The term "projection" describes cases where some element of meaning unexpectedly escapes from the scope of another expression, such as a negation or conditional. For example, "John didn't see his sister" is typically understood as denying that there was an event of John seeing his sister, but not as denying that John has a sister -- even though syntactically the expression "his sister" is under the scope of the negation. An enormous range of expressions yield projection, including presupposition triggers, conventional implicature triggers, approximatives, and some inferences associated with Gricean implicatures. This project will carry out the first systematic empirical and theoretical analysis of the full range of projective meanings, to establish why projection occurs, and how subclasses of projective meaning differ. To this end, it will develop templates for diagnostic tests that can be used to study properties of projective meanings cross-linguistically, with theoretically naive consultants. The researchers will conduct detailed investigations of projective meanings in English and Paraguayan Guarani, and disseminate the techniques to fieldworkers studying other languages, through publications and workshops.

Modern linguistic semantics centers around Fregean compositionality, the way in which the meanings of parts are combined to give the meanings of larger expressions. While there has been great progress in compositional semantics, projective meaning does not obey ordinary compositional rules. Until now, there has been no uniform account of why projective meanings behave differently from ordinary content. This project seeks to explain projection, based on the working hypothesis that aspects of meaning project if and only if they are not at-issue. The notion of at-issueness is a pragmatic one, based on what questions are under discussion in a discourse. Thus the project will place a relatively little-studied pragmatic notion at the heart of work on linguistic meaning. A first area where the innovations in the project are of broad significance is in its cross-linguistic focus: a theory of human language cannot be based on one language alone, and the project will apply its empirical techniques cross-linguistically. Beyond linguistic semantics, the issues studied are fundamental to Philosophy of Language, since the work subtly redraws the boundary between Semantics and Pragmatics. Further the project is of practical significance in the field of Natural Language Processing (specifically, the subfield of Computational Semantics). Text processing systems, e.g. computer systems for automatically answering a user's questions, must take account of inferences which arise independently of standard composition. A question answering system which does not take account of what is projected will not be able to identify what the user already knows, and will not be able to identify what question needs to be answered.

Project Report

To understand how people communicate, for example in order to build machines that can interact with humans effectively, we must understand the inferences people make. The overarching goal of this project is to study a particular type of inference, what we term projective inference. Projective inferences occur when part of the content of a sentence is understood as a commitment of the speaker even though the sentence itself is a negation, a statement of mere possibility, a question, or other construction which typically is not expected to produce such speaker commitment. Consider, for example, the standard case below: A speaker who uttered any of the sentences in (1-4) would normally be understood to believe that Jane used to teach linguistics, even though this is not entailed by any of (2-4). Indeed utterance of (4) gives rise to this implication of speaker commitment even though it is not an assertion at all. 1. Jane stopped teaching linguistics. 2. Jane didn’t stop teaching linguistics. 3. Jane might have stopped teaching linguistics. 4. Has Jane stopped teaching linguistics? Projective inferences are pervasive in language, and yet poorly understood. Various sub-types of projective inference have been studied over the last 40 years, and much progress has been made in studying several of these sub-types. But the main premise of the project was that there has never previously been a sustained scholarly effort to understand the phenomenon of projectivity as a whole. The primary goals of the project were to develop diagnostic methods that can be used to investigate projective meanings cross-linguistically, and to use those methods to develop a new theoretical model of the phenomenon of projection. In the project, we developed diagnostic methods for studying projective meanings both in a fieldwork setting, and in a laboratory or-web-based experimental setting. We applied these methods in a range of languages, including not only English, but also Guarani (spoken in Paraguay), Kich’e Mayan (Guatemala), German, Hungarian, and French. The results we obtained show a range of features of projective meanings which are common across all these languages. We reported some of these results in a paper which presents a new, theoretically motivated taxonomy of different types of projective meaning, published in the leading general linguistics journal, Language. Other empirical and theoretical results from the project are reported in 4 further journal articles, and 21 book chapters and papers appearing in conference proceedings. In addition, the project provided substantial training opportunities for new researchers, both through training of graduate student research assistants, and through 3 courses and 4 workshops organized by the PIs. The results of the project are of significance beyond linguistics, within the disciplines of philosophy, psychology, and computer science. In philosophy, the results of the project relate to standard problems in the philosophy of language that date back to work of Frege and to the famous debate on descriptions between Russell and Strawson. This debate (still much discussed) hinged on the meaning of definite descriptions, which are one of the best known types of projective expression. The results of the current project suggest a new way of looking at this old debate, in terms of whether or not the descriptive content of definite descriptions is what we term "at-issue". In psychology, the project results relate to how human attention is directed in communicative interactions. Specifically, our results suggest that as conversation partners negotiate their way through a conversation, their attention should be modeled in terms of questions which are currently under discussion, and furthermore that projective meanings, which are not at-issue, should be processed differently than other types of meaning. It remains open whether psycho-linguistic work can reveal patterns of processing activity that distinguish the at-issue meaning from not-at-issue meaning. As regards computer science, inference is a hot topic within the Natural Language Processing community. We suggest that our success in unifying an important subset of inferences under the single umbrella of projective meaning offers a pathway that future Natural Language Processing systems should follow if they are to model human inference, and thus take human-computer interaction to a new level.

National Science Foundation (NSF)
Division of Behavioral and Cognitive Sciences (BCS)
Standard Grant (Standard)
Application #
Program Officer
William J. Badecker
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Carnegie-Mellon University
United States
Zip Code