School of Languages and Linguistics - Theses
Permanent URI for this collection
Search Results
1 - 2 of 2
-
ItemPerceptual modification of nonnative phonemic sequences( 2023)This thesis investigates how native language (L1) segmental phonology and phonotactics interfere with the perception of phonological structures in a nonnative or second language (L2). Previous research has shown that nonnative listeners at times modify the phonological structures of an L2 based on the regularities of their L1 phonology, and that they sometimes experience perceptual difficulties in distinguishing L2 contrasts. The present thesis presents a total of five case studies focusing on different kinds of L1-L2 phonological mismatch, all of which trigger different corresponding modification strategies, including neutralisation (contrastive phonemes become non-contrastive), substitution (replacing a target phoneme with a different segment), epenthesis (perceiving an illusory segment when there is no target), and deletion (failing to perceive a target segment). These perceptual modification strategies are investigated through a series of psycholinguistic experiments relying on established methods in the field (e.g., categorisation, identification, and discrimination), and newer methods such as mouse tracking for triangulating on the cognitive processes involved in perceptual modification strategies. The thesis also explores whether extensive experience with the target language and especially the expansion of the L2 vocabulary, is predictive of L2 listeners’ ability to accurately perceive novel phonotactic structures. The findings of the present thesis have strong implications for extending the prevalent theories of nonnative speech perception, including the Perceptual Assimilation Model (PAM, and its extension, PAM-L2; Best, 1995; Best & Tyler, 2007), the Vocabulary Model of Rephonologisation (Vocab; Bundgaard-Nielsen et al., 2011a, 2011b, 2012), and the Automatic Selective Perception Model (ASP; Strange & Shafer, 2008; Strange, 2011). On the basis of the experiments conducted, this thesis argues that while perceptual assimilation is the fundamental mechanism for understanding nonnative (L2) segmental perception, the current frameworks must be extended to also allow cross-language category mapping for more complex phonological structures (i.e., phonemic sequences) in order to understand how L1 phonotactic expectations interfere with segmental perception in L2 speech perception.
-
ItemCan machine translation be used for literary texts? Evidence from a reception study( 2022)Despite the growing use of machine translation (MT) systems to translate texts in various domains, there remain considerable doubts about using this technology for the translation of literary texts. However, the outlook for using MT for literary translation has been enhanced by the advent of neural machine translation (NMT), a newly proposed paradigm for machine translation that outperforms the previously dominant paradigms such as phrase-based machine translation. To examine whether neural machine translation can be used to render (or help human translators to render) literary texts, this study compares how literary readers receive free-verse poems and excerpts from novels that have been translated from English into Chinese in three modalities: the raw machine translations (MTs) produced by a freely available generic NMT system, the post-edited machine translations (MTPEs) produced by a novice translator with no professional experience in literary translation, and the published fully human translations (HTs) produced by professional human translators. In total, fifteen translations (five start texts x three modalities) were investigated on the basis of actual reception by 131 readers. Given the complexity of literary reception, three research methods are used to triangulate the readers’ receptions of the literary translations: a translation-rating task in which the readers rated their reading experience of the given translations, a translation-annotation task in which the readers annotated the text items in the translations with likes and dislikes, and semi-structured interviews in which the readers compared and commented on the different translations of each start text. A total of 131 subjects performed the first two tasks, with each subject responding to one of the three translations of each start text. Subsequently, 15 volunteers from the sample participated in the post-hoc interviews. In this study, “annotation heatmaps” were proposed as an innovative method to visualise and analyse the readers’ liked and disliked annotations. Drawing on the heatmaps frequently used in eye-tracking studies, the annotation heatmaps visualise the frequency of each translation segment being liked or disliked by the readers with different shades of the highlighting colours of the translation segments. By showing the patterns in which the renditions in a translation were appreciated by the subjects and comparing the appreciation patterns of the corresponding renditions in different translation versions, the heatmaps provide fine-grained understanding of how the translation solutions in a target text affect the readers’ literary reception and how the human- and machine-produced solutions differ from each other in their effects on the reception side. Since the annotation maps present empirical data on the frequency of each translation segment be approved or rejected by its readers, rich discussions are elicited on the communicative risks associated with literary translation. Methodologically, the annotation heatmaps well complemented the rating and interview data. Altogether, the three types of reception data provide consistent and mutually enriching insights into the readers’ reception processes. First, although the raw MTs generally received a less positive reception than the other translation modalities, each raw MT involved a considerable number of textual elements that were appreciated positively by most readers. This finding to a large extent challenges the widespread conviction that machine-translated literary texts are not worth reading. Second, although produced by a novice post-editor in a short amount of time, most of the MTPEs not only resulted in a better reading experience than did the raw machine translations but also received similar or even more positive appreciation than did the fully human translations. Given that the MTPEs were produced by a novice in a short amount of time whereas the HTs were taken from published translations rendered by professional literary translators, the small gap between MTPEs and HTs supports the feasibly of using machine translation and post-editing for literary translation. Third, for each start text, there are cases in which the renditions in the published HTs worked less well than did the solutions in the MTs or MTPEs. Thus, human translations cannot be assumed to be a flawless gold standard for the translation of literary text.