Final Report Summary - LANGEVO (Cultural and Biological Bases of Language Evolution)
This project has made a significant breakthrough for the question why human languages have syntactic structure and by what strategies such structures could emerge in a population of agents. Syntactic structure means here both phrase structure (hierarchical grouping of words and ordering based on their parts of speech) and grammatical agreement (for example between subject and verb or adjective and noun). The breakthrough has come from studying the combinatorial complexity of parsing utterances referring to multiple objects and expressing relations between them. It has been shown that without syntax, the complexity of possible interpretations increases as a double exponential. It has also been shown how different features and grammaticalisation processes seen in human languages all contribute to dampen the combinatorial explosion so that the size of utterances can scale up. For example, grouping words that introduce properties of the same object already reduces complexity from double exponential to exponential. Complementary to analytic arguments, computer simulations have been constructed in the form of populations of agents endowed with strategies for negotiating a communication system with syntactic properties. These simulations show that the communication system does not need to be innately given but can progressively self-organise through invention, adoption and alignment operations carried out by each individual agent but leading collectively to a shared grammar.
For agreement systems, five different strategies were examined: (a) adding formal markers to all words that refer to the same object, (b) adding meaningful makers, (c) adding meaningful markers recruited from existing words, (d) optimising articulatory complexity by phonological erosion, and (e) conventionalising agreement features by coercion. For phrase structure, the following strategies have been examined: (a) grouping words belonging to the same constituents, (b) imposing an order on the words leading to stored n-grams, and (c) introducing abstract patterns by categorising words in terms of part of speech. So far only first-order phrase structure has been investigated, i.e. phrase structure with only one level up from words.
In addition to concrete theoretical and computational work on language evolution, we have also pushed forward the state of the art in computational construction grammar. Specifically, we have worked on the integration of charts (a form of dynamic programming) in Fluid Construction Grammar and on the visualisation of the match and merge operations using visualisations developed in molecular dynamics.
The second phase of the research primarily focused on experiments for the origins of syntactic structure in the domain of phrase structure. Our goal is to analyse the selection criteria and come up with strategies that would allow a population of agents to self-organise communication systems with syntactic structure, and to reconstruct a series of stages so that research on the origins of language strategies could proceed. Intense work has also been performed in building the computational and formal linguistic tools to be able to do language evolution experiments.
Phrase structure means that words which are semantically related are grouped together into phrases. Words are assigned lexical categories (for example noun or verb), and these categories appear in a particular order in the phrase. Phrase structure can be hierarchical in the sense that a phrase can itself be assigned a phrasal category and patterns exist in the language which require phrases of a particular category to appear. In the second year of the project we looked only at first order phrase structure (one level up from words). In the final year, we investigated higher order phrase structure grammar as well as recursive structures.
The solution to this question is to come up with a minimal set of grammar building operators that allow individuals in a population to jointly bootstrap a recursive phrase structure grammar, starting from a vocabulary. Because it is far from obvious to understand the ramifications of such operators, as we are dealing with a complex adaptive system where a set of simple rules can have unexpected emergent properties, we need to prove the adequacy of the operators through an agent-based simulation.
The simulation implements a model of a population of `linguistic agents' playing language games about situations presented to them.
Some key publications:
-Steels, L. (2015). The Talking Heads Experiment. Berlin: Language Science Press. http://langsci-press.org/catalog/book/49
-Steels, L., & Casademont, E. G. (2015). Ambiguity and the origins of syntax. The Linguistic Review, 32(1), 37–60. doi:10.1515/tlr-2014-0021
-Beuls, K., & Steels, L. (2013). Agent-Based Models of Strategies for the Emergence and Evolution of Grammatical Agreement. PLOS ONE, 8(3), e58960. http://dx.plos.org/10.1371/journal.pone.0058960
-Steels, L. (ed.) (2012) Experiments in Cultural Language Evolution. John Benjamins Pub., Amsterdam.
-Steels, L. (2011) Modeling the cultural evolution of language. Physics of Life Reviews. 8(4) 330-356. http://www.sciencedirect.com/science/article/pii/S157106451100145X
-Steels, L. (ed.) (2011) Design Patterns in Fluid Construction Grammar. John Benjamins Pub. Amsterdam
For agreement systems, five different strategies were examined: (a) adding formal markers to all words that refer to the same object, (b) adding meaningful makers, (c) adding meaningful markers recruited from existing words, (d) optimising articulatory complexity by phonological erosion, and (e) conventionalising agreement features by coercion. For phrase structure, the following strategies have been examined: (a) grouping words belonging to the same constituents, (b) imposing an order on the words leading to stored n-grams, and (c) introducing abstract patterns by categorising words in terms of part of speech. So far only first-order phrase structure has been investigated, i.e. phrase structure with only one level up from words.
In addition to concrete theoretical and computational work on language evolution, we have also pushed forward the state of the art in computational construction grammar. Specifically, we have worked on the integration of charts (a form of dynamic programming) in Fluid Construction Grammar and on the visualisation of the match and merge operations using visualisations developed in molecular dynamics.
The second phase of the research primarily focused on experiments for the origins of syntactic structure in the domain of phrase structure. Our goal is to analyse the selection criteria and come up with strategies that would allow a population of agents to self-organise communication systems with syntactic structure, and to reconstruct a series of stages so that research on the origins of language strategies could proceed. Intense work has also been performed in building the computational and formal linguistic tools to be able to do language evolution experiments.
Phrase structure means that words which are semantically related are grouped together into phrases. Words are assigned lexical categories (for example noun or verb), and these categories appear in a particular order in the phrase. Phrase structure can be hierarchical in the sense that a phrase can itself be assigned a phrasal category and patterns exist in the language which require phrases of a particular category to appear. In the second year of the project we looked only at first order phrase structure (one level up from words). In the final year, we investigated higher order phrase structure grammar as well as recursive structures.
The solution to this question is to come up with a minimal set of grammar building operators that allow individuals in a population to jointly bootstrap a recursive phrase structure grammar, starting from a vocabulary. Because it is far from obvious to understand the ramifications of such operators, as we are dealing with a complex adaptive system where a set of simple rules can have unexpected emergent properties, we need to prove the adequacy of the operators through an agent-based simulation.
The simulation implements a model of a population of `linguistic agents' playing language games about situations presented to them.
Some key publications:
-Steels, L. (2015). The Talking Heads Experiment. Berlin: Language Science Press. http://langsci-press.org/catalog/book/49
-Steels, L., & Casademont, E. G. (2015). Ambiguity and the origins of syntax. The Linguistic Review, 32(1), 37–60. doi:10.1515/tlr-2014-0021
-Beuls, K., & Steels, L. (2013). Agent-Based Models of Strategies for the Emergence and Evolution of Grammatical Agreement. PLOS ONE, 8(3), e58960. http://dx.plos.org/10.1371/journal.pone.0058960
-Steels, L. (ed.) (2012) Experiments in Cultural Language Evolution. John Benjamins Pub., Amsterdam.
-Steels, L. (2011) Modeling the cultural evolution of language. Physics of Life Reviews. 8(4) 330-356. http://www.sciencedirect.com/science/article/pii/S157106451100145X
-Steels, L. (ed.) (2011) Design Patterns in Fluid Construction Grammar. John Benjamins Pub. Amsterdam