William Croft - The relationship between grammar and language in an evolutionary model
University of Manchester - University of New Mexico
The standard formal syntactic view of a language and a grammar is that the best model of the human language capacity is as a formal grammar that generates an infinite number of sentences (the language), and does not generate other sentences (that is, there is a sharp line between a sentence that is in the language and one that is not). But a real-world language is actually finite: it is the population of utterances produced by members of the speech community. The population of utterances and the speech community are temporally bounded individuals in Hull's (1988) sense: they arose at a point in space and time and will end by their extinction or splitting. We think a language has to be infinite because of what meanings a language is intended to communicate (it is a general-purpose communication system).
A standard formal grammar cannot generate all and only the population of utterances in any interesting way. A real-world grammar that is, what is actually in the head of speakers, is a mental structure developed inductively on the basis of a finite sample of the language, namely the utterances to which the speaker has been exposed throughout her lifetime. This grammar allows the speaker to produce new utterances that she or anyone else may never have produced before, but not necessarily like a standard formal grammar. Instead, a plausible grammar for a real language must allow for frequency-driven, probabilistic interactions between the construction schemas in a speaker's mind that are employed to communicate a meaning in a particular conversational interaction. It is these processes that are of at least as much interest in understanding the nature of grammar as its apparent infinite generative capacity.
There is an interplay between the changing population of the language - the utterances being produced at any time slice - and the grammars - the speaker's knowledge about (not of!) their language at the same time - that leads to the emergence of linguistic structures over the time period of the population of utterances. This process is best modeled not by a formal language system, but by an evolutionary framework that models the replication of structures in finite populations (Hull 1988; Croft 2000, 2005).
- Croft, William (2000). Explaining language change: an evolutionary approach. London: Longman.
- Croft, William (2005). The origins of grammaticalization in the verbalization of experience. Available here.
- Hull, David L. (1988). Science as a process: an evolutionary account of the social and conceptual development of science. Chicago: University of Chicago Press.
Joachim De Beule and Benjamin Bergen - Emerging argument-structure constructions through communicative interactions
AI Lab Brussels - University of Hawaii
All human languages have ways to encode which participants in an event play which roles, whether through word order, case-marking, or argument-verb agreement. In principle, it would be possible to express role bindings of a finite set of scene types through completely lexical means - one word per set of arguments plus verb and their bindings (complex lexical items). Why instead do all languages use constructions that operate over simple lexical items and enforce bindings? One possibility, derived from the Universal Grammar paradigm, is that humans are endowed with innate, and specifically linguistic, capacities, including a set of fundamental predisposition that facilitate the expression of argument structure. However, an alternative explanation proposes that the use of argument structure constructions is a consequence of requirements imposed by communication tasks. In particular, this explanation proposes that when a population of agents is of a particular size, and describes a world with particular properties, maintaining complex lexical items for all possible events becomes, driving agents to a factored, constructional approach.
We test whether expressive and communicative requirements drive agents to greater use of argument structure constructions through a series of experiments with simulated agents engaging in communicative interactions. The agents in a language community play a language game in which they each observe a set of events. One of the agents is the speaker in each trial, and selects one of the scenes to describe. It verbalizes the scene and the hearer points to the event that it thinks was indicated by the language expressed. The game is successful when the hearer points to the scene intended by the speaker.
The results from these experiments demonstrate first that as the complexity of the task increases, so does the likelihood that the population will converge on common argument structure constructions – for example, when they are required to make subtle role-binding distinctions, such as differentiating a scene where John kissed Mary from one where Mary kisses John, they are more likely to evolve argument structure constructions. Second, as the communicative agent population size increases, so does the speed with which the community turns to argument-structure constructions. Larger populations results in faster recruitment of argument structure constructions because of the greater need in such cases to have shared, factored language structures. Third, we observe that as the complexity of the world increases, so does the rate at which a population converges on a common set of argument structure constructions, for the same reason that increased population size yields greater construction use. Finally, our simulations suggest that a population communicating about simple scenes will evolve common tools for encoding argument-structure under almost all conditions, including those most likely to have occurred during the beginnings of human language evolution.
These experiments demonstrate that argument structure evolves automatically in populations of agents communicating about types of scenes with sufficient complexity. Universal aspects of human grammar, it seems, need not be hard-wired, but may be emergent characteristics of communicative systems used in the appropriate cognitive, social, and physical environment.
