Category: Talks

Learning (im)possible number syncretisms: investigating innate featural representations

Naomi Lee, Department of Linguistics, New York University

Friday, 17.07.2020
15:30 – 16:30
Room: [virtual Zoom talk]

Prominent feature-based theories of number, primarily Harbour (2014, 2016) and Smith et al. (2019), assert that dual and plural form a natural class to the exclusion of singular, a prediction borne out in the cross-linguistic prevalence of dual-plural syncretisms. The differently structured theories make opposing predictions for the natural class membership of the singular value, with Harbour ruling out singular- plural (ABA) syncretism and Smith et al. excluding singular-dual (AAB) syncretism. However, a survey of 30 singular-dual-plural languages reveals that typological evidence for either pattern is relatively sparse.

I instead test these predictions experimentally using artificial language learning. 149 adults were trained on one of three fragment grammars: all had three number distinctions on nouns, but each displayed a different logically possible syncretism in verbal number agreement. Participants were asked to select the correct cardinality of referent on the basis of stimuli containing only the syncretic verbal forms. Learners of the dual-plural (ABB) syncretism performed significantly better than chance (p < 0.001), which bears out the common prediction of all discussed theories, thereby supporting the general relevance of feature- derived natural classes in morphological learning. Additionally, the significant relative disadvantage observed for the singular-dual (AAB) syncretism (p < 0.001) suggests specific support for a Smith et al.-like system where singular and dual are not a natural class.

The Grammar of the Body and the ABCs of Language

Wendy Sandler, Sign Language Research Lab, University of Haifa

Tuesday, 30.06.2020
11:00 – 12:00
Room: [virtual Zoom talk]

Sign languages teach us that humans have the capacity to create language in two different physical modalities. What can we learn from this dual capacity? In this talk, I support the idea that the body-to-grammar correspondence of sign languages shows us the ABC’s of language form and emergence. Sign languages allow us to see directly the Analytic properties shared with spoken languages. By tracking the Body-grammar correspondence in young sign languages, we witness the emergence of linguistic form. Different types of communities show the effect of Culture on this emergence path. And finally, Culture, in the form of sign language theatre, spotlights the range of linguistic expression by stretching it to its limits and beyond.

Complexity/informativeness trade-off in the domain of indefinite pronouns

Milica Denic, Universiteit van Amsterdam

Tuesday, 23.06.2020
11:00 – 12:00
Room: [virtual Zoom talk]

The vocabulary of human languages has been argued to support efficient communication by optimizing the trade-off between complexity and informativeness (Kemp & Regier, 2012). The argument has been based on cross-linguistic analyses of vocabulary in semantic domains of content words such as kinship, color, and number terms. The present work extends this analysis to a category of function words: indefinite pronouns (e.g. someone, anyone, no-one, cf. Haspelmath, 2001). We establish the meaning space and feature-based representations for indefinite pronouns, and show that indefinite pronoun systems across languages optimize the complexity/informativeness trade-off. This demonstrates that pressures for efficient communication shape both content and function word categories, thus tying in with the conclusions of recent work on quantifiers (Steinert-Threlkeld, 2019). Furthermore, we argue that the trade-off may explain some of the universal properties of indefinite pronouns, thus reducing the explanatory load for linguistic theories.

Assessing integrative complexity as a predictor of morphological learning in the presence of phonological cues

Tamar Johnson, CLE, The University of Edinburgh

Tuesday, 16.06.2020
11:00 – 12:00
Room: [virtual Zoom talk]

Morphological paradigms differ widely across languages: some feature relatively few contrasts, and others, dozens. Under the assumption that languages evolve to maximise their learnability and that simpler systems are generally easier to learn (e.g. Chater & Vitanyi, 2003), this variation is surprising. Recent work on morphological complexity has resolved this paradox by arguing that certain features of even very large paradigms make them easy to learn and use. Specifically, Ackerman and Malouf (2013) propose an information-theoretic measure, i-complexity, which captures the extent to which forms in one part of a paradigm predict others. They contrast this measure with e-complexity, which is commonly used as a measure of morphological complexity in the literature (e.g., Bickel & Nichols, 2005) and captures the number of distinctions made by the language and the different ways to mark each grammatical function.

