Experimental approaches to language universals in structure and meaning
(ExUni virtual Workshop at ESSLLI2021)

Programme

Here is the programme of talks for the Workshop on Experimental approaches to language universals in structure and meaning which will take part virtually at ESSLLI 2021 in August 2nd-3rd 2021. All times are in CEST.

Abstracts for keynote and invited talks are below.

Keynote & invited speaker abstracts

Dorothy Ahn (Rutgers): Semantic and pragmatic universals in definiteness
Definiteness is a core property of natural language that has been investigated extensively. In this talk, I give an overview of relevant semantic and pragmatic tendencies that have been identified in the literature and show how recent typological and experimental studies from the last two decades have updated our view of how languages encode definiteness. To do so, I will focus on two case studies: a) the marking of familiarity in Mandarin, and b) the use of space to mark different referents in American Sign Language. What we will see from experimental studies is that many of the cross-linguistic patterns observed in marking definiteness result not from hard-wired semantic principles but from more general pragmatic tendencies.

Isabelle Dautriche (Aix-Marseille University): Looking for the roots of semantic primitives in infancy and in non-humans
Human languages share a number of universal architectural features, from atomic meaning properties (e.g., connectedness, Gärdenfors, 2000) to how these meanings combine to generate more complex senses (i.e., compositionality). In this line of work, we investigate experimentally the cognitive origins of these shared features of language in pre-linguistic infants and non-human primates to determine whether these features reflect properties of the language faculty or rather domain-general forces potentially shared across cognitive systems and species.

Jeremy Kuhn (ENS Paris): Iconic biases on logical meaning
In this talk, I contend that looking at iconic signs and gestures can shed light on the cognitive representation of abstract logical meaning. If pre-linguistic cognitive pressures influence semantic typology (i.e. what is attested and what is not), the same pressures should appear in other, extra-linguistic communicative settings, and, in particular, in the interpretation of iconic signs and gestures. We can thus get insights into these cognitive biases by looking at the semantic and iconic typology of sign languages, as well as by looking at the production and interpretation of gestures by non-signers. I explore this hypothesis in two case studies. First, I look at the case of quantification in sign languages. Sign languages, like spoken languages, show semantic variation, but, surprisingly, this variation populates a specific corner of the full typological landscape: distributive concord is common, but negative concord is rare. I argue that these preferences arise from iconic biases — the sign language typology is explained based on what is easy and hard to represent in space. Second, I look at boundarihood of events and objects. Boundarihood has been shown to be involved in a motivated mapping in sign language: telic verbs are associated with gestural boundaries. In a series of experiments, I show that non-signing subjects, too, show an abstract, iconic bias to associate bounded forms with bounded meanings.

Wataru Uegaki (University of Edinburgh): Testing ‘universals’ in meaning-selection correlations in the domain of clausal embedding
In the recent formal semantic literature, there have been several attempts at explaining the selectional restrictions of clause-embedding predicates in terms of their lexical semantics (e.g., Uegaki 2015; Ciardelli & Roelofsen 2015; Theiler et al. 2019; Mayr 2019; Uegaki & Sudo 2019). These studies identify certain correlations between lexical-semantic properties of clause-embedding predicates and their selectional patterns, and propose to explain these correlations by resorting to the notion of logical triviality that is predicted to arise as a result of semantic composition. Although these studies have so far focused primarily on English, the underlying theory predicts that the correlations between lexical semantics and selection are universal, assuming that the underlying logic is universal across human languages. In this talk, I will report on two lines of empirical investigation that are aimed to test the meaning-driven explanations of selectional restrictions, focusing on Uegaki & Sudo’s (2019) analysis of the anti-rogativity of non-veridical preferential predicates. These are: (i) systematic cross-linguistic data collection to examine the meaning-selection correlation; and (ii) experimental investigation of the correlation between semantic and selectional properties across predicates, drawing on a similar study on NPIs by Chemla, Homer & Rothschild (2011).

Dionysia Saratsli (University of Delaware): Pragmatic Effects on the Learnability of Semantic Distinctions: Insights from Artificial Language Learning studies on Evidentiality
Across different languages, prevalent semantic distinctions are widely assumed to be easier to learn, due to the naturalness of the underlying concepts. We propose that pragmatic pressures can also shape the cross-linguistic prevalence of semantic distinctions and offer evidence from evidentiality (the encoding of information source). Cross-linguistically, languages with grammatical evidential systems overwhelmingly encode indirect sources (especially, reported information or hearsay) but very rarely mark direct, visual experience. Conceptually, humans reason naturally about what they see and use linguistic devices to encode visual perception. However, on pragmatic grounds, if one is to encode a single evidential source, reported information is more informative because it is more indirect, potentially unreliable and consequently more marked. Since evidentiality is not grammaticalized in English, we taught it to adults without native language interference in an Artificial Language Learning experiment. Our earlier work has shown that learners preferentially encode reportatives compared to other sources but left open the nature of the asymmetry because of design-specific issues. In our more recent studies, we eliminated these concerns by directly comparing two simple evidential systems, each marking only a single source (visual vs. reportative) and leaving the other source unmarked. Once again, participants learned more easily to mark reportative information sources over visual/direct sources of information. Overall, our results provide support for a pragmatic bias that shapes both the cross-linguistic frequency and the learnability of evidential semantic distinctions.

Yasamin Motamedi (University of Edinburgh): From simultaneous to segmented structure: investigating preferences for manner-path expressions in improvisation and learning
Most languages in the world use segmented, linear forms to express the manner and path properties of motion events. Research into Nicaraguan Sign Language, an emerging sign language, demonstrated the emergence of segmented motion events (in which manner and path are signed separately), from simultaneous productions in earlier cohorts of the language, (in which manner and path components were articulated within the same manual sign). This change over cohorts has been argued to be driven by learning mechanisms available during childhood, further supported by silent gesture work that suggests hearing adult gesturers overwhelming produce simultaneous structures while hearing children produce a much higher proportion of segmented structures. Here, we hypothesise that preferences for simultaneous versus segmented productions are shaped by pressures for semantic transparency and structural simplicity that operate differently during improvisation (where conventions have not yet been established) and in learning. In the absence of conventions, simultaneous articulations allow individual events to be represented iconically and holistically. However, when those articulations are learnt as a system of event-signal mappings, segmented articulations allow users to map relations between meanings onto the structure of the signals they produce (i.e. through recombination of component parts). We test this hypothesis using a set of online artificial gesture learning experiments with that examine adults’ preferences for simultaneous versus segmented gestures in improvisation, and in an iterated learning task in which participants learn previously established conventions over successive generations.

Contact

If you have questions or comments, please email us at syncog.edinburgh[at]gmail.com.