's and for the user's assumed knowledge, where the latter is initialised from a user stereotype and updated following each query and explanation.</Paragraph>
    <Paragraph position="6"> Of course, the knowledge representation and inference rules described in this section are by no means definitive - there is no implied claim that people really use these rules rather than others in learning from descriptions. They simply provide a starting point for exploring how explanation generation may take into account possible learning and inference rules, and thus better select statements in a description given knowledge of the domain and of</Paragraph>
  </Section>
  <Section position="6" start_page="0" end_page="0" type="metho">
    <SectionTitle>
EXAMPLE DESCRIPTIONS
</SectionTitle>
    <Paragraph position="0"> This section will give two examples of how descriptions of bicycles may be generated using this approach. We will assume that the system's knowledge includes the hierarchy given in figure 4, and (for simplification) the user's knowledge includes all the items except the 'Cascade', but includes the fact that Alison's bike has shimano indexed gears.</Paragraph>
    <Paragraph position="1"> The first example will show how the system will select utterances to economically convey information given some target attribute values, while the second will show how biased descriptions may be generated given a specification of the desired prop-erty of inferrable attributes.</Paragraph>
    <Paragraph position="2"> Suppose the user requests a description of the Cascade and that the communicative goal set by the system (by some other process) is to convey the following attributes: 4Note that Collins' and Michalski's theory does not appear to allow multiple examples to be used by generalising the inferred values.</Paragraph>
    <Paragraph position="4"> There are many possible statements which could be made about the Cascade. The user knows Alison's bike, so this example could be mentioned. It could be described as an instance of a mountain bike, or just as a bicycle; a comparison could be made with the Trek-800; or any one of the bikes attributes could be mentioned. In this case if it is identified as an instance of a mountain bike the system guesses that the user could infer the first two attributes, which gives the highest score given the target s. A comparison with the Trek-800 also gives two possible inferrable attributes, {though one incorrect value, which is currently allowed}, and this is the next choice. Finally the system informs the user of the number of gears, blocking the incorrect inference in the previous utterance. The resulting short description is the followingS: aThe Cascade is a kind of mountain bike.</Paragraph>
    <Paragraph position="5"> It is a bit like the Trek-800.</Paragraph>
    <Paragraph position="6"> It has 18 gears.&amp;quot; If the scoring function is changed so that it is biased further towards highly certain inferences, rather than efficient presentation of information, then given the same communicative goal the description may end up as an explicit list of all the attributes of the bike, or in a less extreme case, a class identification and three explicit attributes. This scoring function therefore allows for further variation in descriptions, given a communicative goal, and different scoring functions should be used depending on the type of description required.</Paragraph>
    <Paragraph position="7"> Suppose now that the same bike is to be described, but the communicative goal is that the user has a positive impression of the Cascade. If the user regards it to be good for a bike to be black with 21 * shimano index gears then the following description will be generated.</Paragraph>
    <Paragraph position="8">  information with the target, preferring more certain inferences, and inferences bring the knowledge of the object closer to the target (given the attribute value hierarchy}. For example, an inference that the bike had 18-21 gears~ or an uncertain inference that it had 18, would be given a lower score than a certain inference that it had 18 gears. The total score is the sum of the scores of each possibly inferred value.</Paragraph>
    <Paragraph position="9"> eOf course this description would be more coherent if a higher level cornpare-contra~t relation was used to generate the last two inferences, with resulting text: Ult is a bit like the Trek-800 but has 18 gears.&amp;quot;. Allowing these higer level strategies within an inference-based approach is the subject of further work.</Paragraph>
    <Paragraph position="10"> aThe Cascade is a bit like the Trek-800.</Paragraph>
    <Paragraph position="11"> Alison's bike is a Cascade.</Paragraph>
    <Paragraph position="12"> The Cascade has Shimano Index Gears. ~ Here the system evaluates each statement by comparing the plausible inferences against an evaluation form {Jameson, 1983). The evaluation form describes how far different attribute values are appreciated by different classes of user. Instead of comparing inferred values with some target attribute values the scoring function will score each against the evaluation form. For example, the first utterance (comparison with the Trek-800) is selected because the attributes which might be plausibly inferred from this statement by this user are rated highly on the evaluation form for that class of user. In this case the system assumes that this type of user will prefer a bike with a large number of indexed gears. Of course, one of the plausible inferences which can be made will be incorrect (the fact the Cascade has 21 gears). The system is not required to block such false inferences if they contribute to its goals {though the ethics of generating such leading descriptions might be doubted!).</Paragraph>
    <Paragraph position="13"> It should be clear from this that the descriptions generated by the system are very sensitive to the assumptions about the user's prior knowledge, and the inference rules and the scoring function used, as well as to the communicative goal set. There is much possibility for error (and further research required) in each of these. However, the approach still seems to provide the potential for generating improveddescriptions, and provides a new principled way of making choices in a description which is absent, in schema-based (and RST-based) approaches. It gives a simple example of how, given a model of how people update their beliefs, utterances may be strategically generated to change those beliefs.</Paragraph>
  </Section>
class="xml-element"></Paper>