Florian Schwarz

(full list of papers and talks with download links here)

Meaning in Context: formal and cognitive perspectives

When encountering linguistic utterances, our minds face a juggling act of sizable proportion, as they have to integrate a variety of sources of information to piece together the overall message that is being conveyed. This ranges from abstract grammatical knowledge to many different dimensions of the speech signal that guide the processor in decoding it, as well as consideration of the speaker’s intentions and a wealth of information concerning the utterance context, both linguistic and non-linguistic. While we experience this process, for the most part, as effortless and thus can focus on the overall content of the message conveyed, much recent work in the scientific study of language has begun to untangle its many ingredients and to better understand their interplay. In the study of meaning, two related areas of particular interest are the rich variety of different aspects of meaning and the integration of contextual information into the interpretation process.

My research program investigates foundational issues in these areas by combining formal tools from linguistic semantics and pragmatics with experimental methods from psycholinguistics. The empirical and theoretical components in this work crucially inform one another. Careful experimental manipulation is key for identifying core properties of different aspects of meaning, such as presuppositions and implicatures. Furthermore, insights into the actual cognitive processes involved in language comprehension in context help to constrain theoretical options in light of what is cognitively real. At the same time, theoretical proposals are a key source for hypotheses about what cognitive representations might be at play and can be uncovered us- ing online processing measures such as eye tracking. Ultimately, this two-pronged approach to investigating language comprehension and pragmatic inferencing in context serves to settle one of the central questions in the cognitive science of language for any given phenomenon, namely whether it results from conventional encoding and computation of representations specific to linguistic knowledge, or from domain general cognition and reasoning about language use in context. In the following, I sketch some core areas of my research that contribute to this overall research enterprise.

Presuppositions & Processing

Presuppositions, a sub-type of meaning consisting of back- grounded content that is typically taken for granted, provide excellent testing grounds for the interplay of contextual and linguistically encoded information and their integration in online processing. On the one hand, many of their aspects lend themselves to domain-general explanations based on fundamental aspects of human communication. On the other hand, they exhibit highly complex interactions with the linguistic contexts and structures they appear in, in particular in their characteristic ‘projection’ behavior (reflected in surviving embedding under various entailment-canceling operators).

After initial work opening up first directions for exploring presupposition processing , I have pioneered several new directions in this now vibrant area of research (see my edited volume and a more recent survey in a handbook article. ) First studies using visual world eye-tracking reveal presupposed information to be rapidly accessed. However, in embedded positions, where they exhibit their hallmark projection behavior, joint work with Sonja Tiemann detected slight delays in processing, which we interpret as corresponding to the added complexity in discourse representations, as posited in theoretical frameworks such as DRT, at the cognitive level. Another central line of research in this area has been concerned with the potential need to differentiate different types of presupposition triggers, which I have investigated both in terms of online processing and their theoretical profiles.

More recent work has turned to the question of the role of linear order for presupposition projection phenomena, which exhibit well-known asymmetries, relating to whether material supporting a presuppositional expression linearly precedes or follows it. The nature and source of these asymmetries remains contested. One view is that they result from superficial aspects of language use unfolding in time; alternatively, they could be directly encoded at the level of linguistic representations governing the integration of linguistic and contextual information. While influential proposals in the recent theoretical literature on projection provide a detailed account of the role of the time-course associated with comprehending language ‘from left to right’, relatively little remains known about the real-time cognitive processes involved in comprehending presupposition projection. Our experimental findings on projection from conjunction argue that the observed linear order asymmetries need to be encoded in the linguistic representation proper, rather than just resulting from linear order and the temporal unfolding of linguistic material in comprehension. This conclusion constitutes a case where preferring the view that the relevant interpretive properties are linguistically encoded over other, more domain-general (and conceptually appealing) approaches is forced upon us by the experimental data. Furthermore, there are important methodological and conceptual ramifications for purely pragmatic approaches to projection, which are hard to reconcile with our data. In addition to extending these approaches to more direct measures of online processing, one key direction of ongoing work is to extend this paradigm to other embedding environments to further assess potential lexical variation with regards to linear order effects in projection, which would further bolster the case for lexical encoding.

Varieties of Meaning

Beyond presuppositions, there is a wealth of dimensions of meaning coming into play in language use. While some of these have already received a lot of attention, there are also many new directions to be pursued. Scalar implicatures have been the focus of much of experimental pragmatics going on two decades, yet much remains controversial. One line of my own work on these implicatures explores their processing properties in contrast to presuppositions, with a particular focus on so-called indirect implicatures , as well as looking across various populations (e.g., in acquisition and Broca’s Aphasics ). While the inclusion of indirect implicatures, which allow for a more direct comparison with presuppositions in terms of the details of the experimental paradigm, makes some of the contrast between the two types of meaning less clear-cut, the evidence on the whole best aligns with a traditional view where their nature and source continue to be distinct. In terms of online processing, work extending the covered box paradigm (a picture selection task with some of the displayed items hidden behind a black box, shown to be highly suitable for detecting non-dominant interpretations of various types of expressions) to visual world eye tracking has shed new light on the relative inaccessibility of so-called literal interpretations (that do not factor in the relevant scalar implicature).

