My main research interests lie in the area of semantics, pragmatics, and computational linguistics. The overarching goals are (i) to integrate and contribute to methodological advances and empirical findings in neighboring disciplines; (ii) to deepen our understanding of the interactions between different semantic dimensions, such as modality and temporality; and (iii) to enhance the empirical base of semantic and pragmatic theory through cross-linguistic comparisons and experimental studies. Within the general area of meaning, I pursue a number of sub-projects on topics which can be broadly classified into three categories:

• The language of uncertainty
• Conditionals and causality
• Temporality and modality
• Free choice and modality
• Rational linguistic behavior
• Data-driven approaches to meaning

The language of uncertainty

My ongoing work in this area focuses on conditionals, modality, tense, and aspect. The general goal is to advance our understanding of their meaning and use, the inferences involved in their interpretation, and their semantic interactions with other grammatical categories.

Conditionals and causality.

Recent years have seen a convergence of approaches to counterfactuals in Philosophy, Psychology, and Artificial Intelligence. Causal relations are being recognized as an important factor in their interpretation, and research in these disciplines is producing new mathematical models of causation, model-theoretic accounts of the interpretation of conditionals, and psychological experimental results.

The goals of this project are to explore the implications of these advances for the formal apparatus employed in linguistic analyses, and to address questions of specifically linguistic interest from this perspective. In a series of recent talks, I argued that Premise Semantics – developed by Veltman and Kratzer and among the dominant formal approaches in Linguistics – is a well-suited general framework in which causal dependencies can be integrated as a building block instead of, or in addition to, other notions such as Kratzer's “Lumping” relation.

Based on this formal groundwork, the questions this project will address next are of more immediate linguistic interest: How do the linguistic properties of sentences affect their interpretation in causal models? For instance, most speakers would assent to a sentence like If the sprinkler had been on, the weather would have been dry more readily than to its counterpart If I had turned the sprinkler on, the weather would have been dry. Based on earlier analyses of causal elements in the semantics of aspectual classes and thematic roles, the goal in the long run is to derive such contrasts via subsentential compositional analysis from the lexical properties of the expressions involved.

Selected publications:

Recent slides

Kaufmann, Stefan. 2008. Conditionals right and left: Probabilities for the whole family. Journal of Philosophical Logic. [Online first]

Kanazawa, Makoto, Stefan Kaufmann, and Stanley Peters. 2005. On the lumping semantics of counterfactuals. Journal of Semantics 22:129-151.

Kaufmann, Stefan. 2005. Conditionals. In Brown, Keith, editor-in-chief, Encyclopedia of Language and Linguistics, 2nd edition, Volume 3, pages 6-9. Elsevier.

Kaufmann, Stefan. 2005. Conditionals predictions: A probabilistic account. Linguistics and Philosophy 28:181-231.

Kaufmann, Stefan. 2004. Conditioning against the grain: Abduction and indicative conditionals. Journal of Philosophical Logic 33:583-606.

Temporality and modality.

This is a cross-linguistic exploration of the interplay between temporal and modal semantic dimensions in the interpretation of various linguistic expressions.

Jointly with Yukinori Takubo (Kyoto University), I am analyzing Japanese expressions of temporal precedence in comparison with their English counterparts. While English A before B does not entail B (in contrast to B after A; cf. Mozart died before he finished the Requiem), it does imply that B is “likely” in some sense prior to the reference time of A. This gives rise under certain conditions to the counterfactual implication that B would have been true, had it not been for A. This modal element has been studied closely in the English case (the most recent analysis is due to Beaver and Condoravdi). This project contributes to this line of research on both the empirical and the theoretical sides by contrasting English A before B with two Japanese counterparts, B mae-ni A (`before B, A') and B-nai uti-ni A (lit. `while not yet B, A'). The Japanese expressions differ with regard to the counterfactual inferences they give rise to. Our work includes a novel analysis of the interplay between aspectual properties, tenses and temporal connectives in Japanese, and of the modal properties that are responsible for the modal differences.

In another collaboration in this vein, with Setsuko Arita (Osaka Shoin Women's University) I analyze the uses of Japanese conditional connectives with special attention to the ways in which they interact semantically with the temporal morphology in the constituent clauses. This is an extension of previous work of my own on the temporal interpretation of English conditionals to the Japanese case.

Selected publications:

Kaufmann, Stefan, Cleo Condoravdi, and Valentina Harizanov. 2006. Formal approaches to modality. In Frawley, William, editor, The Expression of Modality, pages 71-106. Mouton de Gruyter.

Kaufmann, Stefan and Yukinori Takubo. 2005. Non-veridical uses of Japanese expressions of temporal precedence. In McGloin, Naomi H. and Junko Mori (eds.), Proceedings of the Fifteenth Conference in Japanese/Korean Linguistics (JK 15), pages 358-369. Stanford, CA: CSLI Publications.

Kaufmann, Stefan. 2005. Conditional truth and future reference. Journal of Semantics 22:231-280.

Under review:

On the Temporal Interpretation of Japanese Temporal Clauses. With Misa Miyachi (University of Chicago).

Free choice and modality.

Jointly with Min-Joo Kim (Texas Tech), I study Korean indeterminates and their interactions with various particles, currently focusing on the Free Choice Items (FCI) nwukwu-na and amwu-na, composed of the indeterminates nwukwu/amwu and the particle -na. These forms are particularly interesting from the perspective of the theory of FCI for two main reasons: First, while both share the general meaning of “free choice,” they exhibit important differences in detail which highlight the multifaceted nature of this category. Second, their relatively transparent morphological structure calls for a compositional analysis which locates the source of the differences between them in the indeterminates nwukwu and amwu and gives a unified analysis to the particle -na. This is a challenge because it means that while the particle is the source of the free-choice flavor of the compounds, it cannot be held responsible for certain properties that are considered typical of FCI in general, yet are not shared by both nwukwu-na and amwu-na. In this paper, we survey the semantic difference between the items and develop a formal analysis of the differences.

