This study investigated Taiwanese university students’ and teachers’ perceptions of foreign language anxiety in the English as a foreign language (EFL) classroom. The main aims were: (a) to identify the situations, sources, effects of, and coping tactics for the anxiety of Taiwanese tertiary students and (b) to examine tertiary English teachers’ perceptions of their students’ anxiety and how they deal with it. An anxiety scale, the ELCAS, was first administered to English major and non-English major students to identify the ten most anxious students in each group for individual semi-structured interviews. English majors’ teachers and those of non-English majors also had semi-structured interviews. The students’ degree of anxiety was statistically analyzed and revealed by IBM SPSS 20. The interview data from students and teachers were transcribed verbatim, coded, categorized, and then thematized in order to obtain the patterns of their perceptions on the issues. The summarized findings are:  The English major and non-major students were similar to each other in the situations, effects of, and coping strategies for anxiety, except for the sources of their anxiety.  The students reported a number of strategies, but most of these only helped them cope with individual anxious situations, not the root causes of their anxiety.  The English majors and their teachers had similar perceptions of anxiety in class although the latter revealed less specific situations than the former.  The similarity was also quite high between the non-majors’ and their teachers’ perceptions of their anxiety. These teachers also reported relatively broader contexts than their students.  Both groups of teachers employed quite diverse strategies for reducing students’ anxiety. Their tactics demonstrated their attempts to address problems at their sources.

No description available.

In this thesis, we describe extensive experiments on the classification of emotions from speech using acoustic features. This area of research has important applications in human computer interaction. We have thoroughly reviewed the current literature and present our results on some of the contemporary emotional speech databases. The principal focus is on creating a large set of acoustic features, descriptive of different emotional states and finding methods for selecting a subset of best performing features by using feature selection methods. In this thesis we have looked at several traditional feature selection methods and propose a novel scheme which employs a preferential Borda voting strategy for ranking features. The comparative results show that our proposed scheme can strike a balance between accurate but computationally intensive wrapper methods and less accurate but computationally less intensive filter methods for feature selection. By using the selected features, several schemes for extending the binary classifiers to multiclass classification are tested. Some of these classifiers form serial combinations of binary classifiers while others use a hierarchical structure to perform this task. We describe a new hierarchical classification scheme, which we call Data-Driven Dimensional Emotion Classification (3DEC), whose decision hierarchy is based on non-metric multidimensional scaling (NMDS) of the data. This method of creating a hierarchical structure for the classification of emotion classes gives significant improvements over other methods tested. The NMDS representation of emotional speech data can be interpreted in terms of the well-known valence-arousal model of emotion. We find that this model does not give a particularly good fit to the data: although the arousal dimension can be identified easily, valence is not well represented in the transformed data. From the recognition results on these two dimensions, we conclude that valence and arousal dimensions are not orthogonal to each other. In the last part of this thesis, we deal with the very difficult but important topic of improving the generalisation capabilities of speech emotion recognition (SER) systems over different speakers and recording environments. This topic has been generally overlooked in the current research in this area. First we try the traditional methods used in automatic speech recognition (ASR) systems for improving the generalisation of SER in intra– and inter–database emotion classification. These traditional methods do improve the average accuracy of the emotion classifier. In this thesis, we identify these differences in the training and test data, due to speakers and acoustic environments, as a covariate shift. This shift is minimised by using importance weighting algorithms from the emerging field of transfer learning to guide the learning algorithm towards that training data which gives better representation of testing data. Our results show that importance weighting algorithms can be used to minimise the differences between the training and testing data. We also test the effectiveness of importance weighting algorithms on inter–database and cross-lingual emotion recognition. From these results, we draw conclusions about the universal nature of emotions across different languages.

