This thesis proposes a generalized probabilistic approach to modelling document collections along the combined axes of both semantics and syntax. Probabilistic topic (or semantic) models view documents as random mixtures of unobserved latent topics which are themselves represented as probabilistic distributions over words. They have grown immensely in popularity since the introduction of the original topic model, Latent Dirichlet Allocation (LDA), in 2004, and have seen successes in computational linguistics, bioinformatics, political science, and many other fields. Furthermore, the modular nature of topic models allows them to be extended and adapted to specific tasks with relative ease. Despite the recorded successes, however, there remains a gap in combining axes of information from different sources and in developing models that are as useful as possible for specific applications, particularly in Natural Language Processing (NLP). The main contributions of this thesis are two-fold. First, we present generalized probabilistic models (both parametric and nonparametric) that are semantically and syntactically coherent and contain many simpler probabilistic models as special cases. Our models are consistent along both axes of word information in that an LDA-like component sorts words that are semantically related into distinct topics and a Hidden Markov Model (HMM)-like component determines the syntactic parts-of-speech of words so that we can group words that are both semantically and syntactically affiliated in an unsupervised manner, leading to such groups as verbs about health care and nouns about sports. Second, we apply our generalized probabilistic models to two NLP tasks. Specifically, we present new approaches to automatic text summarization and unsupervised part-of-speech (POS) tagging using our models and report results commensurate with the state-of-the-art in these two sub-fields. Our successes demonstrate the general applicability of our modelling techniques to important areas in computational linguistics and NLP.

Trijų dalių tapybos kolekcija „Šeimos portretas“ – tai tremtį išgyvenusių vyro ir žmonos psichologinis portretas. Kūrybinė interpretacija paremta sovietinio laikotarpio nuotraukomis. Tapybos darbuose keliama problema – skaudžios tremtinių patirties vaizdavimas. Politiniu, istoriniu, teisiniu atžvilgiu svarbi tema lemia archyvuojamų dokumentų saugojimą, žiniasklaidos aktyvumą, asmeninių patirčių pasakojimus, - atvirą galimybę menininkams įvairiai interpretuoti tremties temą – vaizduoti įvykius, procesus, padarinius, abstrahuoti, apibendrinti tikrovę ir individualias patirtis. Kolekcijos „Šeimos portretas“ temai atspirties tašku tapo fotografo V. Stanionio 1947 metų fotosesija, skirta naujų pasų nuotraukoms. To meto įvykiai susiję su masiniu trėmimu. Kūrybinė medžiaga kaupiama įvairiais būdais: skaitmenine spauda, vaizdą perkeliant ant drobės, grafiškų atspaudų kūrimu, tapybos variantų ieškojimu, schemų sudarymu kompiuterinėmis programomis (Photoshop, Corel Draw). Galutinis darbas atliekamas drobėje, aliejiniais dažais. Trijų dalių tapybos kolekcija atliekama 140 x 100 cm formate. Kompozicija artima altorinei, ji yra centrinė, uždara. Koloritas analogiškas nespalvotai fotografijai. Dramatiška šviesokaita, niaurių šešėlių pritvinkusiu fonu, siekiama išryškinti trapią nuotraukų šviesą. Baltais, tuščiais fragmentais kalba nyki tyla. Tremtis – šeimos, žmonijos laikmečio tamsi dėmė, prisiminimų ir vilties šviesoje. / The three-part paintings' collection called "The Family Portrait" is the psychological portrait of the family witch survived the exile. Creative interpretation is based on the Soviet period photographs. The problem posed by paintings is to depict painful experience of persons who survived an exile. Politically, historically, legally important issue determined archived documents storage, information motion, media activity, personal experiences. Stories became the open opportunity for the artists to interpret the theme of exile in different various ways – as the events, the processes or the effections. Abstract and summed up reality becames an individual experience. The starting point for the collection is 1947-th photo session for the new passports organised by photographer V. Stanionis. The recent events related to the mass deportations. The creative material is collected in various ways: by digital printing, an image transphered into the canvas, by graphically print design, by search options for painting plasics, schemes projected by computer software (Photoshop, Corel Draw). The final work is done on canvas using oil paints. Every three-piece collection takes 140 x 100 cm size. Composition close to the altar, it is a central and enclosed. The colours were chosed similar to the blac-white photography. The dramatic change of the concrasts, somber shadows filled up the background highlights the delicate images of the light. White and empty fragments speaks about the... [to full text]

