Artificial intelligence,language, and the study of knowledge

This paper studies the relationship of Artificial Intelligence to the study of language and the representation of the underlying knowledge which supports the comprehension process. It develops the view that intelligence is based on the ability to use large amounts of diverse kinds of knowledge in procedural ways, rather than on the possession of a few general and uniform principles. The paper also provides a unifying thread to a variety of recent approaches to natural language comprehension. We conclude with a brief discussion of how Artificial Intelligence may have a radical impact on education if the principles which it utilizes to explore the representation and use of knowledge are made available to the student to use in his own learning experiences.     

Answer Sets and Qualitative Decision Making

Logic programs under answer set semantics have become popular as a knowledge representation formalism in Artificial Intelligence. In this paper we investigate the possibility of using answer sets for qualitative decision making. Our approach is based on an extension of the formalism, called logic programs with ordered disjunction (LPODs). These programs contain a new connective called ordered disjunction. The new connective allows us to represent alternative, ranked options for problem solutions in the heads of rules: A × B intuitively means: if possible A, but if A is not possible then at least B. The semantics of logic programs with ordered disjunction is based on a preference relation on answer sets. We show that LPODs can serve as a basis for qualitative decision making.     

A general possible worlds framework for reasoning about knowledge and belief

In this paper non-normal worlds semantics is presented as a basic, general, and unifying approach to epistemic logic. The semantical framework of non-normal worlds is compared to the model theories of several logics for knowledge and belief that were recently developed in Artificial Intelligence (AI). It is shown that every model for implicit and explicit belief (Levesque), for awareness, general awareness, and local reasoning (Fagin and Halpern), and for awareness and principles (van der Hoek and Meyer) induces a non-normal worlds model validating precisely the same formulas (of the language in question).     

Wittgenstein,Korsgaard and the Publicity of Reasons

In The Sources of Normativity (Korsgaard, Christine. The Sources of Normativity. Cambridge: Cambridge University Press, 1996), Christine Korsgaard tried to argue against what she called the ‘privacy’ of reasons, appealing to Wittgenstein's argument against the possibility of a private language. In recent work she continues to endorse Wittgenstein's perspective on the normativity of meaning, although she now emphasizes that her own argument was only meant to be analogous to the private language argument. The purpose of the present paper is to show that the Wittgensteinian perspective is not only not useful in support of Korsgaard's general project, but that it is positively inimical to it, in two ways. First, Wittgenstein opposes views on which principled or rule-following behavior requires that one be guided by anything like a mental representation of a rule or principle. But for Korsgaard, human action essentially requires this. Second, Wittgenstein systematically attempts to de-emphasize the importance of the first-personal perspective, and to emphasize the social functions even of concepts that might seem deployed primarily from that perspective: for example, concepts of sensations and intentions. This is the reverse of Korsgaard's emphasis. The paper also argues, however, that the private language argument does have some implications for a theory of rationality and reasons.     

Motives,mechanisms, and emotions

Abstract

Intelligent animals are solutions to a design problem posed by: the varying requirements of individuals, the more permanent requirements of species and social groups, the constraints of the environment, and the available biological mechanisms. Analysis of this design problem, especially the implications of limited knowledge and a continuous flow of information in a rapidly changing environment, leads to a theory of how new motives are processed in an intelligent system. The need for speed leads to architectures and algorithms that are fallible in ways that explain why intelligent agents are susceptible to emotions and errors. This holds also for intelligent robots. A study of such mechanisms and processes is a step towards a computational theory of emotions, attitudes, moods, character traits, and other aspects of mind so far not studied in Artificial Intelligence. In particular, it turns out that no special emotional subsystem is required. This framework clarifies and refines ordinary concepts of mental processes, and suggests a computational approach to psychotherapy.     

Cognitive development as an executive process – in part: a homeopathic dose of Piaget

Piaget’s theory of cognitive development retains its importance through showing us how the exercise of agency is necessary to the development of self–world dualism and to the developing ability to frame explicit judgements about the physical and mental world. I begin by describing the Piagetian position in my own terms (Agency: Its role in mental development, 1996, for a fuller account) and then set it in the context of research on the executive functions. I also argue, however, that the theory lacks the resources to explain second-order mental representation. The theory was insufficiently nativist in general and insufficiently nativist about symbolic capacities in particular. But adopting a nativist view does not preclude one from taking a Piagetian line on the essential contribution of the first-person experience of agency to cognitive development.     

Deranging the Mental Lexicon

This paper offers a defense of Davidson’s conclusion in ‘A Nice Derangement of Epitaphs’, focusing on the psychology and epistemology of language. Drawing on empirical studies in language acquisition and sociolinguistics, I problematize the traditional idealizing assumption that a person’s mental lexicon consists of two distinct parts—a dictionary, comprising her knowledge of word meanings proper, and an encyclopedia, comprising her wider knowledge of worldly affairs. I argue that the breakdown of the dictionary–encyclopedia distinction can be given a cognitive and functional explanation: facts regarding language learning and the challenges of coping with linguistically diverse environments require that dictionary and encyclopedia remain deeply integrated rather than categorically distinct dimensions of the mental lexicon. This argument provides support for a psychologized version of Davidson’s conclusion in ‘Derangement’: there is no such thing as a language, in the sense that there is no diachronically stable and uniquely specifiable object that could constitute the language which a person knows. I then apply this conclusion to the question of whether the concept of a public language—understood as a more or less stable body of conventions shared by a group of speakers—could nonetheless retain an important explanatory role in philosophy of language and linguistics.     

