Search

Agglutinating morphology has often been described as easier to learn than fusional morphology, in large part because it is more transparent (e.g., Brown, 1976; Goldschneider & DeKeyser, 2001; Igartua, 2015). Such claims mainly rest on data from natural languages, where transparency is often correlated with additional factors such as regularity, compositionality and frequency of morphemes in the system, making it difficult to quantify the effect of transparency alone. To address this issue, this thesis presents a series of artificial language learning experiments which make it possible to investigate the impact of transparency, and more specifically, separative exponence, on learnability in isolation. The artificial language instantiates a nominal paradigm with two binary features, CLASS and NUMBER, expressed through suffixes. In addition to controlling for factors like regularity, this paradigm size makes it possible to hold the number of morphemes to be learned constant. Transparency was manipulated between conditions by varying whether the morphemes exhibited one-to-one mappings between form and meaning (separative exponence, as in agglutinating systems) or one-to-many mappings (cumulative exponence, as in fusional systems). Experiments were conducted with three different groups of native speakers to represent different morphological techniques in the L1, namely native speakers of English (isolating), Turkish (agglutinating) and Spanish (fusional). This way, it was possible to test whether specific morphological experience from the L1 differentially influences the transparency benefit in the L2. Across all experiments, with over 600 participants in total, no learning advantage of the transparent condition over the non-transparent condition was found. Only a slight bias for form-meaning mapping transparency was revealed in participants' error patterns. These results suggest that the role of transparency in the acquisition of morphology is more limited than commonly assumed and that it may be better understood as a benefit only in combination with other characteristics of morphological paradigms such as regularity, generalisability and frequency of morphemes.

The aim of this thesis is to investigate experimentally whether cross-linguistic variation in the structure of languages can be motivated by their external environment. It has been suggested that variation does not only result from cultural drift and language-internal mechanisms but also from social or even physical factors. However, from observational data and correlations between variables alone, it remains difficult to infer the exact underlying mechanisms. Here, I present a novel experimental approach for studying the relationship between language and environment under controlled laboratory conditions. I argue that to arrive at a causal understanding of linguistic adaptation, we can use a cultural evolutionary approach and simulate the emergence of linguistic structure with humans in the lab. This way, it can be tested which pressures shape linguistic features as they are used for communication and transmitted to new speakers. I focus primarily on cases where linguistic conventions emerge in referential communication games in direct face-to-face interaction. In these settings, I test whether specific conventions are more adaptive to solve the same problem under different conditions or affordances imposed by the environment. A series of silent-gesture experiments shows that systematicity (the design feature giving language its compositional power) is sensitive to the communicative environment: Dyads creating novel gestural communication systems to communicate pictorial referents are more likely to systematize traits and create categories that are functionally relevant in the given environment. Additionally, environmental features, such as the size of the meaning space and visibility of referents, affect the degree to which participants rely on systematic rather than simple holistic gestures. This 'experimental semiotics' approach thus models how environmental factors could motivate basic linguistic structure. However, for complex real-world phenomena, such as the hotly debated relationship between spatial language and environment, it is difficult to design simple experiments that isolate variables of interest but retain the necessary level of realism. It has been proposed that topography (e.g., landmarks like rivers, slopes) and sociocultural factors (e.g., bilingualism, subsistence style, population density) can affect whether speakers rely on an egocentric or geocentric Frame of Reference (FoR) to encode spatial relations, but it remains hard to disentangle the exact contribution of these variables to the cross-linguistic variation we observe. I tackle this issue with a novel paradigm: interactive Virtual Reality (VR) experiments that allow for an unprecedented combination of ecological validity and experimental control. In networked VR settings, participants are immersed in realistic settings such as a forest or a mountain slope. By having dyads solve spatial coordination games, I show that speakers of English, which is usually associated with an egocentric FoR, are less likely to use egocentric language (e.g., "the orb is to your left") if there are strong environmental affordances that make geocentric language more viable (e.g., "the orb is uphill from you"). Further experiments address whether the cultural 'success' of egocentric left/right could be motivated by its applicability across environments. For this, I combine VR with the 'experimental semiotics' approach, where the game is solved via a novel visual communication channel. I show how the movement data in the 3D world can be correlated with invented signals to measure which FoR participants rely on. In contrast to the English data, I did not find an advantage for geocentric systems in the slope environment, and overwhelmingly egocentric systems emerged. I discuss how this could relate to task-specificity and native language background. More generally, I show how this new way of studying spatial language with interactive VR games can be used to test hypotheses about linguistic transmission and material culture that could help explain the origins of the egocentric FoR system, which is regarded a fairly recent cultural innovation. Taken together, the thesis comprises several studies testing the relationship between linguistic and environmental variables. Additionally, VR is presented as a novel tool to study spatial language in controlled large-scale settings complementing more traditional fieldwork. More generally, I suggest that VR can be used to study the evolution of language in complex, multimodal settings without sacrificing experimental control.