Caregiver-child conversation is an important platform for children’s development of language and conversational skills and can form the basis for an approach to intervention for children with pragmatic difficulties. While most intervention studies have focused on reporting overall improvements in children’s language and conversational skills as a result of changes in caregiver behaviour, there is limited fine-grained understanding of children’s ability to achieve conversational topic contingency as a result of specific changes in their caregivers’ use of conversational topic turns and/or the facilitative techniques employed by therapists. The studies described in this thesis aim to address this limitation. The following questions are raised: (i) What is the impact of caregiver training intervention programmes on caregivers’ conversational topic turns and facilitative techniques? (ii) What is the impact of caregiver training intervention programmes on children’s conversational topic turns? and (iii) How do specific facilitative techniques impact children’s conversational skills? The first part of this thesis involves establishing the methodology for data transcription and data coding used in the four studies conducted for this thesis. Conversational recordings were transcribed using the CHAT format and conventions. Then a conversational coding system and a facilitative technique coding system were used to code the communicative acts transcribed. The conversational acts coded were topic change (TC), topic extension (TE), topic maintaining responses (TM), and non-relevant responses (NR). The facilitative techniques coded were imitation, expansion, follow-in questions, and follow-in cloze procedures. Inter-rater reliability levels of the transcription and coding of conversational acts and facilitative techniques were high. Study 1 (presented in Chapter 3) was designed to investigate changes following an ‘It Takes Two to Talk’ Hanen programme in the conversations of four caregiver-child dyads where the children had identified language delay. It employed a single subject design and the outcome measures were analysed in three phases: baseline, intervention and follow-up. The outcome measures were the rates of (i) caregivers’ conversational topic turns (i.e., TC, TE and TM), (ii) caregivers’ facilitative techniques (i.e., imitation, expansion, follow-in questions, and follow-in cloze procedures) and (iii) children’s conversational topic turns (i.e., TC, TE and TM). Results showed that all caregivers produced fewer TCs and higher rates of facilitative techniques, while all children produced significantly more TEs following intervention. Individual caregiver patterns of change appeared to be reflected in their children’s conversational skills. This study supports the effectiveness of caregiver group training programmes in improving the quality of caregiver-child conversations and highlighted the importance of investigating individual variations in intervention. Study 2 (presented in Chapter 4) was designed to investigate the changes in conversational skills of three children with features of autism spectrum disorders (ASD) following the dyads’ participation in caregiver-child individual training as part of a multidisciplinary programme for children with ASD. This study employed a case series design and the outcome measures were analysed in two phases (i.e., intervention and follow-up). Similar to Study 1, the outcome measures were rates of the (i) caregivers’ conversational topic turns (i.e., TC, TE and TM), (ii) caregivers’ facilitative techniques (i.e., imitation, expansion, follow-in questions, and follow-in cloze procedures) and (iii) children’s conversational topic turns (i.e., TC, TE and TM). Results of Study 2 showed that one of the three caregivers decreased the rates of TC, and one of them increased the rates of TM following intervention. The caregivers also increased their rates of facilitative techniques (i.e., imitation, expansion and follow-in questions). Concurrently, the children whose caregivers showed positive changes following intervention increased their rates of TCs and TMs. Consistent with the findings of Study 1, high individual variations were observed in the changes exhibited by the caregivers. Study 2 supports the effectiveness of individual training programmes in improving the quality of caregiver-child conversations and emphasizes the importance of investigating individual variations in intervention. Study 3 (presented in Chapter 5) was designed to investigate the qualitative changes made by a child from each of the first two studies. The two children presented with different language levels and aetiologies but both were receiving individualised programmes designed to enhance their conversational abilities. Taking a functional approach to communication development, Study 3 examined how each child (i) collaborated on an activity; (ii) expanded an activity; and (iii) returned to a previous activity or proposed a new activity, through conversational topic turns. Results suggest that the children learnt to collaborate on and expand activities through their caregivers’ repeated use of contingent topic turns (i.e., TM and TE) and facilitative techniques (i.e., expansion, follow-in questions and follow-in cloze procedures). However, they tended to return to a previous activity or propose a new activity when they did not attend to the preceding act or topic, seemed to not comprehend or were not interested in the preceding act or topic, or when their caregivers failed to attend to their preceding act or topic. These findings highlights that while caregivers’ topic turns that are contingent and facilitative help children to advance their activities, caregivers’ topic turns that are non-contingent have the potential to cause the children to end the preceding activity and switch to another activity. Study 4 was designed to compare the effectiveness of expansion, as a technique for facilitating children’s conversational topic turns, with expansion combined with other techniques when implemented by speech and language therapists (SLTs). Using a repeated measures design, this study aimed to compare the effects of expansion alone (EA); expansion combined with wh-questions (EQ); and expansion followed by a cloze procedure (EC) on the conversational skills of eight preschool children with conversational difficulties in conversation with their regular speech-language therapists (SLTs). Results showed that while there were no significant differences in child verbal topic maintaining responses across all techniques, EA elicited a significantly higher number of TEs, more non-verbal TMs and fewer NRs from the children, than either EQ or EC. The positive effects of each technique on the pragmatic appropriateness in conversations suggest that they could be used strategically in language intervention to enhance therapeutic effect. This thesis suggests that caregiver training programmes that focus on following the child’s lead and support caregivers and therapists to use contingent topic turns and facilitative techniques have positive outcomes for children’s conversational development. It also suggested that caregivers’ and therapists’ facilitative strategies that do not obligate responses from the child (i.e., expansion) have better potentials to help the child to expand the scope of conversations than strategies that obligate a response from the child (i.e., wh-questions and cloze procedures). Finally this thesis suggests that family-focussed intervention that follows the child’s lead appears likely to improve the conversational skills of children with a range of diagnoses by helping to address the common underlying features of conversational difficulties.