For the past three decades, psychological research has repeatedly shown that it is not always necessary for us to be conscious of events in order to perceive them, a phenomenon referred to as implicit cognition (Underwood & Bright 1996). Although this has been the subject of much research in the disciplines of psychology and social psychology, sociolinguists have only recently begun to examine how implicit cognition functions with regards to how we perceive speech (Campbell-Kibler 2012). Consistent with social psychology research on implicit responses to visually-derived social information (Greenwald et al 1998; Karpinski & Hilton 2001), recent sociolinguistic research suggests that listeners make differing conscious and unconscious social evaluations upon hearing different regional and foreign-accented speech varieties (Kristiansen 2009; Pantos & Perkins 2013), and that this is at least partly driven by socially-marked phonetic variation (Campbell-Kibler 2012, 2013). While previous research has investigated this phenomenon in relation to different regional or international varieties of English, the current study investigates the conscious and unconscious associations listeners make towards different social accents in Glasgow. This was achieved over three experiments by adapting an established psycholinguistic eye tracking methodology for sociolinguistic research. The first experiment (N=32) was conducted without eye tracking, relying on pencil and paper responses. Participants were tasked with choosing between on-screen ‘working-class’ and ‘middle-class’ target images (determined via a separate norming task) of brand logos and objects while recordings of different speakers uttering words semantically related to both images were heard. Non-significant trends were found in the data, with participants more likely to choose ‘working-class’ brand logos when a working-class speaker was heard and ‘middle-class’ logos when a middle-class speaker was heard. A second experiment (N=42) recorded listener eye movements in real time towards the same experimental stimuli, finding listeners to have been significantly (p < .05) more likely to fixate upon ‘working-class’ brand logos when hearing a working-class speaker than when hearing a middle-class speaker. Listeners’ verbal choices of brand logos showed no significant effect of speaker heard, showing a divergence between the on-line and off-line responses made towards speakers. Conversely, the speaker heard was found to have had a significant (p < .05) effect on the images of objects verbally chosen by listeners, but no effect on fixations made towards objects. A third experiment (N=54) investigated listener fixations towards brand logos while hearing words containing different socially-marked phonetic variants. Socially-marked phonetic realisations of CAT, post-vocalic/post-consonantal /l/, and non-prevocalic /r/ were all found to have elicited significant (p < .05) effects on listener fixation behaviour, with response times ranging from 300-700ms. A supplemental subjective reaction test (N=60) found participants to have evaluated middle-class Glaswegian speakers significantly (p < .05) more favourably in terms of Zahn & Hopper’s (1985) status attributes than working-class Glaswegian speakers, in line with the findings of previous language attitude studies (Preston 1999; Zahn & Hopper 1985; Kristiansen 2001). Overall, the results indicate that speech varieties with varying levels of perceived social status elicit differing conscious and unconscious social evaluations in listeners, and that socially-marked phonetic variation plays a role in this.

This thesis is an investigation of how written words are processed and represented for output. A single case study of an acquired dysgraphic patient is presented who produced a serial position effect in spelling tasks characterised by an increase in error rate from word beginning of word end. This pattern is assumed to reflect a deficit in the retrieval of stored orthographic representations. It is suggested that the order of output of letters may be encoded by an ordering of activation values. The nature of the deficit in terms of distinctions between input and output and access and storage are considered. The findings are discussed in relation to an existing connectionist model of spelling which was implemented and lesioned in an attempt to reproduce certain aspects of the patients data. Furthermore, a detailed analysis of the patients spelling errors suggests that orthographic representations consist of representational units other than the single letter and whole word. Finally, the role of the semantic system in lexical retrieval was investigated with regards to the distinction between proper names and common nouns. The ability of two patients to write/read proper names semantically was studied. Differences in the representational properties of proper names may result in them being selectively spared or impaired.

This thesis investigated the fundamental representations within the mental-lexicon and whether such representations are fixed or differ according to the characteristics of various languages. It looked at whether syllable structure is represented at distinct levels of linguistic representation at phonological and phonetic levels, with phonology governed by the demands of a combinatorial system (the need to create many distinct words from a small number of symbols) and phonetics governed by articulatory complexity (the need to keep motor programming as simple as possible). Empirical evidence as well as computational work was used to investigate whether syllable structure may be present as an abstract unit within the lexicon and not just computed online at the phonetic level. Three languages were explored in this work: English, Hindi and Italian. This project found evidence from English and Hindi patients with acquired language disorders to support the data previously collected from Italian patients. The empirical data was supported by computational work that considered the rates of resyllabification and storage costs based on the assumptions of different speech production models. Both the empirical and computational data support the hypothesis that syllable structure may be stored within the mental lexicon.

It is hypothesised that the development of concepts in formal education can be understood through the ideas of non-literal language and conceptual integration networks. The notions of concept, understanding and meaning are examined in some depth from philosophical, psychological and linguistic standpoints. The view that most concepts are grasped through non-literal means such as metaphor and conceptual blend is adopted. The central contention is that this applies both to everyday ideas and to those presented to students in formal educational contexts, and that consequently such learning is best seen in those terms. Such learning is not founded upon literal language, but a construction by the student of a complex network of metaphor and conceptual blends. This is examined in the context of students learning programming, in particular in the language Java. The hypothesis is tested by analysing transcribed interviews with a wide range of students, triangulated with an examination of teaching materials, and the data is shown to be consistent with the hypothesis. However the approach is fundamental and is not concerned with specific features of programming or Java, so that conclusions are relevant across a wide range of disciplines, especially mathematics, science and engineering. The thesis provides a new way of examining course design and learning materials including lectures and textbooks. Discourse which might seem to be literal is in fact metaphorical and blended, since it is in that way that the expert community understands the ideas. The students’ construction of corresponding blends is on the basis of their learning experience, and course design features such as examples can be explained and evaluated in such terms.