Research on cross-linguistic differences in morphological paradigms reveals a wide range of variation on many dimensions, including the number of categories expressed, the number of unique forms, and the number of inflectional classes. This typological variation is surprising within the approach that languages evolve to maximise learnability (e.g., Christiansen and Chater 2008; Deacon 1997; Kirby 2002). Ackerman and Malouf (2013) argue that there is one dimension on which languages do not differ widely: in predictive structure. Predictive structure in a paradigm describes the extent to which forms predict each other, sometimes called i-complexity. Ackerman and Malouf (2013) show that although languages differ ac- cording to surface paradigm complexity measures, called e-complexity, they tend to have low i-complexity. While it has been suggested that i-complexity affects the task of producing unknown forms (the Paradigm Cell Filling Problem, Ackerman, James P. Blevins, et al. 2009; Ackerman and Malouf 2015), its effect on the learnability of morphological paradigms has not been tested. In a series of artificial language learning tasks both with human learners and LSTM neural networks, I evaluate the hypothesis that learners are sensitive to i-complexity by testing how well paradigms which differ on this dimension are learned. In Part 1, I test whether learners are sensitive to i-complexity when learning inflected forms in a miniature language. In Part 2, I compare the effect of i-complexity on learning with that of e-complexity and assess the relationship between these two measures, using randomly con- structed paradigms. In Part 3, I test the effect of i-complexity on learning and generalisation tasks, manipulating the presence of extra-morphological cues for class membership. Results show weak evidence for an effect of i-complexity on learning, with evidence for greater effects of e-complexity in both human and neural network learners. A strong nega- tive correlation was found between i-complexity and e-complexity, suggesting that paradigms with higher surface paradigm complexity tend to have more predictive structure, as mea- sured by i-complexity. There is no evidence for an interaction between i-complexity and extra-morphological cues on learning and generalisation. This suggests that semantic or phonological cues for class membership, which are common in natural languages, do not enhance the effect of i-complexity on learning and generalisation. Finally, i-complexity was found to affect generalisation in both human and neural network learners, suggesting that i-complexity could, in principle, shape languages through the process of generalisation to unknown forms. I discuss the difference in the effects of i-complexity on learning and generalisation, the similarities between the effect of i-complexity in human learners and neural networks, and cases the two types of learner differed. Finally, I discuss the role that i-complexity is likely to have in language change based on the results.