The experimental study of other dimensions of meaning is still just in its beginnings. Some relevant lines of work include the study of so-called implicated presuppositions (or anti- presuppositions), which seem to exhibit a mix of properties of implicatures and presuppositions, but may well form their own kind of meaning. In ongoing joint work with Nadine Bade , we have studied key examples such as the anti-uniqueness inferences of indefinites (e.g., ‘A sun’ giving rise to the sense that there is more than one (relevant) sun, due to the avoidance of ‘the’, associated with a uniqueness presupposition), and found them to be much more elusive in experimental settings than the theoretical literature would lead us to expect, and crucially different from implicature controls. Another joint project in my lab, led by Muffy Siegel , has assessed certain types of relevance implicatures, where the inferred meaning is directly contradictory to the literal meaning, as in ‘I’m not suggesting that you’re going too slowly, but…’, which pragmatically conveys that you ARE going to slowly. We uncovered a variety of factors impacting the prevalence of these pragmatic inferences, most importantly whether or not participants have access to the verbatim information of the earlier utterance at the time they have to assess its meaning. The fact that the absence of verbatim information increases pragmatic response rates has direct implications for the mental representations stored in memory, with pragmatic inferences seemingly outlasting inconsistent verbatim information. Finally, new work with Andrea Beltrama investigates the impact of social information on pragmatic interpretation effects, namely in the realm of imprecise interpretations of numerals. We find that the social construct of the persona of the speaker, as conveyed by their visual representation (e.g., as ‘nerdy’ or ‘chill’), impacts the extent to which comprehenders grant ‘pragmatic slack’ (e.g., associating an utterance of ‘It costs $300’with a depiction of an actual price of $292), as well as the time course of responses as measured in reaction times. This work opens up promising new avenues for studying pragmatic aspects of interpretation in relation to an even wider range of contextual factors.

Reference, Quantification & Domain Restriction

A third area of my research is concerned with the interpretation of noun phrases in context, again using both theoretical and experimental approaches. In my dissertation work , I investigated different types of definites in German, and spell out an analysis in terms of different relations to the context, distinguish- ing between anaphoric forms, which formally relate back to a linguistic antecedent, and forms requiring contextual uniqueness. Both approaches have a rich tradition in the literature, and I argue that they are both right – for certain forms, which can be morphologically distinguished. This line of work has led to a fairly extensive, and continually growing, literature on definites cross-linguistically.

One key aspect of my analysis of uniqueness-based definites consists of the role of situations relative to which nominal descriptions are evaluated. Technically, this is implemented by positing situations pronoun inside of the syntactic structure of the noun phrase. There are various ways for providing a value for these pronouns, including through a topic situation, which can be construed based on the formal notion of a question under discussion. This provides a framework for the integration of contextual information into the interpretation process. I have explored the predictions of this account experimentally by investigating the interpretation of definite descriptions relative to displays containing multiple candidates for reference, in varying linguistic contexts. Both online processing data from visual world eye tracking and response data reveal that the discourse structure as established by the linguistic context affects the extent and the time-course of which referents are considered when interpreting a definite description, in ways that align well with the situation-based framework.

Yet another important factor affecting domain restriction, in particular for quantificational noun phrases, is that of presuppositional elements in the sentence itself. In a number of studies on both children and adults , we investigated the interplay of domain restriction and presupposition projection from various quantifiers. While presuppositional constraints make relevant domain restriction options particularly accessible, careful experimental manipulation can help to tease apart whether a given interpretive choice results from one or the other, thereby crucially informing the debate on the quantificational force of projection from the scope of quantifiers (whether it is universal or existential).

Research Methodology

As witnessed by recent developments in the field, the integration of experimental and theoretical approaches is key for making progress in understanding both the formalisms underlying our knowledge of linguistic structures and their use, as well as the cognitive processes involved in online interpretation. My work has developed and enhanced various new research methods for studying linguistic comprehension in context. This includes a variety of picture matching tasks for both behavioral data and visual world eye tracking , as well as custom-tailored solutions for a given phenomenon, such as assessing a single feedback utterance from a ‘guide’ in a decoy lexical decision experiment in a post-experiment survey. At a more general level, PCIbex, a free and open-access tool for online experiment implementation and data collection, developed under the lead of Jeremy Zehr in my lab, is now widely used by researchers in linguistic meaning and beyond. While especially relevant in current times with likely longer term limitations for in-lab research due to the COVID-19 crisis, another more general and valuable addition of tools such as these is that they allow for the relatively easy collection of data on a wide range of languages, thereby making it easier to conduct cross- linguistic experimental research, and thus allowing for the integration of two of the most active research areas in the study of meaning. Finally, these free and easy to use tools are also of great use in teaching and student research, as they provide free access to research tools that used to only be available in laboratory settings, and greatly facilitate sharing of experimental stimuli and implementations in the spirit of open science to counter the replication crisis. I am committed to further supporting and promoting the development of tools such as these in order to contribute to the continued integration of empirical and theoretical research in the cognitive science of meaning.