In this talk I will present our findings showing that in morphological systems with no semantic and phonological cues for class membership, e-complexity is a better predictor of the learnability of an inflectional paradigm than i-complexity, both in neural networks and human learners. Moreover, in human learners, we find only weak evidence (if any) that low i-complexity paradigms are easier to learn. Based on previous studies showing that human learners need redundant phonological or semantic cues in addition to distributional cues to form classes (Frigo and McDonald, 1998; Gerken, Wilson & Gomez, 2009), we test the hypothesis that in the presence of phonological cues for class membership, low i-complexity facilitates learning the inflectional system. I will discuss the method used in this study and present preliminary results.

The learnability consequences of Zipfian distributions

Ori Lavi-Rotbain and Inbal Arnon, Hebrew University, Jerusalem

Tuesday, 9.06.2020
11:00 – 11:40
Room: [virtual Zoom talk]

One of the striking commonalities between languages is the way word frequencies are distributed. Across languages, word frequencies follow a Zipfian distribution, showing a power law relation between a word’s frequency and its rank (Zipf, 1949). Intuitively, this means that languages have relatively few high-frequency words and many low-frequency ones. While studied extensively, little work has explored the learnability consequences of the greater predictability of words in such distributions. We propose such distributions confer a learnability advantage for word segmentation, a foundational aspect of language acquisition. We capture the greater predictability of words using the information-theoretic notion of efficiency, which tells us how predictable a distribution is relative to a uniform one. We first use corpus analyses to show that child-directed speech is similarly predictable across fifteen different languages. We then experimentally investigate the impact of distribution predictability on children and adults. We show that word segmentation is uniquely facilitated at the predictability levels found in language, compared both with uniform distributions and with skewed distributions that are less predictable than those of natural language. We further show that distribution predictability impacts learning more than distribution shape, and that learning is not improved further in distributions more predictable than natural language. These novel findings illustrate learners’ sensitivity to the overall predictability of the linguistic environment; suggest that the predictability levels found in language provide an optimal environment for learning; and point to the possible role of cognitive pressures in the emergence and propensity of such distributions in language.

Linguistic Prescription, Ideological Structure and the Actuation of Linguistic Changes: Grammatical Gender in French Parliamentary Debates

Heather Burnett and Olivier Bonami, Université de Paris, LLF, CNRS

Tuesday, 2.06.2020
11:00 – 11:40
Room: [virtual Zoom talk]

In this presentation, we give a new study of the role that social meaning and speaker ideologies play in variation and change in g(rammatical) gender in French. More specifically, we study noms de métiers et de fonctions ‘professional nouns’ which have the following g-gender assignment pattern: when they are used to refer to socially female individuals, they can have either masculine or feminine g-gender (i.e. le ministre or la ministre for a female minister); whereas, when they are used to refer to socially male individuals, they can have only masculine g-gender (only le ministre for a male minister). We present the first quantitative study of the linguistic and social factors that condition the use of masculine vs feminine g-gender with reference to women, focusing on variation in the transcripts of the debates of the Assemblée Nationale (AN, French House of Representatives).

The use of grammatical gender in expressions referring to women has been the subject of enormous amounts of prescription and language planning in France and within the Assemblée Nationale itself (see Houdebine 1987, 1998, Burr 2003, Viennot 2014, among others). These efforts can be naturally divided into two phases of activism: First, in 1986, Prime Minister Laurent Fabius legislates the use of feminine grammatical gender and (certain) feminized forms in the AN and similar government institutions. However, we show in our data that this prescription had little to no effect on the speech of the politicians at the time (see also Yaguello 1989, Brick & Wilks 1994 for qualitative observations). Second, in 1998, Prime Minister Lionel Jospin issues a statement reiterating Fabius’ policy. We show that, unlike 12 years earlier, use of the feminine form (eg. la ministre) successfully replaces use of the masculine form (eg. le ministre) within the space of a year in the AN. This striking difference raises the question: What changed from 1986 to 1998 which allowed the feminine form to take over, possibly aided by (the exact same) language policy?