Why do languages look the way they do? This question lies at the core of much of linguistics research, and answering it can shine a light on the relationship between individual cognitive preferences and linguistic structure. One area that has attracted particular attention is basic word order. Many mature languages exhibit a fixed or dominant ordering of subject (S), object (O), and the verb (V). However, evidence from restricted communication systems, such as emerging sign languages, shows that even before such conventions have been established, language producers show strong ordering preferences. It has been suggested that SOV is the natural ordering of entities in an event, and the default order used by all newly emerging languages. Over the past decade or so, a growing body of research has endeavoured to investigate this question using the silent gesture paradigm in which participants describe events using only their hands. This work has been instrumental in uncovering a range of factors that influence the way people convey information about events in the absence of linguistic conventions, challenging the view that there is a single natural order. In this thesis, I present a series of experimental studies, implementing new techniques for data collection and analysis, showing how properties of individual referents influence the word orders people use to convey information about simple transitive events. I start, in Chapter 2, with a detailed review of the silent gesture literature, highlighting the numerous accounts that have been offered to explain word order preferences. One theme common to many of these accounts is that there is a direct relationship between word order and the structural and semantic properties of events. In Chapter 3, I report a silent gesture experiment in which I investigate an additional, complementary factor, namely, the salience of entities in an event. In the data analysis, I develop a novel computational method for inferring word order preferences based on incomplete gesture strings. The results of this study suggest that the relationship between salience and word order is not necessarily linear. Rather, manipulating the salience of referents influences the perspective from which a producer frames an event, which in turn influences structural choices. The results, however, are inconclusive about whether these structural choices reflect a direct mapping from conceptual structure to word order. In Chapter 4, I investigate the role of another operational factor: biases specific to the gestural modality. I report three experiments in which participants conveyed information about events by selecting pictorial representations of event components. Although the findings from this study are inconclusive, they nevertheless highlight important questions about the effects of other task-specific factors such as the way elicitation stimuli are presented. In Chapter 5, I focus on the relationship between word order and one of the most fundamental determiners and drivers of linguistic structure - animacy. Using a series of artificial language learning experiments, I test two existing accounts that have been proposed to explain animacy-based word order variation. One emphasizes communicative pressures arising from the potential ambiguity of events involving two human referents; the other focuses on the salience of humans relative to inanimate objects. The results of this study offer tentative support for the salience-based hypothesis. Nevertheless, I suggest that further work is required to understand how language producers negotiate the communicative challenge of accurately conveying information about events where the role played by each of the noun referents is ambiguous. Echoing the conclusions from the previous two studies, I also highlight the need for more research to better understand the role played by other factors, such as modality and native language. Overall, the studies reported in this thesis demonstrate that word order in newly developing languages not only reflects structural and semantic properties of events, but is also influenced by properties of referents interacting in an event, for example, salience and humanness. While the overall findings are inconclusive about the precise nature of this relationship, they nevertheless add to a growing body of literature showing that structural choices in the absence of linguistic conventions do not conform to a single natural order, but are subject to the effects of a range of potentially interacting factors.

Languages evolve in response to various pressures, and this thesis adopts the view that two pressures are especially important. Firstly, the process of learning a language functions as a pressure for greater simplicity due to a domain-general cognitive preference for simple structure. Secondly, the process of using a language in communicative scenarios functions as a pressure for greater informativeness because ultimately languages are only useful to the extent that they allow their users to express - or indeed represent - nuanced meaning distinctions. These two fundamental properties of language - simplicity and informativeness - are often, but not always, in conflict with each other. In general, a simple language cannot be informative and an informative language cannot be simple, resulting in the simplicity-informativeness tradeoff. Typological studies in several domains, including colour, kinship, and spatial relations, have demonstrated that languages find optimal solutions to this tradeoff - optimal solutions to the problem of balancing, on the one hand, the need for simplicity, and on the other, the need for informativeness. More specifically, the thesis explores how inductive reasoning and communicative interaction contribute to simple and informative structure respectively, with a particular emphasis on how a continuous space of meanings, such as