Unarticulated meaning

It is a common experience of mental life that we come to articulate meanings which we had initially grasped in only a sketchy way. In this paper, I consider how this idea of an initially unarticulated meaning may fit in a general theory of mental representation. I propose to identify unarticulated meanings with what I callspecific concepts, which are quite similar to Rosch's categories of basic objects and are distinct both from images and generic concepts (which come to articulate meanings). I argue that unarticulated meaning is non-representational in an important respect, a claim which relies on a distinction amonglevels of representation.     

Nap‐dependent learning in infants

Sleep has been shown to aid a variety of learning and memory processes in adults ( Stickgold, 2005 ). Recently, we showed that infants’ learning also benefits from subsequent sleep such that infants who nap are able to abstract the general grammatical pattern of a briefly presented artificial language ( Gomez, Bootzin & Nadel, 2006 ). In the present study, we demonstrate, for the first time, long‐term effects of sleep on memory for an artificial language. Fifteen‐month‐old infants who had napped within 4 hours of language exposure remembered the general grammatical pattern of the language 24 hours later. In contrast, infants who had not napped shortly after being familiarized with the language showed no evidence of remembering anything about the language. Our findings support the view that infants’ frequent napping plays an essential role in establishing long‐term memory.     

Logical Foundations for Belief Representation

This essay presents a philosophical and computational theory of the representation of de re, de dicto, nested, and quasi-indexical belief reports expressed in natural language. The propositional Semantic Network Processing System (SNePS) is used for representing and reasoning about these reports. In particular, quasi-indicators (indexical expressions occurring in intentional contexts and representing uses of indicators by another speaker) pose problems far natural-language representation and reasoning systems, because—unlike pure indicators—they cannot be replaced by coreferential NPs without changing the meaning of the embedding sentence. Therefore, the referent of the quasi-indicator must be represented in such a way that no invalid coreferential claims are entailed. The importance of quasi-indicators is discussed, and it is shown that all four of the above categories of belief reports can be handled by a single representational technique using belief spaces containing intensional entities. Inference rules and belief-revision techniques for the system ore also examined.     

Meaning and Interpretation. I

The paper is an attempt at a logical explication of some crucial notions of current general semantics and pragmatics. A general, axiomatic, formal-logical theory of meaning and interpretation is outlined in this paper.In the theory, accordingto the token-type distinction of Peirce, language is formalised on two levels: first as a language of token-objects (understood as material, empirical, enduring through time-and space objects) and then – as a language of type-objects (understood as abstract objects, as classes of tokens). The basic concepts of the theory, i.e. the notions: meaning, denotation and interpretation of well-formed expressions (wfes) of the language are formalised on the type-level, by utilising some semantic-pragmatic primitive notions introduced on the token-level. The paper is divided into two parts.In Part Ia theoryof meaningand denotation is proposed, and in Part II - its expansion to the theory of meaning and interpretation is presented.The meaninga wfe is defined as an equivalence class of the relation possessing the same manner of using types (cf. Ajdukiewicz [1934], Wittgenstein [1953]). The concept of denotation is defined by means of the relation of referring which holds between wfe-types and objects of reality described by the given language. Presented by Wojciech Buszkowski     

Temporal Interaction of Information and Belief

The temporal updating of an agent’s beliefs in response to a flow of information is modeled in a simple modal logic that, for every date t, contains a normal belief operator B _t and a non-normal information operator I _t which is analogous to the ‘only knowing’ operator discussed in the computer science literature. Soundness and completeness of the logic are proved and the relationship between the proposed logic, the AGM theory of belief revision and the notion of plausibility is discussed. A first draft of this paper was presented at the Workshop on Belief Change in Rational Agents: Perspectives from Artificial Intelligence, Philosophy and Economics, Dagstuhl (Germany), August 2005. Special Issue Formal Epistemology II. Edited by Branden Fitelson     

Induction: Representation,strategy and argument

In order to be a general theory of human cognition, the theory of mental models needs to accommodate a variety of forms of reasoning in addition to deduction. The mental model theory of induction is a crucial step in establishing generality. After suggesting that the theory of mental models can also account for abduction and analogy, the paper points out that inductive performance is likely to be constrained both by the nature of the representation used and by strategic factors. Since human cognition involves the communication of arguments, a final section explores the relationship between the theory of mental models and the notion of an argument. It proposes that models can contain tokens of inductive arguments which can be referred to in the course of other arguments.     

