This dissertation presents a survey of center-embedding across languages, the factors involved in the processing difficulties, and implications for the evolution of human language.First, I discuss prior typological, experimental, and theoretical research on syntactic center-embedding, with a particular focus on how its acceptability and processability differs between SVO languages and SOV languages. The evidence shows that both universal factors and language-specific factors are involved, including clause types, referential form of subjects, and case marking. Building on this cross-linguistic survey, I present the results of a corpus study demonstrating that center-embedding is more common in verb-final languages. In addition, I examine sentences from the corpus search, and other observed English examples of multiple center-embedding, with regard to the facilitating semantic and syntactic factors observed in experiments. In addition to linguistic data, I report on experimental results including a series of nonlinguistic iterated sequence learning experiments intended to elicit the formation of center-embedding. While embedded structures did appear, they rarely showed up at more than chance rates, but there were (sometimes significant) differences between conditions. Specifically, repeated icons in initial strings did seem to prompt the formation of embedding. Additionally, I report on a pair of artificial language experiments involving dependency structures. The results suggest that center-embedding is particularly hard to process, possibly even harder than crossed dependencies (which are very rare in natural language). The results thus stand at odds with natural language typology. Next I put these findings into a broader perspective, exploring the roots of center-embedding in nonlinguistic (and not exclusively human) cognitive abilities. Based on a range of observational and experimental evidence, I argue that center-embedding is an inheritance from nonlinguistic cognition, limited in its scope in language by specific features of language (i.e. time pressure and semantics). Drawing on previous animal cognitive research, I present the results of another novel experiment, an attempt to teach birds to recognize a form of simple center-embedded pattern; one subject showed signs of successful learning. Finally I discuss the greater implications of these findings for the evolution of center-embedding in particular and language evolution in general.

Language researchers have long been interested in the extent to which performance factors play a central role in language use, and how these factors interact with linguistic representations of the competence grammar. Here I address this issue in three ERP experiments, investigating the role that domain-general neuro-cognitive systems play in language acquisition and comprehension.

In Experiment 1, I test the hypothesis that the formation of both long-distance syntactic dependencies and referential dependencies is underpinned by the same cognitive operations. I show that second elements in each dependency type elicit brain responses associated with the domain-general working memory system. Though the constraints on the formation of these two dependency types have traditionally been treated as distinct in the theoretical literature, I show that the same basic working memory processes are operative in each case. Nonetheless, subtle differences in the brain responses elicited are consistent with construction-specific properties of these dependencies as well.

In Experiment 2, I employ a traditional training-testing paradigm to show that selective attention plays a subtle, nuanced role in implicit language learning. After an hour of training, the brain showed domain-general responses consistent with implicit learning of grammatical systems unattended during training, though behavioral performance on judging the grammaticality of violations of these systems remained below chance, suggesting that the representations generated under brief implicit learning conditions are not sufficiently robust to influence behavior. I discuss these results in the context of studies of both implicit learning of non-linguistic information and those investigating second-language learning of grammatical and semantic information.

Lastly, in Experiment 3, I utilize a sentence-processing experiment and a task that measures attentional efficiency and show that the elicitation of early negativity (or eLAN) in sentence-processing contexts depends to a large degree on attentional efficiency, and therefore that these brain responses are not underlyingly linguistic, but should rather be interpreted as attentional modulations of domain-general N100 responses.

In general, I contend that the notion of “linguistic ERP components” may be misguided, and that what initially appear as language-specific responses should rather be interpreted as indices of domain-general cognitive systems operating over structured, statistically rich linguistic input.