Selected publications:

Kim, Min-Joo and Stefan Kaufmann. 2006. Domain restriction in freedom of choice: A view from Korean indet+na items. In Puig-Waldmüller (ed.), Proceedings of Sinn und Bedeutung 11, pages 375-389, Universitat Pompeu Fabra, Barcelona, Spain.

Rational linguistic behavior

This project explores the ways in which speakers' and hearers' intentions and mutual beliefs influence the use and interpretation of linguistic forms (e.g., accent placement and implicatures). In 2005, the “Game Theory and Meaning” project at Northwestern's Department of Linguistics began as a reading group and soon developed its own research agenda. The current core members are Brady Clark (Linguistics) and graduate students James German (Linguistics) and Eyal Sagi (Psychology). Our research focuses on the application of methods from Game Theory to problems in the study of natural language meaning. Recent work (e.g. by Jäger, Merin, Parikh, and van Rooij) has recast pragmatic analyses in Game-Theoretic terms. This theoretical work suggests new experimental paradigms in which to test the predictions of Game-Theoretic models, but experimental research has thus far not kept up with the pace of the theoretical advances. The goal of this project is to close this gap through experimental and agent-based simulation work.

Currently, we focus on two sub-projects investigating the role played by situational factors like mutual beliefs and speakers' intentions in determining (i) the use and interpretation of accent placement, and (ii) Gricean implicatures listeners draw from speakers' utterances.

Selected publications:

German, James B., Eyal Sagi, Brady Clark, Stefan Kaufmann, and Min-Joo Kim. 2006. The role of hearers' beliefs in the interpretation of logical connectives. Second Conference on Games and Decisions in Pragmatics (GDP II), Zentrum für Allgemeine Sprachwissenschaft, Berlin, Germany.

German, James B., Janet Pierrehumbert, and Stefan Kaufmann. 2006. Evidence for phonological constraints on nuclear accent placement. Language 82:151-168.

Data-driven approaches to meaning

My work in computational linguistics focuses on the application of computational methods for extracting semantic information from large text corpora in areas like education, politics, and business. The overarching goal is to test and improve the applicability of methods which have been widely and successfully used in information retrieval, topic detection and related practical applications, for purposes such as the detection of knowledge and conceptual change, opinions and ideology.

Computational evaluation of interview data in science education.

This project grew out of joint work with Gregory Dam (then a graduate student in Learning Sciences), published in Dam and Kaufmann (to appear). There, we applied Latent Semantic Analysis (LSA), a method for deriving a measure of similarity between words and larger units of text from word occurrence patterns in a training corpus, in classifying transcripts of interviews with school children before and after instruction on a science topic (in this case, the explanation for the existence of different seasons on earth). The goal was to determine whether students' responses reflected knowledge of the correct explanation or common misconceptions. Our computational method reached a high level of accuracy with the judgments of human annotators.

Computational tools for this sort of analysis promise to have practical benefits for the development and evaluation of teaching techniques – removing the “bottleneck” currently involved in manually annotating large data sets, and placing the evaluation on a more objective and principled methodolgical base – and at the same time contributes to work in computational linguistics – tackling a new area of application with its own specific challenges, which point to new directions for the development of computational methods. Jointly with Bruce Sherin (Learning Sciences) and David Uttal (Psychology), we have since developed this work into a larger project whose goal is to explore the use of computational linguistics in detecting and tracing conceptual change in education.

Selected publications:

Dam, Gregory and Stefan Kaufmann. 2007. Computer assessment of interview data using Latent Semantic Analysis. Behavior Research Methods 40(1):8-20.

Kaufmann, Stefan. 2000. Second-order cohesion. Computational Intelligence 16:511-524.

Takayama, Yasuhiro, Raymond Flournoy, and Stefan Kaufmann. 1998. Information Mapping: Concept-Based Information Retrieval Based on Word Associations. CSLI, Stanford University.

Linguistic expressions of ideology and opinion.

This joint project with Daniel Diermeier (Kellogg), the postdoctoral researchers Bei Yu and Beata Klebanov-Beigman, and Jean-François Godbout, a graduate student in Political Science, started in Summer, 2006, as part of the Northwestern Institute on Complex Systems (NICO). Its goal is to apply computational methods to the task of classifying text documents according to the opinions expressed in them, rather than the topic they are about. In the political domain, we used a Support Vector Machine (SVM) classifier trained on senatorial speeches from the 101st–107th Congresses to detect the ideological positions of Senators in the 108th Congress based on their speeches. This work was presented at a NICO seminar in February and at the annual meeting of the Midwest Political Science Association in April, 2007, and a journal paper reporting the results is currently under review. More recently, we have begun to apply similar methods to detect the opinions expressed in news articles on corporate entities. In this domain, the goal is to trace negative “issues” as they arise and spread through the media. We are currently preparing a manually annotated corpus of news texts related to Wal-Mart for use in training and evaluating automated analysis tools.

Selected publications:

Yu, Bei, Daniel Diermeier, and Stefan Kaufmann. 2008. Classifying party affiliation from political speech. Journal of Information Technology in Politics 5:33-48.

Yu, Bei, Stefan Kaufmann, and Daniel Diermeier. 2008. Exploring the characteristics of opinion expressions for political opinion classification. Ninth Annual Internation Conference on Digital Government Research (dg.o 2008), Montreal, Canada.