Back to schedule - Back to top
Holger Diessel - A usage-based approach to grammatical development
University of Jena
This paper is concerned with the usage-based approach to language acquisition. The paper draws on recent work on the acquisition of relative clauses to characterize the approach. It is argued that the acquisition of relative clauses involves the construction of a network of relative clauses. Children acquire this network of constructions in a piecemeal bottom-up fashion. The development originates from subject-relative clauses that are similar to simple sentences and it ends with genitive relative clauses that are structurally and semantically distinct from all other relative clause constructions. The paper argues that the acquisition process is driven by multiple factors: the frequency of the relative construction in the ambient language, the information structure of the sentence in which the relative clause is embedded, the pragmatic function of the construction, and, most importantly, the similarity between the various relative clauses and other constructions in mental grammar.
Back to schedule - Back to top
Pat Healy - Graphical Language Games: Interactional Constraints on Representational Form
Queen Mary College of London
The emergence of shared symbol systems is considered to be a pivotal moment in human evolution and human development. These changes are normally explained by reference to changes in people's internal cognitive processes. I will present two experiments which provide evidence that changes in the external, collaborative processes that people use to communicate can also affect the structure and organisation of symbol systems independently of cognitive change. I propose that mutual-modifiability -- opportunities for people to edit or amend each other's contributions -- is a key constraint on the emergence of complex symbol systems. I will discuss the implications for models of language development and the origins of compositionally.
Back to schedule - Back to top
Vittorio Loreto - Complex Systems Approaches to Semiotic Dynamics
University of Rome
What processes can explain how very large populations are able to converge on the use of a particular word or grammatical construction without global coordination? Answering this question helps to understand why new language constructs usually propagate with a rather sudden transition towards global agreement. In this talk I will discuss a class of microscopic models of communicating autonomous agents performing language games without any central control. These systems undergo a disorder/order transition, going trough a sharp symmetry breaking process to reach a shared set of conventions. Before the transition, the system builds up non-trivial scale-invariant correlations, for instance in the distribution of competing synonyms, which make the system ready for the transition towards shared conventions. If observed on the time-scale of collective behaviors, the transition becomes sharper and sharper with system size. These results not only help explaining why human language can scale up to very large populations but also suggest ways to optimize artificial semiotic dynamics and design new technologies that support or orchestrate self-organizing communication systems. As a concrete example I will discuss the case of a recent social tagging systems for the web.
Back to schedule - Back to top
Luc Steels - Fluid Construction Grammars and Embodiment
AI Lab Brussels - Sony CSL Paris
Very substantial progress has been made recently in computational and robotic experiments that show how lexical communication systems might emerge in situated interactions among embodied agents. For the emergence of grammar, there are some promising initial results but much more research is needed. More specifically, we need powerful linguistic formalisms that can support emergent grammars and learning operators that can give rise to interesting grammar dynamics. We also need powerful ways for conceptualising and interpreting meanings by agents embodied in the world through a complex sensori-motor apparatus. This talk presents some recent steps in both directions, reporting on Fluid Construction Grammar (http://arti.vub.ac.be/FCG/), an operational linguistic framework that has specifically been designed to support emergent grammar, and on IRL, a computational framework that uses genetic programming techniques to evolve second order meanings.
Back to schedule - Back to top
Eors Szathmary - Level Formation in Biological Systems
Collegium Budapest
No abstract submitted yet.
Back to schedule - Back to top
Paul Vogt - On the acquisition and evolution of compositional languages: Sparse input and the productive creativity of children
University of Edinburgh
This paper investigates the productive creativity of children in a computational model on the emergence and evolution of compositional structures in language. In the model, agents try to generalise over initially unstructured and unanalysed expressions based on a usage-based grammar learning technique, where agents search for regularities in both the meaning and signal spaces. When such regularities are found, agents can break up expression-meaning pairs to form compositional structures.
In previous models it was shown that compositional structures can emerge in language when the language is transmitted from one generation to the next through a transmission bottleneck. Due to the fact that in these models language is transmitted only in a vertical direction where adults only speak to children and children only listen, this bottleneck needs to be imposed by the experimenter. In the current study, this bottleneck is removed and instead of having a vertical transmission of language, the language is -- in most simulations -- transmitted horizontally (i.e. any agent can speak to any other agent). It is shown that such a horizontal transmission scenario does not need an externally imposed bottleneck, because the children face an implicit bottleneck when they start speaking early in life.
The model is compared with the recent development of Nicaraguan Sign Language, where it is observed that children are a driving force for inventing grammatical (or compositional) structures, possibly due to a sparseness of input (i.e. an implicit bottleneck). The results show that in the studied model children are indeed the creative driving force for the emergence and stable evolution of compositional languages, thus suggesting that this implicit bottleneck may -- in part -- explain why children are so typically good at acquiring language and, moreover, why they may have been the driving force for the emergence of grammar in language.
The model will be demonstrated using a simulation toolkit of the Talking Heads experiment, called THSim, which is available at my web-page.
Back to schedule - Back to top