James Kirby (Edinburgh) & Morgan Sonderegger (McGill)

Population dynamics in the actuation of sound change

Sound change arises from the pronunciation variability ubiquitous in every speech community, but most such variability does not lead to change. Hence, an adequate model must allow for stability as well as change. Existing theories of sound change tend to emphasize factors at the level of individual learners promoting one outcome or the other, such as channel bias (which favors change) or inductive bias (which favors stability). Here, we consider how the interaction of these biases can lead to both stability and change in a population setting.

First, we show that while population structure itself can act as a source of stability, both stability and change are possible outcomes only when both types of bias are active, suggesting that it is possible to understand why sound change occurs at some times and not others as the population-level result of the interplay between forces promoting each outcome in individual speakers. We then discuss how this account of actuation may be generalized to at least two other cases where change can occur without production bias: contact between subpopulations, and phonetic variants which bear different levels of prestige. We show how phonetic systems can remain stable under a variety of perturbations, such as lenition, contact, and social prestige, depending on the relative magnitudes of the biases involved. This result suggest a more unified account of the mechanics of actuation, with the crucial factor being the relative magnitude, rather than the specific source, of the displacement event.

Kenny Smith (work with Deborah Kerr, MSc ELC 2014/15)

The Spontaneous Emergence of Linguistic Diversity in an Artificial Language

I will present an experimental paradigm, combining artificial language learning with the Minimal Group method borrowed from social psychology, and demonstrate the spontaneous emergence of linguistic diversity despite the absence of functional pressures for social differentiation.

Longer abstract here

Christine Cuskley

Frequency and Stability in Linguistic Rules

Frequency and stability exhibit an interesting relationship in language: the more frequent a linguistic construction is, the less it tends to change over time. Despite this evident relationship, it is less clear what specific social and cognitive factors cause increased stability in more frequent constructions. This talk will present work which aims to examine the specific factors underlying the frequency-stability relationship using the test case of verb regularity in English: highly frequent verbs are more likely to be irregular, while less frequent verbs tend to destabilise to the regular form. To investigate this, I will present analysis from a historical corpus of English showing that vocabulary growth underlies the most marked increases in regularity over time, and verbs transition from irregular to regular and visa versa within a particular frequency band. In another approach to the problem, an adaptation of the Naming Game sheds light on the dynamics of rules and irregular exceptions across a population of interacting agents. Finally, I present an experiment which contrasts how native and non-native speakers inflect novel verbs to investigate how differences in population structure might affect regularity in a language system. Together, these results help us to better understand what causes the persistence of irregular exceptions to regular rules in language more generally, and how sociolinguistic and demographic processes may effect regularity.

Thursday 14 January, 11:00–12:30
Lecture Theatre 3, 7 Bristo Square

Judith Degen, Department of Psychology, Stanford University

Context in pragmatic inference

In the face of underspecified utterances, listeners routinely and without much apparent effort make the right kinds of pragmatic inferences about a speaker’s intended meaning. I will present a series of studies investigating the processing of one type of inference – scalar implicature – as a way of addressing how listeners perform this remarkable feat. In particular, I will explore the role of context in the processing of scalar implicatures from “some” to “not all”. Contrary to the widely held assumption that scalar implicatures are highly regularized, frequent, and relatively context-independent, I will argue that they are in fact relatively infrequent and highly context-dependent; both the robustness and the speed with which scalar implicatures from “some” to “not all” are computed are modulated by the probabilistic support that the implicature receives from multiple contextual cues. I will present evidence that scalar implicatures are especially sensitive to the naturalness or expectedness of both scalar and non-scalar alternative utterances the speaker could have produced, but didn’t. In this context I will present a novel probabilistic and contextualist account of scalar implicature processing that has roots in both constraint-based and information-theoretic accounts of language processing and that provides a unified explanation for a) the varying robustness of scalar implicatures across different contexts, b) the varying speed of scalar implicatures across different contexts, and c) the speed and efficiency of communication.

Wednesday 13 January, 11:00–12:30
Room G32, 7 George Square

Florian Jaeger, University of Rochester (work with Dan Gildea, Masha Fedzechkina, Lissa Newport, and John Trueswell)