This thesis presents the results of an experimental study designed to examine whether children raised bilingually in Japanese and English from birth in Christchurch, New Zealand, exhibit the same morphological case and topic marking knowledge in Japanese as monolingual children in Japan. The participants were 34 children aged between five and eleven years who have been raised in a one-person one-language environment in an English dominant community. The study replicated previous studies on monolingual Japanese children, and involved two widely used paradigms for assessing a child’s grammar: picture selection, and elicited imitation. The responses of the children in this study were different from those reported in studies of monolingual children. In the picture selection tasks, some children in this study interpreted the agent-patient relationship based on the word order cue in the object-initial types of transitive sentences, whereas previous studies have demonstrated that monolingual children five years and older are able to interpret the agent-patient relationship in the same way as adults, using the case marking cue. Moreover, in the elicited imitation tasks, many children in this study re-analysed the topic-comment construction as a genitive possessive when the particles in the stimuli were masked with noise. This pattern has not been reported in any previous study. The results also revealed that there was a great degree of individual variation. The study suggests cross-linguistic influence from English on Japanese as a possible explanation for the difference between the children in this study and monolinguals. The phenomena observed in the results satisfies two conditions for cross-linguistic influence proposed by Hulk and Müller (2000) and Müller and Hulk (2001), because (i) English and Japanese overlap at the surface level in terms of the agent position in a canonical sentence and the possessive structure, and (ii) the problematic structures for some children in this study involved the interface between syntax and pragmatics in the C-domain. The study, however, has no principled explanation for the individual variation found because of a lack of data on the Japanese input and the child’s fluency, both of which are likely to affect simultaneous bilingual development.