the colour spectrum, may be divided into discrete labelled categories. The thesis first describes information-theoretic perspectives on learning and communication and highlights the fact that one of the hallmark feature of conceptual structure - which I term compactness - is not subject to the simplicity-informativeness tradeoff, since it confers advantages on both learning and use. This means it is unclear whether compact structure derives from a learning pressure or from a communicative pressure. To complicate matters further, some researchers view learning as a pressure for simplicity, as outlined above, while others have argued that learning might function as a pressure for informativeness in the sense that learners might have an a-priori expectation that languages ought to be informative. The thesis attempts to resolve this by formalizing these different perspectives in a model of an idealized Bayesian learner, and this model is used to make specific predictions about how these perspectives will play out during individual concept induction and also during the evolution of conceptual structure over time. Experimental testing of these predictions reveals overwhelming support for the simplicity account: Learners have a preference for simplicity, and over generational time, this preference becomes amplified, ultimately resulting in maximally simple, but nevertheless compact, conceptual structure. This emergent compact structure remains limited, however, because it only permits the expression of a small number of meaning distinctions - the emergent systems become degenerate. This issue is addressed in the second part of the thesis, which compares the outcomes of three experiments. The first replicates the finding above - compact categorical structure emerges from learning; the second and third experiments compare artificial and genuine pressures for expressivity, and they show that it is only in the presence of a live communicative task that higher level structure - a kind of statistical compositionality - can emerge. Working together, the low-level compact categorical structure, derived from learning, and the high-level compositional structure, derived from communicative interaction, provide a solution to the simplicity-informativeness tradeoff, expanding on and lending support to various claims in the literature.

Languages are culturally transmitted through a repeated cycle of learning and communicative interaction. These two aspects of cultural transmission impose (at least) three interacting pressures that can shape the evolution of linguistic structure: a pressure for learnability, a pressure for expressivity, and a pressure for coordination amongst users in a linguistic community. This thesis considers how these sometimes competing pressures impact linguistic complexity across cultural time. Using artificial language and iterated learning experimental paradigms, I investigate the conditions under which complexity in morphological and syntactic systems emerges, spreads, and reduces. These experiments illustrate the interaction of transmission, learning and use in hitherto understudied domains - morphosyntax and word order. In a first study (Chapter 2), I report the first iterated learning experiments to investigate the evolution of complexity in compositional structure at the word and sentence level. I demonstrate that a complex meaning space paired with pressures for learnability and communication can result in compositional hierarchical constituent structure, including fixed combinatorial rules of word formation and word order. This structure grants a productive and productively interpretable language and only requires learners to acquire a finite lexicon and a finite set of combinatorial rules (i.e., a grammar). In Chapter 3, I address the unique effect of communicative interaction on linguistic complexity, by removing language learning completely. Speakers use their native language to express novel meanings either in isolation or during communicative interaction. I demonstrate that even in this case, communicative interaction leads to more efficient and overall simpler linguistic systems. These first two studies provide support for the claim that morphological and syntactic complexity are shaped by an overarching drive towards simplicity (or learnability) in language learning and communication. Chapter 4 reports a series of experiments assessing the possibility that the simplicity bias found in the first two studies operates at a different strength depending on the linguistic level. Studies in natural language learning and in pidgin/creole genesis suggest that while morphological variation seems to be highly susceptible to regularisation, variation in other syntactic features, like word order, appears more likely to be reproduced. I test this experimentally by comparing regularisation of unconditioned variation across morphology and word order in the context of artificial language learning. I show that language users in fact regularise unconditioned variation in a similar way across linguistic levels, suggesting that the simplicity bias may be driven by a single, non-level-specific mechanism. Taken together, the experimental evidence presented in this thesis supports the hypothesis that the cultural and cognitive pressures acting on language users during learning and communicative interaction - for learnability, expressivity and coordination - are at least partially responsible for the evolution of linguistic complexity. Specifically, they are responsible for the emergence of linguistic complexity which maximises learnability and communicative efficiency, and for the reduction of complexity which does not. More generally, the approach taken in this thesis promotes a view of complexity in linguistic systems as an evolving variable determined by the biases of language learners and users as languages are culturally transmitted.