Pressures for processing and communicative efficiency bias language development

Functional biases have been hypothesized to affect language change and explain typological patterns. I’ll focus on two specific pressures on language processing and production that are well-established. The first pressure relates to the fact that linguistic communication takes place in the presence of noise, so listeners need to infer intended message from noisy input —making less probable message harder to infer (e.g., Levy, 2008; Norris & McQueen, 2008; Bicknell & Levy, 2012; Gibson et al., 2013; Kleinschmidt & Jaeger, 2015). The second pressure relates to memory demands during language processing, where longer dependencies are associated with slower processing (Gibson, 1998, 2000; Lewis et al., 2006; Vasishth & Lewis, 2005). Both pressures are well-known to affect language processing, including evidence from both experimental data (e.g., McDonald & Shillcock, 2003; Grodner & Gibson, 2005) and broad-coverage corpus studies (e.g., Demberg & Keller, 2008; Boston et al., 2010; Smith & Levy, 2013). This means that an ideal speaker (in the sense of ideal observers) should a) support low-probability —i.e., high information— messages with ‘better’ linguistic signals to the extent that this is warranted against the effort in implies (e.g., due to aiming for more precise articulations or due to articulation additional words, cf. Lindblom, 1990, Jaeger, 2006, 2013; Gibson et al., 2013) and b) aim for short dependencies (e.g., by reordering constituents, Hawkins, 2004, 2014).

Case study 1 asks whether actual natural languages have syntactic properties that increase processing efficiency, as would be expected if the processing efficiency biases language learning and/or change. Using data from five large syntactically annotated corpora, I show that natural languages have lower information density and shorter dependency lengths than expected by chance (Gildea & Jaeger, in prep; for dependency length, see also Gildea & Temperley, 2010). Previous work has found similar properties for phonological and lexical systems (e.g, Manin, 2006; Piantadosi et al., 2011, 2012; Wedel et al., 2013). The present work is the first to find that the same properties affect even the syntactic system (which involves considerably more complex latent structure and has often been assumed to be encapsulated from functional pressures).

Case studies 2 and 3 employ an miniature language learning approach to the same question. I show that learners of such languages restructure them in a way that improves both the inferability of messages and the dependency length (Fedzechkina et al., 2011, 2013, under review; Fedzechkina & Jaeger, 2015).Unlike approaches that rely on statistical modeling of typological data, miniature language learning does not suffer from data sparsity and can —if applied correctly— assess causality by directly manipulating the relevant factors.

Some references to related work from my lab, available at https://rochester.academia.edu/tiflo/Papers:

Fedzechkina, M., Chu, B., and Jaeger, T. F. submitted. ‘Long before short’ preference in a head-final artificial language: In support of dependency minimization accounts.
Fedzechkina, M., Newport, E., and Jaeger, T. F. accepted for publication. Balancing effort and information transmission during language acquisition: Evidence from word order and case-marking. Cognitive Science.
Fedzechkina, M., Jaeger, T. F., and Newport, E. 2012. Language learners restructure their input to facilitate efficient communication. Proceedings of the National Academy of Sciences 109(44), 17897-17902. [doi:10.1073/pnas.1215776109]
Gildea, D. and Jaeger, T. F. submitted. Language structure shaped by the brain: Human languages order information efficiently.

Cathleen O’Grady (work with Christian Kliesch, Kenny Smith, Thom Scott-Phillips)

The ease and extent of recursive mindreading

Mindreading, also called theory of mind, is the ability to mentally represent the mental states of other individuals, e.g. “Sarah believes in fairies.” Recursive mindreading is the ability to mentally represent mental states that themselves are representations of other mental states, e.g. “Joey knows that Monica thinks that Phoebe doesn’t know that Chandler and Monica love each other.”

Various accounts of pragmatics require not only that interlocutors are able to make inferences about each other’s mental states, but suggest that communicative intentions necessarily require recursive mindreading. However, despite a wealth of research on first-level mindreading and the developmental trajectory of mindreading, the extent of the adult human capacity for recursive mindreading is relatively under-studied. Existing research on the topic suggests that recursive mindreading is both limited and more effortful than other complex memory tasks, but this research suffers from substantial methodological flaws.

Based on the success in developmental studies of using implicit rather than explicit mindreading tasks, we present an implicit test of high levels of recursive mindreading. We show experimentally that adult human recursive mindreading abilities are more advanced than has previously been shown, with high accuracy up to seven levels of embedding. We further show that presentation of the task as implicit rather than explicit improves performance, explaining the difference in results from previous research, and suggesting that a more ecologically valid task is a more accurate test of recursive mindreading capacity. Our results lend support to pragmatic accounts requiring recursive mindreading.

Marieke Schouwstra (work with Kenny Smith and Simon Kirby)

From natural order to convention in silent gesture

Silent gesture, an experimental paradigm in which adult hearing participants describe events using only their hands, has been valuable for investigating the origins of word order. Goldin-Meadow et al. (2008) found a language-independent preference for SOV for extensional transitive events (e.g., boy-ball-throw), but participants prefer SVO for intensional events (e.g., boy-search-ball; Schouwstra & de Swart, 2014).