Children''s Representational Theory of Language:The Problem of Opaque Contexts

Around four years of age, children recognize that action is less a consequence of the way the world is than the way it is represented by the actor. This understanding is characterized as a “theory of mind.” This study examines the possibility of the development of a parallel theory of language; specifically, the understanding that, in opaque contexts, terms do not simply map on to the referent of the expression, but rather indicate how that object is to be represented. 120 3- to 7-year-olds were tested on their theory of mind (using false belief tasks) and sensitivity to opaque contexts. Children who passed false belief tasks performed more successfully on the opacity measure than those who did not, even when age was partialled out (r (117) = .2453, p < .01). It is concluded that children come to realize that language does not refer to the world directly, but rather via one's representation of it. The results are consistent with the view that both abilities are manifestations of a more general understanding of representation, and that children's theories of mind and language follow similar developmental paths.     

Hume’s naturalistic theory of representation

Hume is a naturalist in many different respects and about many different topics; this paper argues that he is also a naturalist about intentionality and representation. It does so in the course of answering four questions about his theory of mental representation: (1) Which perceptions represent? (2) What can perceptions represent? (3) Why do perceptions represent at all? (4) Howdo perceptions represent what they do? It appears that, for Hume, all perceptions except passions can represent; and they can represent bodies, minds, and persons, with their various qualities. In addition, ideas can represent impressions and other ideas. However, he explicitly rejects the view that ideas are inherently representational, and he implicitly adopts a view according to which things (whether mental or non-mental) represent in virtue of playing, through the production of mental effects and dispositions, a significant part of the causal and/or functional role of what they represent. It is in virtue of their particular functional roles that qualitatively identical ideas are capable of representing particulars or general kinds; substances or modes; relations; past, present, or future; and individuals or compounds.     

Tracking, Epistemic Dispositions and the Conditional Analysis

According to Nozick’s tracking theory of knowledge, an agent a knows that p just in case her belief that p is true and also satisfies the two tracking conditionals that had p been false, she would not have believed that p, and had p been true under slightly different circumstances, she would still have believed that p. In this paper I wish to highlight an interesting but generally ignored feature of this theory: namely that it is reminiscent of a dispositional account of knowledge: it invites us to think of knowledge as a manifestation of a cognitive disposition to form true beliefs. Indeed, given a general account of dispositions in terms of subjunctive conditionals, the two tracking conditionals are satisfied just in case the belief in question results from some cognitive disposition to form true beliefs. Recently, such a conditional account of dispositions has, however, been criticised for its vulnerability to so-called ‘masked’, ‘mimicked’ and ‘finkish’ counterexamples. I show how the classical counterexamples to Nozick’s theory divide smoothly into four corresponding categories of counterexamples from epistemic masking, mimicking and finkishness. This provides strong evidence for the thesis that satisfaction of the two tracking conditionals is symptomatic of knowledge and that knowledge is instead constituted by a dispositional capability to form true beliefs. The attempt to capture such a cognitive, dispositional capability in terms of the tracking conditionals, although providing a good approximation in a wide variety of cases, still comes apart from the real thing whenever the epistemic layout is characterised by masking-, mimicking- and finkish mechanisms. In the last part of the paper I explore the prospect of improving the tracking theory in the light of these findings.     

The Occasion-Sensitivity of Thought

On the most common interpretation of occasion-sensitivity what varies cross-contextually is the truth-conditional content of representations. Jerry Fodor argues that when extended to mental representation this view has some problematic consequences. In this paper I outline an approach to occasion-sensitivity which circumvents Fodor’s objections but still maintains that the aspect of thought that guides deliberation and action is occasion-sensitive. On the proposed view, what varies cross-contextually are not truth conditions but rather the conditions for accepting a (true) representation as true relative to a practical goal that is pursued on an occasion. I show that although the proposal entails an error theory this theory is not problematic since it is meant to compensate for the over-generating nature of semantic competence, namely, the fact that not all of the representation’s truth-makers are conducive to a given contextually salient goal.

     

Recent Scholarship on Hume's Theory of Mental Representation

In a recent paper, Karl Schafer argues that Hume's theory of mental representation has two distinct components, unified by their shared feature of having accuracy conditions. As Schafer sees it, simple and complex ideas represent the intrinsic imagistic features of their objects whereas abstract ideas represent the relations or structures in which multiple objects stand. This distinction, however, is untenable for at least two related reasons. Firstly, complex ideas represent the relations or structures in which the impressions that are the objects of their simple components stand. Secondly, abstract ideas are themselves instances of complex ideas. I draw two important conclusions from these facts. Firstly, contra Schafer and Garrett (to whom Schafer responds), the Copy Principle, properly emended, constitutes the entirety of Hume's theory of mental representation. Secondly, whereas paradigm examples of complex ideas, e.g. ideas of spatial and temporal complexes, are structured by relations of contiguity, abstract ideas are those complex ideas instead structured by relations of resemblance. As such, they represent their objects not as spatially or temporally contiguous but rather as resembling.