Based on Ullman’s (2004) hypothesis that declarative memory impairment will contribute to language impairment, this thesis presents two experiments that test familiarity and recollection in children with Autism Spectrum Disorder (ASD) and Intellectual Disability (ID). Four experimental groups comprised children and adolescents with ASD with language impairment (ALI); ASD without language impairment (ALN), intellectually disabled children without ASD (ID) and typically developing children (TD). Children were tested on two forced choice recognition tests of familiarity and recollection and a shape recognition and cued action-recall test. The relation between familiarity and conceptual semantic knowledge was investigated whilst controlling for visuo-perceptual abilities and fluid intelligence. Findings confirmed an association between familiarity and conceptual semantic knowledge in the ASD population as well as the use of visuo-perceptual skills to enhance familiarity. The broader role of declarative memory in language is addressed in ASD. Implications for future methods of testing in ASD populations are considered, as are implications of declarative memory anomalies in both ASD and ID populations in educational settings.

One of the phenomena to emerge from the study of human spoken interaction is accommodation or the tendency of an individual’s speech patterning to shift relative to their interlocutor. Whilst the experimental approach to the detection of accommodation has a solid background in the literature, it tends to treat the process of accommodation as a black box. The general approach for the detection of accommodation in speech has been to record the speech of a given speaker prior to interaction and then again after an interaction. These two measures are then compared to the speech of the interlocutor to test for similarity. If the speech sample following interaction is more similar then we can say that accommodation has taken place. Part of the goal of this thesis is to evaluate whether it is possible to look into the black box of speech accommodation and measure it ‘in situ’. Given that speech accommodation appears to take place as a result of interaction, it would be reasonable to assume that a similar effect might be observable in other areas contributing to a communicative interaction. The notion of an interacting dyad developing an increased degree of alignment over the course of an interaction has been proposed by psychologists. Theories have posited that alignment occurs at multiple levels of engagement, from broad levels of syntactic alignment down to phonetic levels of alignment. The use of speech accommodation as an anchor with which to track the evolution of change in the brain signal may prove to be one approach to investigating the claims made by these theories. The second part of this thesis aims to evaluate whether the phenomenon of accommodation is also observable in the form of electrical signals generated by the brain, measured using Electroencephalography (EEG). However, evaluating the change in the EEG signal over a continuous stretch of time is a hurdle that will need to be tackled. Traditionally, EEG methodologies involve averaging the signal over many repetitions of the same task. This is not a viable option when investigating communicative interaction. Clearly the evaluation of accommodation in both speech and brain activity, especially for continuously unfolding phenomena such as accommodation, is a non-trivial task. In order to tackle this, an approach from speech recognition and computer science has been employed. The implementation of Hidden Markov Models (HMM) has been used to develop speech recognition systems and has also been used to detect fraudulent attempts to imitate the voice of others. Given that HMMs have successfully been employed to detect the imitation of another person’s speech they are a good candidate for being able to detect the movement towards or away from an interlocutor during the course of an interaction. In addition, the use of HMMs is non-domain specific, they can be used to evaluate any time-variant signal. This adaptability of the approach allows for it to also be applied to EEG signals in conjunction with the speech signal. Two experiments are presented here. The behavioural experiment aims to evaluate the ability of a HMM based approach to detect accommodation by engaging pairs of female, Glaswegian speakers in the collaborative DiapixUK task. The results of their interactions are then evaluated from both a traditional phonetic standpoint, by assessing changes in Voice Onset Time (VOT) of stop consonants, formant values of vowels and speech rate over the course of an interaction and using the HMM based approach. The neural experiment looks to evaluate the ability of a HMM based approach to detect accommodation in both the speech signal and in brain activity. The same experiment that was performed in Experiment 1 was repeated, with the addition of EEG caps to both participants. The data was then evaluated using the HMM based approach. This thesis presents findings that suggest a function for speech accommodation that has not been explored in the past. This is done through the use of a novel, HMM based, holistic acoustic-phonetic measurement tool which produced consistent measures across both experiments. Further to this, the measurement tool is shown to have possible extended uses for EEG data. The use of the presented HMM based, holistic-acoustic measurement tool presents a novel contribution to the field for the measurement and evaluation of accommodation.