Abstracts of HCRC Research papers, 1994

Enric Vallduví and Ron Zacharski
Accenting phenomena, association with focus, and the recursiveness of focus-ground
Recent work in formal semantics argues that the interpretation of a number of logico-semantic operators is crucially defined in function of the traditional focus-ground partition. For this proposal to hold, one needs to assume that sentences with more than one of these operators contain multiple focus-ground partitions in an overlapping or recursive fashion. This paper shows that such an assumption is unwarranted. A careful analysis of the Eng\-lish facts and a contrastive look at languages that realize focus-ground syntactically, like Catalan, reveals that not all accented constituents are foci in a focus-ground partition but that operators can nevertheless associate with them. The fact that so-called focus-sensitive operators operate on partitions that are not focus-ground invalidates any accounts that crucially define the semantics of these operators in terms of the semantics of focus-ground.
(June 1994; 23 pages)
Ref. No. HCRC/RP-49 Price: UKL 1.10

Andrew John Merrison, Anne H. Anderson and Gwyneth Doherty-Sneddon
An investigation into the communicative abilities of aphasic subjects in task oriented dialogue
This paper reports on research into three mild to moderate non-fluent Broca-type aphasics engaged with a non-aphasic partner in task oriented dialogue. The research aimed to address the relationship between linguistic and communicative skills of aphasic subjects, in particular, how it is that they can sometimes communicate better than their linguistic abilities would initially suggest. The results indicate that at least some aphasic individuals can compensate for their linguistic deficits (a) by effective and sensitive use of non-verbal gesture (producing on average up to eighteen times as much information carrying gesture than do non-aphasics engaged in the same task) and (b) by relying on the communicative expertise of their non-impaired partner.
(June 1994; 25 pages)
Ref. No. HCRC/RP-50 Price: UKL 1.10

Keith Stenning
Logic as a foundation for a cognitive theory of modality assignment
A cognitive theory of modality allocation has to explain the different cognitive consequences of assigning the same information to different modalities of expression---crudely, which picture is worth which thousand words? Most theories of the benefits of graphical representations have focussed on {\it visual} properties of graphics. This talk will sketch a theory which focusses on general logical/computational properties of graphics. The theory is based on the proposition that graphical representations are weakly expressive, in a logical sense, and that this explains graphics' processing advantages. The theory distinctively predicts that graphics will be worse than language when graphics' logical inexpressiveness impedes abstraction that is necessary for a task.
The general theory will be illustrated by an analysis of the use of Euler's Circles to solve syllogisms. Earlier misinterpretations of Euler's system by psychologists illuminate the distinctions the present theory draws between minimally abstractive representation systems (MARS), limited abstraction representation systems (LARS), and unlimited abstraction representation systems (UARS). The analysis reveals model-theoretic properties of this logical fragment which explain why it can be captured by Euler's system. Euler's and Venn's diagrammatic systems are both based on the analogy of spatial containment for set membership. The basis of the difference in their expressive power is revealed by this approach. Euler's graphical algorithm is then generalised and the cognitively fruitful geometrical limitation to circles is related to the logical limitations of the generalised system.
The example analysis of Euler's Circles is then related back to the cognitive theory of graphics and the role that logic plays in this approach to modality allocation.
(June 1994; 15 pages)
Ref. No. HCRC/RP-51 Price: UKL 1.00

E. G. Bard, D. Robertson, and A. Sorace
Magnitude Estimation of Linguistic Acceptability
Judgments of linguistic acceptability constitute an important source of evidence for theoretical and applied linguistics, but are typically elicited and represented in ways which limit their utility. This paper describes how @c[magnitude estimation], a technique used in psychophysics, can be adapted for eliciting acceptability judgments. Magnitude estimation of linguistic acceptability is shown to solve the measurement scale problems which plague conventional techniques; to provide data which makes fine distinctions robustly enough to yield statistically significant results of linguistic interest; to be usable in a consistent way by linguistically naive speaker-hearers, and to allow replication across groups of informants. Methodological pitfalls are discussed and suggestions are offered for new approaches to the nature and measurement of linguistic acceptability.
(July 1994; 42 pages)
Ref. No. HCRC/RP-52 Price: UKL 1.60

Robert Inder and Jussi Stader
Bags and Viewers: a Metaphor for Intelligent Database Access
We present a way of structuring a database query system to form a bridge between current data handling systems and the data requirements of creative work. The interface is based around specifying the contents of ``bags'' of objects and inspecting them using ``viewers'', which can then be used to launch further queries.
The operation of such an interface is described and illustrated by an example session using a prototype system. A number of features of such an interface are presented, the metaphor is discussed and a number of areas for future work are indicated.
(December 1994; 18 pages)
Ref. No. HCRC/RP-53 Price: UKL 1.00

