Hierarchical centre embeddings (HCEs) in natural language have been taken as evidence that language is not processed as a finite state system (Chomsky, 1957). While phrase structure may be necessary to produce HCEs, finite state, sequential processing may underlie their comprehension (Frank, Bod, & Christiansen, 2012). Under this account, listeners employ surface level cues (e.g. semantic content) to determine the dependencies within an utterance, instead of processing the words in a hierarchy. The acoustic structure of speech reflects the speaker’s syntactic representation during production (Cooper, Paccia & Lapointe, 1978). In comprehension, temporal (Snedeker & Trueswell, 2003) and pitch (Watson, Tanenhaus, & Gunlogson, 2008) cues rapidly influence processing. Therefore, temporal and pitch variation in speech could contain cues to dependencies. We examine whether grouping behaviour may be driven by Gestalt principles. Temporal proximity suggests that individuals group sequential words that occur closer together in time. Pitch similarity states that individuals group sequential words that are similar in pitch. In this thesis, I examine whether these Gestalts support dependency detection in speech, providing a mechanism through which hierarchical structure can be processed non-hierarchically. In Chapter 3, we assessed whether temporal proximity and pitch similarity explicitly relate to the structure of a corpus of spontaneously produced active and passive relative clauses. This was the case for actives; the embedded clause was preceded by a lengthened pause and a large pitch reduction. For passives, a longer pause and pitch reduction occurred after the verb-phrase of the embedded clause, counter to prediction. The results for actives suggest that temporal proximity and pitch similarity cues could be used to group the phrases of the embedded clause, obviating the need to process hierarchically structured speech hierarchically. Two artificial grammar learning studies assessed whether pitch similarity and temporal proximity cues support the acquisition of phrase structure grammar. Chapter 4 emphasised temporal proximity cues, while chapter 5 emphasised pitch similarity cues. In Chapter 5, pitch similarity cues improved classification performance for structures with two levels of embedding. In both, participants did not benefit from temporal proximity cues. However, the results of a cross-species meta-analysis of artificial grammar learning studies (Chapter 2) raised the possibility that reflection-based measures (e.g. grammaticality judgements) are not well suited for assessing processing-based learning, such as online speech processing (Christiansen, 2018). To properly assess the role of Gestalt cues in speech processing therefore requires processing-based measures. To assess the influence of auditory Gestalts on online speech processing, in Chapter 6 we analysed participants’ gaze behaviour in response to pitch similarity and temporal proximity cues using the visual world paradigm. Participants heard speech-synthesised active-object and passive relative clauses, whilst viewing four potential targets. Each sentence had a prosodic structure consistent with either syntactic form (Chapter 3), or two control prosodic structures. Pitch similarity results indicated that these cues facilitated processing. Temporal proximity cues consistent with syntactic structure did not facilitate processing, instead results suggested a general benefit of increased processing time. Overall, these studies suggest that participants can use the pitch similarity Gestalt to group together syntactically dependent phrases in hierarchical speech, offering a mechanism through which individuals could process hierarchical structures non-hierarchically. The results of Chapters 4, 5, and 6 suggest temporal proximity cues did not facilitate performance to the same extent. Thus, we suggest that unfilled pauses in isolation may be insufficient to facilitate groupings on the basis of temporal proximity.