Mechanisms of stress assignment in Greek and English skilled reading

Assignment of lexical stress to the appropriate syllable is an integral part of reading polysyllabic words. The attention of reading research and computational modelling of reading has recently shifted from the monosyllabic to the polysyllabic domain, hence making more apparent the need to understand the stress assignment process. The great amount of cross-linguistic variability regarding the type and the amount of information that each language provides to its readers about stress position indicates that the stress assignment mechanism is not a uniform process among different languages. Research on the stress assignment mechanism investigates how diverse sources of information on stress position in each language interact with and shape the process of assigning stress to polysyllabic words. The main focus of this thesis was to enhance theoretical knowledge regarding stress assignment by collecting and analysing data in two languages with quite distinct linguistic characteristics regarding stress, namely, Greek and English. The sources of information on stress position can be generally summarised in lexical contributions by access to the stored position of stress for each word in the mental representations, the explicit indication of stress position by orthographic stress marks, and language-dependent statistical regularities between orthography and/or phonology and stress position. In Greek, stress position is explicitly and consistently marked orthographically by means of a diacritic over the vowel of the stressed syllable but only in lower and not uppercase words. Hence, it provides an ideal platform for investigations with naturalistic manipulations of the availability of this information through manipulation of case. A default metrical bias to stress a specific syllable, the prefinal, has also been observed. With the utilisation of reading aloud and lexical decision experiments, it was investigated: 1) how the different sources of information interact with each other; 2) how the shift in reading strategies has an impact on this mechanism; and 3) within the framework of a computationally implemented theory of reading, what levels of representation stress is encoded at. Experiments with Greek skilled readers in reading aloud showed that the stress assignment mechanism is directly dependent upon shifts of processing strategies in reading with lexical processing encouraging the extraction of stress position information directly from the lexicon. The default bias was also found to be active and dominant in the absence of other sources but also of relatively limited contributions when the other sources are present. Effects of case type of presentation were also disentangled from any potential effects from the availability of the diacritics. Case type was found to have an independent role in reading regardless whether this coincides with the presence of stress marks (in lowercase) or not (in uppercase). Experiments with skilled readers in visual word recognition showed evidence that stress is a multifaceted feature encoded in several levels of representation in the reading system, including as an independent unit processed at the letter level, stored in the phonological representations, and embedded in the orthographic representations of known words. Additionally, statistical analyses of the statistical characteristics of Greek orthography, which eventually led to the development of a new psycholinguistic database (Kyparissiadis, van Heuven, Pitchford & Ledgeway, 2017), showed that, in similarity with other languages such as English, the orthographic patterns in the endings of Greek words can be associated with specific stress positions. It remains to be investigated whether such regularities have a behavioural dimension and inform the stress mechanism as an additional source of information on stress assignment. This has been shown in other languages that present similar regularities. Furthermore, it was shown that the default metrical bias, which has universally been accounted as the result of a statistical preponderance of stress on a specific syllable, does not present a uniform statistical dominance in Greek. The process of adding part-of-speech information to the database elucidated that not all grammatical categories are predominantly stressed on the default position. In contrast to other languages, in which readers have been shown to be sensitive to the shift of statistical regularities within different subsets of the language’s battery, a lexical decision experiment in Greek showed uniform conformance of readers to the default bias even in a subset or words that does not present this statistical dominance. This raises questions regarding the origins of the default bias and warrants further investigations. In English, in conjunction to lexical retrieval, stress assignment is argued to be heavily affected by associations between orthography and the stressed syllable. Investigations on the default metrical bias in English have given rise to inconsistent results. Direct comparison of the contributions of the default mechanism against lexical activations in stress assignment showed no evidence for a default pattern being active in English reading in the presence of lexical information. However, the shifting of stressing strategies induced by a subsequent implicit priming experiment suggests that stress assignment through lexical access can be still affected by sublexical processes. Computational simulations with a leading computational model of reading showed a heavier reliance of the model on lexical contributions rather than orthographic information when these two sources where contrasted against each other. However, the model showed sub-optimal performance when simulating the behavioural results of the first experiment in English reported above. The reasons behind this inefficiency of a model that is able to simulate a wide range of behavioural patterns are considered and discussed. Finally, results of simulations from a preliminary word recognition model in Greek are presented and future directions for these modelling investigations are discussed. Overall, the results of this body of research indicate the stress assignment mechanism is multi-facetted and language-dependent. This has implications for universal models of written word recognition and reading aloud that future research will need to address