William J. Rapaport

This essay considers what it means to understand natural language and whether a computer running an artificial-intelligence program designed to understand natural language does in fact do so. It is argued that a certain kind of semantics is needed to understand natural language, that this kind of semantics is mere symbol manipulation (i.e., syntax), and that, hence, it is available to AI systems. Recent arguments by Searle and Dretske to the effect that computers cannot understand natural language are discussed, and a prototype natural-language-understanding system is presented as an illustration.

“Contextual” vocabulary acquisition is the active, deliberate acquisition of a meaning for a word in a text by reasoning from textual clues and prior knowledge, including language knowledge and hypotheses developed from prior encounters with the word, but without external sources of help such as dictionaries or people. But what is “context”? Is it just the surrounding text? Does it include the reader’s background knowledge? I argue that the appropriate context for contextual vocabulary acquisition is the reader’s “internalization” of the text “integrated” into the reader’s “prior” knowledge via belief revision.

This project continues our interdisciplinary research into computational and cognitive aspects of narrative comprehension. Our ultimate goal is the development of a computational theory of how humans understand narrative texts. The theory will be informed by joint research from the viewpoints of linguistics, cognitive psychology, the study of language acquisition, literary theory, geography, philosophy, and artificial intelligence. The linguists, literary theorists, and geographers in our group are developing theories of narrative language and spatial understanding that are being tested by the cognitive psychologists and language researchers in our group, and a computational model of a reader of narrative text is being developed by the AI researchers, based in part on these theories and results and in part on research on knowledge representation and reasoning. This proposal describes the knowledge-representation and natural-language-processing issues involved in the computational implementation of the theory; discusses a contrast between communicative and narrative uses of language and of the relation of the narrative text to the story world it describes; investigates linguistic, literary, and hermeneutic dimensions of our research; presents a computational investigation of subjective sentences and reference in narrative; studies children’s acquisition of the ability to take third-person perspective in their own storytelling; describes the psychological validation of various linguistic devices; and examines how readers develop an understanding of the geographical space of a story. This report is a longer version of a project description submitted to NSF. This document, produced in May 2007, is a L ATEX version of Technical Report 89-07 (Buffalo: SUNY Buffalo Department of Computer Science, August 1989), with slightly..

Deliberate contextual vocabulary acquisition (CVA) is a reader’s ability to figure out a (not the) meaning for an unknown word from its “context”, without external sources of help such as dictionaries or people. The appropriate context for such CVA is the “belief-revised integration” of the reader’s prior knowledge with the reader’s “internalization” of the text. We discuss unwarranted assumptions behind some classic objections to CVA, and present and defend a computational theory of CVA that we have adapted to a new classroom curriculum designed to help students use CVA to improve their reading comprehension.

A computer can come to understand natural language the same way Helen Keller did: by using “syntactic semantics”—a theory of how syntax can suffice for semantics, i.e., how semantics for natural language can be provided by means of computational symbol manipulation. This essay considers real-life approximations of Chinese Rooms, focusing on Helen Keller’s experiences growing up deaf and blind, locked in a sort of Chinese Room yet learning how to communicate with the outside world. Using the SNePS computational knowledge-representation system, the essay analyzes Keller’s belief that learning that “everything has a name” was the key to her success, enabling her to “partition” her mental concepts into mental representations of: words, objects, and the naming relations between them. It next looks at Herbert Terrace’s theory of naming, which is akin to Keller’s, and which only humans are supposed to be capable of. The essay suggests that computers at least, and perhaps non-human primates, are also capable of this kind of naming.

In this reply to James H. Fetzer’s “Minds and Machines: Limits to Simulations of Thought and Action”, I argue that computationalism should not be the view that (human) cognition is computation, but that it should be the view that cognition (simpliciter) is computable. It follows that computationalism can be true even if (human) cognition is not the result of computations in the brain. I also argue that, if semiotic systems are systems that interpret signs, then both humans and computers are semiotic systems. Finally, I suggest that minds can be considered as virtual machines implemented in certain semiotic systems, primarily the brain, but also AI computers. In doing so, I take issue with Fetzer’s arguments to the contrary.