The SVO/SOV pattern for intensional/extensional events arises independently of participants’ native language, and, we will claim, represents naturalness, reflecting cognitive preferences to put Agents first (Jackendoff, 2002) and more abstract/relational information last. However, existing languages tend not to condition word order on event type and are instead more regular. Understanding this transition from naturalness to conventionalised regularity is a major goal of language evolution research. We present a new approach to this challenge using a novel experimental paradigm in which silent gesture is both used for communication (Christensen et al, 2015) and culturally transmitted through artificial generations of lab participants (Smith et al, in prep).

I will describe four experiments in which participants communicate about intensional and extensional events, either in a dyadic communication or a gradual turnover setup. Our experiments show that in silent gesture communication and transmission, semantically conditioned word order tends to disappear in favour of regular word order. The frequency of event types determines how regularisation progresses. This suggests that where pressures for naturalness and regularity are in conflict, languages start natural, but naturalness will give way to regularity as signalling becomes conventionalised through repeated usage.

Klaas Seinhorst, Amsterdam Center for Language and Communication, University of Amsterdam

Filling in the blanks – acquisition meets typology

It has often been observed that languages disprefer gaps in their phoneme inventories: they tend to maximally combine their distinctive features. In the late 1960s, the French phonologist André Martinet suggested in passing that this tendency may be rooted in the acquisition process, but so far empirical evidence has mostly been lacking. I present exactly such evidence from experiments with human learners (n = 96), who were exposed to one of eight category structures. These category structures have one binary feature and one ternary feature: as such, they resemble plosive inventories from spoken language, which often have a binary voicing opposition ([±voice] or [±tense]) and a three-way place of articulation contrast (usually [labial] vs. [coronal] vs. [dorsal]).

Given these features, a type I language could have the plosives /p t k/; a type VIII language has either /p t k b d g/ (like English) or /p ph t th k kh/; a type VI language might have /b p d k/. Types I and VIII are predominantly found in the world’s languages.

The results of the experiments show that participants often regularise their input, i.e. they (quite literally) fill in the blanks: for instance, type VI is often classified as type VIII. Consequently, learners considerably reduce the cumulative complexity of the data set. These outcomes, then, seem to support two related hypotheses: (i) biases in pattern learning favour regular systems; (ii) the low degree of complexity that has been attested in spoken languages may be ascribed to the iterated effect of such biases.

Alan Nielsen

Systematicity, contrastiveness, and learnability: Evidence from a growing lexicon experiment

Typically, experiments exploring the degree to which systematicity and motivatedness compare the learnability of complete artificial lexica to one another, concluding that associations between words and meanings that are systematic create learnability penalties at certain sizes. That is, systematic associations between words and meanings, whether they are motivated or not, necessarily create artificial lexica that are more constrained, i.e. words are sufficiently similar to one another that they are easily confused.

In this talk I present the results of an experiment using a paradigm designed to explore this learning penalty directly by teaching participants an artificial language over time, allowing for a comparison between how well words are learned at first exposure, compared to how well they are learned after similar labels enter the lexicon.

Additionally, the experimental design allows for us to test a simple version of the bootstrapping hypothesis, which suggests that the acquisition of motivated tokens bootstraps the acquisition of later non-motivated tokens.

We find support for the first of these hypotheses: as signal spaces become increasingly saturated, individual labels within those lexica become increasingly difficult to learn. We do not, however, find support for the bootstrapping hypothesis.

Simon Kirby (work with Tessa Verhoef and Carol Padden, UCSD)

Naturalness and Systematicity: Evidence from artificial sign language

Language is shaped by cognitive biases. These biases can influence the emergence of linguistic structure through multiple linking mechanisms: improvisation of novel solutions to communicative tasks; repeated interaction between communicating individuals; and iterated learning of linguistic conventions over multiple generations. The study of language evolution seeks an explanation of the origin of linguistic structure in terms of these processes and their interaction with human cognition.

In this talk I will concentrate on two types of bias that shape the emergence of linguistic structure: naturalness, and systematicity. Both are the result of a domain general preference for simplicity, but differ in how this simplicity is realised. Naturalness is a property of the relationship between linguistic form and something non-linguistic, whereas systematicity is a property of the relationship among linguistic elements. I will argue that the former type of bias is felt most strongly during improvisation, whereas the latter is felt most strongly during learning.

To test this, we look at a formal feature that governs the lexicon of sign languages: the so-called instrument vs. handling distinction. This has been recently argued to exhibit “patterned iconicity”, a property that combines both naturalness and systematicity. We show in series of artificial sign language experiments online and in the lab that even participants who have never been exposed to sign languages are sensitive to this feature. We show a very strong naturalness bias for instrument forms to be matched with objects and handling forms to be matched with actions. However, the naturalness bias that favours an iconic relationship between form and meaning can be overturned after a small number of exposures to an anti-iconic artificial sign language. This shows that although naturalness may be important in the early stages of language emergence, it is systematicity that is the driving factor where sets items are being learned and transmitted.