The Emergence of Compositional Languages for Numeric Concepts Through Iterated Learning in Neural Agents

Abstract

Since first introduced by [6], computer simulation has been an increasingly important tool in evolutionary linguistics. Recently, with the development of deep learning techniques, research in grounded language learning has also started to focus on facilitating the emergence of compositional languages without pre-defined elementary linguistic knowledge. In this work, we explore the emergence of compositional languages for numeric concepts in multi-agent communication systems. We demonstrate that compositional language for encoding numeric concepts can emerge through iterated learning in populations of deep neural network agents. However, language properties greatly depend on the input representations given to agents. We found that compositional languages only emerge if they require less iterations to be fully learnt than other non-degenerate languages for agents on a given input representation.

Iterated learning for emergent systematicity in VQA

“…esis. The machine learning community has also recently shown an increasing interest in applying IL towards emergent communication (Guo et al, 2019;Li & Bowling, 2019;Cogswell et al, 2019;Dagan et al, 2020;Ren et al, 2020). Different from previous works, we believe that IL is an algorithmic principle that is equally applicable to recovering compositional structure in more general tasks.…”

On (Emergent) Systematic Generalisation and Compositionality in Visual Referential Games with Straight-Through Gumbel-Softmax Estimator

“…In recent years, research into language emergence and grounding have received increased attention. The former of which raises the question of how to make artificial languages emerge with similar properties to natural languages, or at least 'natural-like' protolanguages, exhibiting compositionality as the primarily-targeted property [4,21,45,60]. Indeed, languages' compositionality has been shown to further the learnability of said languages [34,64,9,45] and promises to increase the generalisation ability of the artificial agent that would be able to wield them.…”

“…Although it has been shown that emerging languages are far from being 'natural'-like [39,12,13], there are some successful cases demonstrating the emergence of compositional languages and learned representations (e.g. Kottur et al [39], Lazaridou et al [44], Choi et al [15], Bogin et al [7], Guo et al [21], Korbak et al [37], Chaabouni et al [14]), relative to a given standard of compositionality. This paper focuses exclusively on the Straight-Through Gumbel-Softmax (ST-GS) approach proposed by Havrylov and Titov [23], as it supposedly allows a richer signal towards solving the credit assignment problem that language emergence poses since the gradient can be backpropagated from the listener agent to the speaker agent, while, in comparison, it cannot be backpropagated when using more commonly adopted approaches based on REINFORCE-like algorithms [69].…”

“…Like Keresztury and Bruni [33], we acknowledge that it ought to be measured more quantitatively than doing qualitative eye-balling of the artificial agents' utterances. Therefore, this paper aims to provide a more complete picture of the phenomena involving the ST-GS approach using two metrics: topographic similarity [11], which is acknowledged by the research community as the main quantitative metric for compositionality [44,21,63,14,60], and zero-shot compositional tests, in which a set of stimuli composed of specific attributes are held-out from the training set, while making sure that the agents are still familiarising themselves with the specific attributes in different contexts/combinations to the held-out ones. The use of these two metrics has been shown to be critical by the concurrent work of Chaabouni et al [14], which shows that learning agents are able to generalise in a referential game in spite of their utterances not being compositional, when measured with topographic similarity.…”

ReferentialGym: A Nomenclature and Framework for Language Emergence & Grounding in (Visual) Referential Games

“…Computational linguists have been researching the emergence of these properties in artificial languages induced by language games [20,39,9,10,40,33,61,41] to better understand the evolution of natural languages. It is only relatively recently that it has also been investigated within the context of deep learning [46,29,45,25,47,7,19,43,21,15,16,49,60,27,50,2,17,3], as the ability to ground into other modalities a natural-like language is thought to be a prerequisite for general AI [69,54,4,18,3].…”