Achieving Compositional Language in a Population of Iterated Learners

Abstract

Iterated learning takes place when the input into a particular individual's learning process is itself the output of another individual's learning process. This is an important feature to capture when investigating human language change, or the dynamics of culturally learned behaviours in general. Over the last fifteen years, the Iterated Learning Model (ILM) has been used to shed light on how the population-level characteristics of learned communication arise. However, until now each iteration of the model has tended to feature a single immature language user learning from their interactions with a single mature language user. Here, the ILM is extended to include a population of immature and mature language users. We demonstrate that the structure and make-up of this population influences the dynamics of language change that occur over generational time. In particular, we show that, by increasing the number of trainers from which an agent learns, the agent in question learns a fully compositional language at a much faster rate, and with less training data. It is also shown that, so long as the number of mature agents is large enough, this finding holds even if a learner's trainers include other agents that do not yet posses full linguistic competence.

Simplicity and Specificity in Language: Domain-General Biases Have Domain-Specific Effects

“…It is an empirical question whether the particular ranking of grammars in terms of simplicity that we can derive from this particular representation matches precisely the ranking that applies in the case of real language learners, but we are confident that the crucial distinction between degenerate < compositional < holistic is correct. This matches behavior of participants in the lab (Kirby et al, 2015 ) and broadly similar results are found in both connectionist and symbolic models of iterated learning (Kirby and Hurford, 2002 ; Brace et al, 2015 ).…”

Understanding Language Evolution in Overlapping Generations of Reinforcement Learning Agents

“…This supports other work that has demonstrated a link between the linguistic bottleneck and the number of linguistic tutors (Brace et al, 2015). Indeed, the behaviour seen in Figure 7 indicates that the factors underpinning the cultural transmission of language change and linguistic variation are perhaps too complicated to be understood by analysing the nature of just inter-and intra-generational transmission; and that further research into linguistic change should focus on the nature of the social network the underpins linguistic populations (Wichmann and Holman, 2009;Lupyan and Dale, 2010;Reitter and Lebiere, 2010;Milroy, 2013).…”

“…A series of computer-based simulations that use the method of iterated learning have shown that the linguistic bottleneck is crucially important in regards to whether or not language can be successfully passed from one generation to the next and, in situations where this transmission can be achieved successfully, show that it is also crucial to the linguistic structure that arises (Kirby, 2002b,a;Kirby and Hurford, 2002;Kirby et al, 2014;Smith, 2002;Smith et al, 2003;Brace et al, 2015).…”

“…Although interesting in their own right, the dyadic setting considered so far limits the conclusions that can be drawn. After all, human language persists in a highly complex social milieu, and it has been shown that the structure and composition of a population can influence the dynamics of language change over time (Brace et al, 2015). Thus, the original reinforcement learning model (R-L) was extended in a number of ways.…”

Understanding Language Evolution in Overlapping Generations of Reinforcement Learning Agents

“…A series of computer-based simulations that use the method of iterated learning have shown that the linguistic bottleneck is crucially important in regards to whether or not language can be successfully passed from one generation to the next and, in situations where this transmission can be achieved successfully, show that it is also crucial to the linguistic structure that arises (Kirby, 2002b,a;Kirby and Hurford, 2002;Kirby et al, 2014;Smith, 2002;Smith et al, 2003;Brace et al, 2015). Although a similar effect to the bottleneck is seen in other types of uni-generational models, such as the naming game (Steels, 1995), the model presented here is novel in that it demonstrates the impact of bottleneck-like behaviour in a generational-based simulation that explores term-based languages.…”