Rosemary J. Stevenson, Rosalind A. Crawley and David Kleinman
Thematic Roles, Focus and the Representation of Events
Two experiments investigated the focusing properties of thematic roles, while a third experiment investigated the view that thematic role preferences reflect a focusing on the consequences of the represented event. Sentence continuation tasks were used in which subjects wrote continuations to sentence fragments containing two antecedents, each occupying a different thematic role. The results of experiments one and two showed a preference for referring to a particular thematic role regardless of the presence or absence of a pronoun at the start of the continuation and regardless of whether the continuation was part of a different sentence from the one containing the antecedents (experiment one) or part of the same sentence (experiment two). These preferences were interpreted as being due to a focus on the consequences of the represented event in a mental model of the sentence. Experiment three tested this interpretation by using sentence fragments that ended in so (a connective that reinforces the focus on consequences) or because (a connective that conflicts with the focus on consequences). The results confirmed the interpretation: the observed preferences were maintained with so but modified by because. The results are discussed in terms of the structure of represented events, top-down and bottom-up processes, and thematic hierarchies. The first mention effect in pronoun comprehension is also discussed.
(August 1994; 29 pages)
Ref. No. HCRC/RP-54 Price: UKL 1.20

Enric Vallduví
Polarity items, n-words, and minimizers in Catalan and Spanish
Catalan minimizers (expressions denoting some minimal quantity or extent) contain an overt negative marker. On the one hand, when subject to a number of diagnostic tests, the behaviour of these minimizers contrasts completely with the behaviour of (negative) polarity items. On the other, minimizers generally behave like n-words (items which appear to have a dual negative/polar nature), except in non-negative polarity-licensing contexts. In these contexts n-words behave like polarity items and minimizers, as noted, never do. This article argues that Catalan minimizers, rather than n-words, instantiate the true behaviour of negative-concord terms. The fact that n-words may appear in nonnegative polar contexts should not be taken as typical of negative-concord terms but rather as an exceptional phenomenon. As for Spanish, it posesses two classes of minimizers. One class is identical to the Catalan type. The other displays the behaviour of true polarity items. The facts argue against identifying negative-concord terms with true polarity items, and, therefore, against subsuming negative concord under polarity licensing.
(August 1994; 29 pages)
Ref. No. HCRC/RP-55 Price: UKL 1.20

Richard Shillcock and Paul Cairns
Neglect dyslexia and the neuropsychology of reading: a connectionist model of neglect dyslexia with no attentional component
Neuropsychological evidence suggests that the fovea's projections are split precisely between the two hemispheres. This splitting means that a word that is directly fixated in normal reading is initially divided between the two hemispheres. We review the central data relevant to neglect dyslexia. Recent accounts of neglect dyslexia involve either the impairment of prelexical and lexical representations, or the interaction between lexical representations and impaired attentional processes over those representations. We describe a connectionist model of neglect dyslexia which embodies the splitting of projections from the fovea. The model captures the central features of neglect dyslexia without requiring an attentional component.
(July 1994; 9 pages)
Ref. No. HCRC/RP-56 Price: UKL 0.70

Rosemary J. Stevenson and Agnieszka Urbanowicz
The Effects of Sentence Subject, Initial Mention and Pragmatic Plausibility on the Accessibility of a Pronoun's Antecedents
Two experiments showed that the accessibility of a discourse entity is jointly affected by the local salience of the antecedent and by two aspects of the discourse structure: the pragmatic plausibility of the interpretation of the clause containing the pronoun and the presence or absence of prior context. Subjects read target sentences containing two clauses. The first mentioned two individuals, while the second contained a pronoun that referred to one of the individuals. Cleft constructions were used in the first clause so that two local signals to salience, subjecthood and initial mention, could be disentangled. When the clefted individual was the subject, then subjecthood coincided with initial mention but when it was the non-subject, subjecthood and initial mention conflicted. The content of the second clause either biased the interpretation of the pronoun to refer to the subject or the non-subject of the first clause or there was no bias. Context sentences preceded the target sentence in Experiment 1, while the targets were presented in isolation in Experiment 2. Results showed that local salience, pragmatic bias, and discourse context all affected the pronoun assignment preferences and the time taken to read the clause containing the pronoun. We suggest, therefore, that models of discourse processing need to take account of how local salience and global discourse structure are integrated during comprehension.
(November 1994; 29 pages)
Ref. No. HCRC/RP-58 Price: UKL 1.20

Corin Gurr
Supporting Formal Reasoning for Safety Critical Systems
Formal methods can significantly assist in the design and modelling of safety-critical systems. However, formal methods are frequently criticised as being unusable through being too complex and requiring expert knowledge to use. We assert that to make formal methods usable they must be able to be presented in a manner which is readily interpretable. However, we must ensure that the inferences which may be drawn from such a presentation are correct with respect to the formal semantics.
Concurrent systems in which communication occurs between asynchronously operating agents are widely used in safety-critical applications. Unfortunately designing and understanding such systems is made difficult by the interactions between the various concurrent agents.
We present an exercise in the specification and modelling of a safety-critical multiprocessing system fragment. This serves to illustrate three issues which are crucial to the design and modelling of a safety-critical system. These are the advantage of a formal approach, particularly for concurrent systems, the importance of ensuring that a formal model correctly represents the real system and the need to provide a user with a clear understanding (or {\em visualisation}) of the formal model. For this latter point we propose, with examples, the efficacy of a well-founded graphical representation in supporting such an understanding.
(August 1994; 13 pages)
Ref. No. HCRC/RP-59 Price: UKL 0.70

