Skip to content
Conference Open access

Communicating in Emergent Language with an Induced Morphological Phrasebook

2026 · Annual Meeting of the Association for Computational Linguistics · pp. 30101-30121 · 0 citations · 62 references
Computer Science

TL;DR

The normalized pointwise mutual information of forms and meanings in the mor-phemes serves as a metric of compositionality that is more closely correlated with the ability of the phrasebook-agents to “speak” and “hear” an EL than existing metrics such as topographic similarity.

Abstract

We build rule-based emergent language (EL) agents using form–meaning mappings induced from ELs (“morphological phrasebooks”) and test their communicative performance in the EL environment with its neural network agents. This contributes three things: First, it assesses the quality of the morphemes discovered by the induction algorithm in situ , which we find to be effective for communicating in the EL. Second, it allows us to uncover morphosyntactic properties of EL through ablating the algorithms which induce and utilize morphemes, showing that the ELs rely on repetition as well as morpheme ordering to convey meaning. Third, we find that the normalized pointwise mutual information of forms and meanings in the mor-phemes serves as a metric of compositionality that is more closely correlated with the ability of the phrasebook-agents to “speak” and “hear” an EL than existing metrics such as topographic similarity.

Read PDF

Similar papers

Preprint Jul 2026

Evolving language compositionality in a frequency-structured meaning space

The iterated learning model was introduced to investigate language evolution: the way in which the characteristic properties of human languages have been shaped, at least partly, by repeated transmission from one language user to another. The key finding is that language compositionality can arise spontaneously as a consequence of language being passed repeatedly through a language learning bottleneck. Here we explore how changing the frequency of different meanings, so that some meanings occur much more frequently than others, affects the character of its compositionality. We find that, as observed in natural languages, high-frequency meanings can escape the pressure to conform to the grammar that characterizes lower-frequency meanings. However, when the frequency structure is instead imposed on parts rather than on whole meaning vectors, the language fails to transmit across generations. This occurs despite the fact that the most frequent elements are reliably learned. These results suggest that frequency can shape emergent linguistic structure only when the frequency distribution is defined over form-meaning units that learners can acquire holistically. When frequency is instead distributed over smaller units, it fails to support the relational structure required for compositional generalisation, thereby preventing stable language transmission.

Fabio De Ponte, Eloise Gaines-White, Conor J. Houghton et al. · 0 citations
Open access Jul 2026

Nominalization Patterns in Tombatu Language: A Generative Morphological Analysis of Affixal Noun Formation

Regional languages preserve complex grammatical systems that reveal how communities organize experience, identity, and cultural knowledge. However, nominalization in Tombatu, an underdescribed Austronesian language of Minahasa, has not been systematically examined through an explicit rule-based morphological framework. This study addresses this gap by investigating the structural patterns, semantic functions, and generative mechanisms of affixal noun formation in Tombatu. Employing a descriptive-taxonomic design, data were collected over six months in Tombatu Village from proficient speakers through naturally occurring speech, elicitation, recording, note-taking, and informant validation. The final dataset found unique nominalized forms. These forms were classified into three major processes comprising 11 noun-forming prefixes, one noun-forming suffix, and 14 affix combinations, yielding 26 identified affixal configurations. The analysis applied Word Combination Rules, Derivation Rules, the Boundary Insertion Convention, and relevant morphological constraints. The findings show that prefixation exhibits the greatest structural and semantic variation in the dataset, producing agentive, person-denoting, instrumental, object-related, and result-related nouns. For example, /tapə-/ combines with /lukuʔ/ ‘to drink’ to form /tapəlukuʔ/ [tapəlukuʔ] ‘drinker’. Suffixation is more restricted, with /-an/ forming locative nouns, as in /tawoi/ ‘to work’ plus /-an/, which produces /tawoian/ [tawoian] ‘workplace’. Complex affix combinations form nouns denoting places, processes, objects, states, results, and entities. This study provides the first systematic generative account of Tombatu nominalization, expands empirical evidence for Austronesian word formation, strengthens regional-language documentation, and offers a linguistic basis for language maintenance and potential contrastive morphological activities in multilingual education.

Verra E. Manangkot, J. Liuw, L. Amalputra et al. · 0 citations
Open access Jul 2026

Clarifying when-constructions: Conversation, morphosyntax, lexicon, and mode and their interactions in language use

The present study investigates clarifying when-constructions (e.g., when I say more, I don’t mean that we should consume more, but really it’s about creating a rich life) in a sample of 971 examples from the Corpus of Contemporary American English. In this construction, a phrase or clause from a different or the same turn is repeated in a when-clause to provide clarification through the main clause, often accompanied by a refining clause. The present research centers around identifying configurations of variables that are correlated with whether a speaker repeats a phrase or clause from a different or the same turn in a when-clause. We do so with a predictive modeling approach that uses the following variables: refining clause, main clause verb, main clause polarity, and mode. We also discuss the prototypical configurations associated with each type of repetition. Our findings go beyond traditional research on self-repetition and clarification by showing how their conversational functions interact with other grammatical domains in language use: morphosyntax, lexicon, and mode. This is in line with recent Usage-Based Construction Grammar work that has shown that constructionhood is a complex intersection of internal and external properties.

Jesús Olguín Martínez, S. Gries · 0 citations
Open access 2026

Population-level linguistic conventionality benefits the robustness, learnability, and cognitive efficiency of emergent languages

Human languages are widely understood to be conventional: for a given meaning, a population of language users expect a certain form to be used. Given the lack of unconventional natural languages with which to compare, there is little direct empirical evidence which explains why linguistic conventionality matters. We carry out a series of emergent language experiments with artificial agents designed to exhibit the effect of conventionality on robustness, learnability, and cognitive efficiency. By experimentally manipulating the interaction dynamics through which agents align their vocabularies, we create conditions that either promote or prevent population-level convergence, resulting in conventional and unconventional emergent languages while holding communicative success constant. We find that conventional languages are indeed more robust, better learnable, and more efficient, when compared with less conventional languages.

Jamie D Wright, Lara Verheyen, Fabio De Ponte et al. · 0 citations
Aug 2026

The Cybernetic Order-word: Tensors, Tensions and LLM Vector Spaces

Why what is really a matter of data analytics and statistical prediction is so readily assumed to be a display of real intelligence and even emergent cognition is explored by genealogically tracing the relationship between machines, organisms and language.

Chantelle Gray · 0 citations
Preprint Aug 2026

Rethinking and formalising the state across languages: a unified computational learning theory account

It is suggested that the state is a systemic, context-dependent morphosyntactic mechanism that selects grammatical templates across synthetic languages and constitutes one instance of a broader class of syntactically conditioned dependencies that also includes agreement and grammatical case.

M. E. Idrissi · 0 citations