Our main proposal in this talk is that changes in the use of feminine grammatical gender and differences in the effectiveness of Fabius/Jospin’s language policy are the result of changes in social gender ideologies that occurred in France between the mid 1980s and mid 1990s. To make this claim maximally explicit, we develop a formal model of the relationship between ideological structure and language use and interpretation based on current work in game theoretic pragmatics (particularly Franke 2009 and Frank & Goodman 2012). More specifically, we use Gärdenfors (2000, 2014)’s Conceptual Spaces framework to formalize speaker ideologies and Burnett (2017)’s Social Meaning Game framework to capture the link between ideological structure, social meaning and language use. Using this model, we show that the failure of Fabius’ policy and Jospin’s subsequent successful use of this policy is predictable from independently motivated assumptions concerning 1) the social meaning of French g-gender (following Livia 2001, McConnell-Ginet 2013), and 2) changes in social discourses surrounding the properties of female politicians associated with the Parité (‘equal representation’) debates in the late 1990s (Ramsay 2003, Scott 2007, Julliard 2012, among others). We therefore conclude that tools from formal semantics and pragmatics can be helpful to understanding both the relationship between social change and linguistic change, and the conditions under which language policies can be effective.

References:

Brick, N., & Wilks, C. (1994). Et Dieu nomma la femme: observations sur la question de la féminisation des noms d’agent et sur les désignations d’Edith Cresson dans la presse. Journal of French Language Studies, 4(2), 235-239.

Burnett, H. (2017). Sociolinguistic interaction and identity construction: The view from game‐theoretic pragmatics. Journal of Sociolinguistics, 21, 238-271.

Burr, E. (2003). Gender and language politics in France. Gender across languages, 3, 119-139.

Frank, M. C., & Goodman, N. D. (2012). Predicting pragmatic reasoning in language games. Science, 336, 998-998.

Franke, M. (2009). Signal to act: Game theory in pragmatics. PhD Thesis. Institute for Logic, Language and Computation, University of Amsterdam.

Gärdenfors, P. (2014). The geometry of meaning: Semantics based on conceptual spaces. MIT Press.

Gärdenfors, P. (2000). Conceptual spaces: The geometry of thought. MIT press.

Houdebine, A. M. (1987). Le français au féminin. La linguistique, 23, 13-34.

Houdebine, A. M. (1998). La féminisation des noms de métiers. Paris: Harmattan.

Julliard V. (2012). De la presse à Internet, la parité en questions, Paris, Lavoisier.

Livia, A. (2001). Pronoun envy: literary uses of linguistic gender. Oxford University Press.

McConnell-Ginet, S. (2013). Gender and its relation to sex: the myth of ‘natural’gender. in Corbett, G. (ed). The Expression of Gender, Berlin: de Gruyter. 3-38.

Ramsay, R. L. (2003). French women in politics. Berghahn Books.

Scott, J. W. (2007). Parite!: Sexual equality and the crisis of French universalism. University of Chicago Press.

Viennot, É. (2014). “Non, le masculin ne l’emporte pas sur le féminin.” Petite histoire des résistances de la langue française. Éditions iXe.

Yaguello, M. (1989). Le sexe des mots. Belfondl.

All-or-nothing tells us little about language evolution

Pedro Tiago Martins, Universitat de Barcelona / UBICS

Tuesday, 26.05.2020
11:00 – 11:40
Room: [virtual Zoom talk]

Language is a biological capacity unique to humans. A major topic of research, especially in recent years, is understanding how this capacity evolved. A prominent hypothesis posits that language has at its core a particular computational operation, which emerged suddenly, as the result of single mutation. The justification for this evolutionary scenario is that this capacity is either present or absent, with no intermediate steps, and as such it could not have evolved through intermediate evolutionary steps. In this talk I will offer a critical look at theories of this kind, arguing that this “all-or-nothing” conception of the evolution of language in general and some of its subcomponents in particular is not productive. Vocal learning will be given special attention as one such case of a sub-component of language which benefits from a more permissive, non-dichotomical view regarding its biological nature and evolution.