We discuss a research project that develops and applies algorithms for computational contextual vocabulary acquisition (CVA): learning the meaning of unknown words from context. We try to unify a disparate literature on the topic of CVA from psychology, first- and secondlanguage acquisition, and reading science, in order to help develop these algorithms: We use the knowledge gained from the computational CVA system to build an educational curriculum for enhancing students’ abilities to use CVA strategies in their reading of science texts at the middle-school and college undergraduate levels. The knowledge gained from case studies of students using our CVA techniques feeds back into further development of our computational theory. Keywords: artificial intelligence, knowledge representation, reading, reasoning, science education, vocabulary acquisition.

John Searle once said: "The Chinese room shows what we knew all along: syntax by itself is not sufficient for semantics. (Does anyone actually deny this point, I mean straight out? Is anyone actually willing to say, straight out, that they think that syntax, in the sense of formal symbols, is really the same as semantic content, in the sense of meanings, thought contents, understanding, etc.?)." I say: "Yes". Stuart C. Shapiro has said: "Does that make any sense? Yes: Everything makes sense. The question is: What sense does it make?" This essay explores what sense it makes to say that syntax by itself is sufficient for semantics.

Cognitive agents, whether human or computer, that engage in natural-language discourse and that have beliefs about the beliefs of other cognitive agents must be able to represent objects the way they believe them to be and the way they believe others believe them to be. They must be able to represent other cognitive agents both as objects of beliefs and as agents of beliefs. They must be able to represent their own beliefs, and they must be able to represent beliefs as objects of beliefs. These requirements raise questions about the number of tokens of the belief representation language needed to represent believers and propositions in their normal roles and in their roles as objects of beliefs. In this paper, we explicate the relations among nodes, mental tokens, concepts, actual objects, concepts in the belief spaces of an agent and the agent's model of other agents, concepts of other cognitive agents, and propositions. We extend, deepen, and clarify our theory of intensional knowledge representation for natural-language processing, as presented in previous papers and in light of objections raised by others. The essential claim is that tokens in a knowledge-representation system represent only intensions and not extensions. We are pursuing this investigation by building CASSIE, a computer model of a cognitive agent and, to the extent she works, a cognitive agent herself. CASSIE's mind is implemented in the SNePS knowledge-representation and reasoning system.

This essay presents and defends a triage theory of grading: An item to be graded should get full credit if and only if it is clearly or substantially correct, minimal credit if and only if it is clearly or substantially incorrect, and partial credit if and only if it is neither of the above; no other (intermediate) grades should be given. Details on how to implement this are provided, and further issues in the philosophy of grading (reasons for and against grading, grading on a curve, and the subjectivity of grading) are discussed

Contextual vocabulary acquisition (CVA) is the deliberate acquisition of a meaning for a word in a text by reasoning from context, where “context” includes: (1) the reader’s “internalization” of the surrounding text, i.e., the reader’s “mental model” of the word’s “textual context” (hereafter, “co-text” [3]) integrated with (2) the reader’s prior knowledge (PK), but it excludes (3) external sources such as dictionaries or people. CVA is what you do when you come across an unfamiliar word in your reading, realize that you don’t know what it means, decide that you need to know what it means in order to understand the passage, but there is no one around to ask, and it is not in the dictionary (or you are too lazy to look it up). In such a case, you can try to figure out its meaning “from context”, i.e., from clues in the co-text together with your prior knowledge. Our computational theory of CVA—implemented in a the SNePS knowledge representation and reasoning system [28]—begins with a stored knowledge base containing SNePS representations of relevant PK, inputs SNePS representations of a passage containing an unfamiliar word, and draws inferences from these two (integrated) information sources. When asked to define the word, definition algorithms deductively search the resulting network for information of the sort that might be found in a dictionary definition, outputting a definition frame whose slots are the kinds of features that a definition might contain (e.g., class membership, properties, actions, spatio-temporal information, etc.) and whose slot-fillers contain information gleaned from the network [6–8,20,23,24]. We are investigating ways to make our system more robust, to embed it in a naturallanguage-processing system, and to incorporate morphological information. Our research group, including reading educators, is also applying our methods to the develop-

In the Fall of 1983, I offered a junior/senior-level course in Philosophy of Artificial Intelligence, in the Department of Philosophy at SUNY Fredonia, after returning there from a year’s leave to study and do research in computer science and artificial intelligence (AI) at SUNY Buffalo. Of the 30 students enrolled, most were computerscience majors, about a third had no computer background, and only a handful had studied any philosophy. (I might note that enrollments have subsequently increased in the Philosophy Department’s AI-related courses, such as logic, philosophy of mind, and epistemology, and that several computer science students have added philosophy as a second major.) This article describes that course, provides material for use in such a course, and offers a bibliography of relevant articles in the AI, cognitive science, and philosophical literature.