Jasper Taylor
Using Hierarchical Autoepistemic Logic to Model Beliefs in Dialogue
The study of beliefs in dialogue has proceeded along two separate lines: the creation of logical formalisms which capture the inferences that it seems reasonable for agents to make, and the construction of implemented models that approximate what agents actually do. In our project to connect the two approaches we have created a belief model based on a reason maintenance system, and here we present a theory in nonmonotonic logic which describes the behaviour of this model.
We start out by examining the consequences of combining the use of modal operators (to represent nested beliefs) with autoepistemic operators (as a means of nonmonotonic inference). We then look at how some general principles of belief modelling can be represented, such as the persistence and ascription of beliefs, before applying our findings to the construction of a theory for the specific domain in which we are interested. We present a worked example of a dialogue in this domain, and outline the restrictions on the inferences sanctioned by the theory.
Although the purpose of the logic-based theory is merely to illustrate the principles behind the inferences made by the implemented model, we describe ways in which an implemented model can be created directly from the theory by the use of a theorem prover, either an off-the-shelf system or one specifically designed for this logic. Finally we look at the extentions that would need to be made to the theory in order to model further inferences available to human agents that can be made in our chosen domain.
(November 1994; 32 pages)
Ref. No. HCRC/RP-60 Price: UKL 1.30

Keith Stenning, Robert Inder and Irene Neilson
Applying semantic concepts to analysing media and modalities
Our long term goal is an understanding of human communication in terms which would provide the basis for rational design of information presentations. The kernel will be a theory of the cognitive consequences of allocating the same information to different media and modalities, based on the user's information processing characterised in computational terms.
Our theory information allocation starts from an analysis of differences in logical expressiveness of graphical and linguistic representations (Stenning \& Oberlander (1994, in press)). \nocite{stenning oberlander theory article} \nocite{stenning oberlander barnden} This semantic approach requires conceptualisations of {\it medium} and {\it modality} that can be related to representation systems. We propose that media are the physical or perceptual aspects of representations; modalities are classes of interpretation function which map media onto meanings. These interpretations of the terms contrast with existing HCI usage.
Graphical modalities are distinguished from sentential languages by the nature of their interpretation functions. A hierarchy of expressiveness of interpretations of graphics is defined, and compared with interpretations of sentential languages. Using the expressiveness of representations to predict their cognitive properties also requires reference to the availability of constraints of their interpretation to users. Further contrasts between graphics and language emerge in the availability of their constraints.
Three example domains of graphical representations are analysed from this perspective---matrix graphics; logic diagrams; and semantic networks. Some empirical evidence of the usability of these notations is reviewed as evidence that the proposed conceptualisation offers powerful generalisations.
(December 1994; 34 pages)
Ref. No. HCRC/RP-61 Price: UKL 1.30

Francis Corblin
Multiple negation processing
This paper considers negative triggers (negative and negative quantifiers) and the interpretation of simple sentences containing more than one occurrence of those items (multiple negation sentences). In the most typical interpretations those sentences have more negative expressions than negations in their semantic representation. It is first shown that this compositionality problem remains in current approaches. A principled algorithm for deriving the representation of sentences with multiple negative quantifiers in a DRT framework (Kamp and Reyle, 1993) is then introduced. The algorithm is under the control of an on-line check-in, keeping the complexity of negation auto-embedding below a threshold of complexity. This mechanisn is seen as a competence limitation imposing (and licensing) the "abrogation of compositionality" (May 1989) observed in the so-called negative concord readings (Labov 1972, Zanuttini 1991, Ladusaw 1992). A solution to the compositionality problem is thus proposed, which is based on a control of the processing input motivated by a limitation of the processing mechanism itself.
(November 1994; 32 pages)
Ref. No. HCRC/RP-62 Price: UKL 1.30

Robert Inder and Jon Oberlander
Applying Discourse Theory to aid Hypertext Navigation
We discuss ways of improving navigation facilities in hypertext systems, considering theoretical and implementation issues, from a natural language processing research perspective. The key claim is that certain ideas from the theory of discourse structure can be exploited to improve the context-sensitivity of navigation facilities. In particular, (i) after a node is reached by a jump, links to other nodes can by dynamically suppressed where these are rendered irrelevant by the user's mode of arrival; and (ii) at any point after a jump (or sequence of jumps), the user can easily return to a limited set of structurally accessible nodes. The approach therefore combines certain advantages of navigation based on document structure with those based on individual interaction histories. The ideas have been tested by altering the implementation of \info, the hypertextual help system built into the \emacs\ text editor, which runs on Unix, Macintosh and {\sc pc} systems. Evaluation studies are currently being planned, and we see further potential for exploiting notions from discourse structure to inform the design of navigation aids.
(December 1994; 12 pages)
Ref. No. HCRC/RP-64 Price: UKL 0.70