Cognitive Novelties, Informational Form, and Structural-Causal Explanations

Andrew Buskell, Department of History and Philosophy of Science, University of Cambridge

Tuesday, 19.05.2020
11:00 – 11:40
Room: [virtual Zoom talk]

Over the last thirty years, an increasing number of researchers have argued that cultural evolution is important for understanding human cognition and evolutionary history. Their claims range from the mundane to the surprising. One of the more radical holds that central cognitive capacities—at least language, mindreading, and imitation—result from developmental interaction with cultural traditions. Researchers like Kim Sterelny, Cecilia Heyes, and Dan Dennett have put forward adaptationist explanations to make plausible these surprising claims. Here I argue for a complementary project based around structural-causal explanations. Analogous to those found in evolutionary developmental biology, these explanations highlight the importance of structured relationships among informational elements. Focusing on how these end up mirrored in cognition, I argue that structural-causal explanations flesh out adaptationist explanations; they give a way of understanding how developmental interaction with culturally evolved bits of information can change cognitive functioning.

The Role of Simplicity in the Word Order Harmony Bias

Danielle Naegeli & Rachel Kindellan, CLE, University of Edinburgh

Tuesday, 12.05.2020
11:00 – 11:40
Room: [virtual Zoom talk]

Previous research has demonstrated that the typologically attested preference for word order harmony, consistent order of dependents relative to their head, is reflected in individual learning behaviour in artificial language learning. To address the hypothesis that the word order harmony bias is driven by a more general bias for simplicity, we compared the strength of the word order harmony bias in one- and two-modifier contexts. It was predicted that the harmony bias would be stronger in the one-modifier conditions than in their two-modifier counterparts given that in one-modifier conditions the harmonic grammar is simpler than non-harmonic grammar, whereas in two-modifier phrases they are equally simple. Participants did not exhibit a harmony bias in any condition based on proportion of majority order production in the one-modifier nor in the two modifier contexts. Shannon entropy was used to measure the variation the output languages. This analysis revealed that there was no significant effect of the interaction between number of modifiers and harmony on regularisation. These unexpected findings prompt further exploration of the word order harmony bias; contexts in which it is active, other linguistic factors that influence it, and alternative paradigms to use in its investigation.

Acquisition of grammatical and sociolinguistic variation: Evidence from artificial language learning

Betsy Sneller, Georgetown University

Tuesday, 21.04.2020
14:00 – 15:00
Room: [virtual Zoom talk]

When faced with inconsistent input, children are found to be master generalizers (Kerswill 2003; Senghas & Coppola 2001; Singleton & Newport 2004). However, natural languages also contain meaningful variation, which a speaker must also acquire in order to be a sociolinguistically competent speaker. Some studies within sociolinguistics suggest children acquire phonological variation at fairly young ages, while others suggest that variation requires additional age or exposure (Smith, Durham & Fortune 2007; Miller 2013; Hendricks, Miller & Jackson 2018).

Here I present an artificial language study designed to test the role of age and linguistic conditioning in the acquisition of phonological variation. In natural languages, variation is typically conditioned by language-internal (grammatical) factors as well as external (social) factors. In these experiments, we test how children of different ages acquire grammatically conditioned and socially conditioned variation. We find that the youngest children show a strong tendency to regularize their inconsistent input, while older children begin to exhibit the conditioning given in the input. However, we also find an effect of the type of conditioning, with socially-conditioned variation acquired more readily by young children than grammatically-conditioned variation. These findings suggest that both speaker age and the type of conditioning both play a role in the acquisition of variable phonology.