It is well known that people from other disciplines have made significant contributions to philosophy and have influenced philosophers. It is also true (though perhaps not often realized, since philosophers are not on the receiving end, so to speak) that philosophers have made significant contributions to other disciplines and have influenced researchers in these other disciplines, sometimes more so than they have influenced philosophy itself. But what is perhaps not as well known as it ought to be is that researchers in other disciplines, writing in those other disciplines' journals and conference proceedings, are doing philosophically sophisticated work, work that we in philosophy ignore at our peril. Work in cognitive science and artificial intelligence (AI) often overlaps such paradigmatic philosophical specialties as logic, the philosophy of mind, the philosophy of language, and the philosophy of action. This special issue offers a sampling of research in cognitive science and AI that is philosophically relevant and philosophically sophisticated.

Hauser argues that his pocket calculator (Cal) has certain arithmetical abilities: it seems Cal calculates. That calculating is thinking seems equally untendentious. Yet these two claims together provide premises for a seemingly valid syllogism whose conclusion - Cal thinks - most would deny. He considers several ways to avoid this conclusion, and finds them mostly wanting. Either we ourselves can't be said to think or calculate if our calculation-like performances are judged by the standards proposed to rule out Cal; or the standards- e.g., autonomy and self-consciousness- make it impossible to verify whether anything or anyone (save oneself) meets them. While appeals to the intentionality of thought or the unity of minds provide more credible lines of resistance, available accounts of intentionality and mental unity are insufficiently clear and warranted to provide very substantial arguments against Cal's title to be called a thinking thing. Indeed, considerations favoring granting that title are more formidable than is generally appreciated. Rapaport's comments suggest that, on a strong view of thinking, mere calculating is not thinking (and pocket calculators don't think), but on a weak, but unexciting, sense of thinking, pocket calculators do think. He closes with some observations on the implications of this conclusion.

SNePS, the Semantic Network Processing System 45, 54], has been designed to be a system for representing the beliefs of a natural-language-using intelligent system (a \cognitive agent"). It has always been the intention that a SNePS-based \knowledge base" would ultimatelybe built, not by a programmeror knowledge engineer entering representations of knowledge in some formallanguage or data entry system, but by a human informing it using a natural language (NL) (generally supposed to be English), or by the system reading books or articles that had been prepared for human readers. Because of this motivation, the criteria for the development of SNePS have included: it should be able to represent anything and everything expressible in NL; it should be able to represent generic, as well as speci c information; it should be able to use the generic and the speci c information to reason and infer information implied by what it has been told; it cannot count on any particular order among the pieces of information it is given; it must continue to act reasonably even if the information it is given includes circular de nitions, recursive rules, and inconsistent information.

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 for 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 are also examined.

We present a computational analysis of de re, de dicto, and de se belief and knowledge reports. Our analysis solves a problem first observed by Hector-Neri Castañeda, namely, that the simple rule `(A knows that P) implies P' apparently does not hold if P contains a quasi-indexical. We present a single rule, in the context of a knowledge-representation and reasoning system, that holds for all P, including those containing quasi-indexicals. In so doing, we explore the difference between reasoning in a public communication language and in a knowledge-representation language, we demonstrate the importance of representing proper names explicitly, and we provide support for the necessity of considering sentences in the context of extended discourse (for example, written narrative) in order to fully capture certain features of their semantics.

Alexius Meinong developed a notion of defective objects in order to account for various logical and psychological paradoxes. The notion is of historical interest, since it presages recent work on the logical paradoxes by Herzberger and Kripke. But it fails to do the job it was designed for. However, a technique implicit in Meinong's investigation is more successful and can be adapted to resolve a similar paradox discovered by Romane Clark in a revised version of Meinong's Theory of Objects due to Rapaport. One family of paradoxes remains, but it is argued that they are unavoidable and relatively harmless.