Dealing with phonological variations is important for speech processing. This article addresses whether phonological variations introduced by assimilatory processes are compensated for at the pre-lexical or lexical level, and whether the nature of variation and the phonological context influence this process. To this end, Swedish nasal regressive place assimilation was investigated using the mismatch negativity (MMN) component. In nasal regressive assimilation, the coronal nasal assimilates to the place of articulation of a following segment, most clearly with a velar or labial place of articulation, as in utan mej "without me" > [MODIFIER LETTER TRIANGULAR COLONtam mejMODIFIER LETTER TRIANGULAR COLON]. In a passive auditory oddball paradigm, 15 Swedish speakers were presented with Swedish phrases with attested and unattested phonological variations and contexts for nasal assimilation. Attested variations - a coronal-to-labial change as in utan "without" > [MODIFIER LETTER TRIANGULAR COLONtam] - were contrasted with unattested variations - a labial-to-coronal change as in utom "except" > *[MODIFIER LETTER TRIANGULAR COLONtLATIN SMALL LETTER OPEN On] - in appropriate and inappropriate contexts created by mej "me" [mejMODIFIER LETTER TRIANGULAR COLON] and dej "you" [dejMODIFIER LETTER TRIANGULAR COLON]. Given that the MMN amplitude depends on the degree of variation between two stimuli, the MMN responses were expected to indicate to what extent the distance between variants was tolerated by the perceptual system. Since the MMN response reflects not only low-level acoustic processing but also higher-level linguistic processes, the results were predicted to indicate whether listeners process assimilation at the pre-lexical and lexical levels. The results indicated no significant interactions across variations, suggesting that variations in phonological forms do not incur any cost in lexical retrieval; hence such variation is compensated for at the lexical level. However, since the MMN response reached significance only for a labial-to-coronal change in a labial context and for a coronal-to-labial change in a coronal context, the compensation might have been influenced by the nature of variation and the phonological context. It is therefore concluded that while assimilation is compensated for at the lexical level, there is also some influence from pre-lexical processing. The present results reveal not only signal-based perception of phonological units, but also higher-level lexical processing, and are thus able to reconcile the bottom-up and top-down models of speech processing.

This thesis examines the phonology of Gangou, a Sinitic (Chinese) variety spoken in Minhe County, Qinghai Province, China. Gangou is a variety of Mandarin, a.k.a. Northern Chinese, that belongs to what has been termed the Amdo Sprachbund, a linguistic area consisting of Sinitic, Bodic, Mongolic, and Turkic languages which have converged structurally towards a common Altaic prototype. Although the languages of the Amdo Sprachbund have started to receive more attention recently, Gangou remains an underdocumented language. Of the few articles that have been published, none so far has been dedicated to phonology, and it is the aim of this thesis to fill this gap. The material for the thesis was elicited during a field trip in Xining and Gangou, Qinghai Province, in September–October 2017. The data has been analysed based on general phonemic theory; no formalist theory of phonology has been adopted. The focus is primarily syn-chronic, but attention is paid to diachronic development where it facilitates the explanation of features that are peculiar to Gangou as compared to other Mandarin varieties such as Standard Mandarin. The primary goal of the present thesis is to determine the phonemes and tonemes of Gangou from a synchronic perspective. A secondary goal is to manifest the peculiarities of Gangou as compared to Standard Mandarin, which is the most widely known member of the Mandarin group of Chinese languages. Attempts have also been made to explain these unique features as possible contact-induced changes. As for the phonology of the languages of the Amdo Sprachbund, they represent either a Bodic or a Sinitic type. The results of this thesis show that Gangou exhibits a Sinitic orientation; it has a phonemic inventory that is close to that of other Mandarin varieties and a C (initial con-sonant) M (medial) V (vowel) F (final) syllable structure that is typical of Sinitic. Further-more, Gangou is tonal, having three tones in syllable-initial position, which is also a non-Bodic (i.e. Sinitic) feature. A comparison between the tone systems of Gangou and Standard Mandarin shows that a correspondence between Gangou and Standard Mandarin tones can be detected, which is to be expected. However, there are also many inconsistencies, especially in non-final syllables. Tone reduction may be one explanation to these inconsistencies.

A string encoding for a subclass of bipartite graphs enables graph rewriting used in autosegmental descriptions of tone phonology via existing and highly optimized finite- state transducer toolkits (Yli-Jyrä 2013). The current work offers a rigorous treatment of this code-theoretic approach, generalizing the methodology to all bipartite graphs having no crossing edges and un- ordered nodes. We present three bijectively related codes each of which exhibit unique characteristics while preserving the freedom to violate or express the OCP constraint. The codes are infinite, finite-state representable and optimal (efficiently computable, invertible, locally iconic, compositional) in the sense of Kornai (1995). They extend the encoding approach with visualisation, generality and flexibility and they make encoded graphs a strong candidate when the formal semantics of autosegmental phonology or non-crossing alignment relations are implemented within the con- fines of regular grammar.