The discourse of older healthy adults is commonly described as lengthy and off-topic and thought to be associated with a general cognitive decline that occurs in healthy aging. This study investigated the overall decline in attention associated with healthy aging and its relationship to instances of off-topic speech (OTS) in a constrained and unconstrained language production task. Ninety cognitively healthy adults participated and comprised three age cohorts (40 – 75+). Participants completed cognitive measures of attention and two discourse tasks that included recounting personal events and describing a procedure. Older adults exhibited poorer scores on measures of selective and shifting attention, and elderly adults produced more OTS as compared to middle-aged (40s) and older (60s) adults in the unconstrained task only. Poorer scores of shifting attention were significantly correlated with more OTS in the older adults (60s) only. Overall, a marked increase in variability of language production was observed with advancing age. Results indicated the need for further research on the relationships between age, attention, OTS, and task type in healthy aging to determine an underlying cause for increasing variability of language production with age.

Given that data is increasing exponentially everyday, extracting and understanding the information, themes and relationships from large collections of documents is more and more important to researchers in many areas. In this paper, we present a cooperative semantic information processing system to help biomedical researchers understand and discover knowledge in large numbers of titles and abstracts from PubMed query results. Our system is based on a prevalent technique, topic modeling, which is an unsupervised machine learning approach for discovering the set of semantic themes in a large set of documents. In addition, we apply a natural language processing technique to transform the “bag-of-words” assumption of topic models to the “bag-of-important-phrases” assumption and build an interactive visualization tool using a modified, open-source, Topic Browser. In the end, we conduct two experiments to evaluate the approach. The first, evaluates whether the “bag-of-important-phrases” approach is better at identifying semantic themes than the standard “bag-of-words” approach. This is an empirical study in which human subjects evaluate the quality of the resulting topics using a standard “word intrusion test” to determine whether subjects can identify a word (or phrase) that does not belong in the topic. The second is a qualitative empirical study to evaluate how well the system helps biomedical researchers explore a set of documents to discover previously hidden semantic themes and connections. The methodology for this study has been successfully used to evaluate other knowledge-discovery tools in biomedicine.

No description available.

<p>It has been increasingly important to develop statistical models to manage large-scale high-dimensional image data. This thesis presents novel hierarchical nonparametric Bayesian models for joint analysis of imagery and text. This thesis consists two main parts.</p><p>The first part is based on single image processing. We first present a spatially dependent model for simultaneous image segmentation and interpretation. Given a corrupted image, by imposing spatial inter-relationships within imagery, the model not only improves reconstruction performance but also yields smooth segmentation. Then we develop online variational Bayesian algorithm for dictionary learning to process large-scale datasets, based on online stochastic optimization with a natu- ral gradient step. We show that dictionary is learned simultaneously with image reconstruction on large natural images containing tens of millions of pixels.</p><p>The second part applies dictionary learning for joint analysis of multiple image and text to infer relationship among images. We show that feature extraction and image organization with annotation (when available) can be integrated by unifying dictionary learning and hierarchical topic modeling. We present image organization in both "flat" and hierarchical constructions. Compared with traditional algorithms feature extraction is separated from model learning, our algorithms not only better fits the datasets, but also provides richer and more interpretable structures of image</p> / Dissertation

While significant progress has been made in automatic speech recognition (ASR) during the last few decades, recognizing and understanding unconstrained conversational speech remains a challenging problem. In this dissertation, five methods/systems are proposed towards a robust conversational speech recognition and understanding system. I. A non-uniform minimum classification error (MCE) approach is proposed which can achieve consistent and significant keyword spotting performance gains on both English and Mandarin large-scale spontaneous conversational speech tasks (Switchboard and HKUST Mandarin CTS). II. A hybrid recurrent DNN-HMM system is proposed for robust acoustic modeling and a new way of backpropagation through time (BPTT) is introduced. The proposed system achieves state-of-the-art performances on two benchmark datasets, the 2nd CHiME challenge (track 2) and Aurora-4, without front-end preprocessing, speaker adaptive training or multiple decoding passes. III. To study the specific case of conversational speech recognition in the presence of competing talkers, several multi-style training setups of DNNs are investigated and a joint decoder operating on multi-talker speech is introduced. The proposed combined system improves upon the previous state-of-the-art IBM superhuman system by 2.8% absolute on the 2006 speech separation challenge dataset. IV. Latent semantic rational kernels (LSRKs) are proposed for spotting the semantic notions on conversational speech. The proposed framework is generalized using tf-idf weighting, latent semantic analysis, WordNet, probabilistic topic models and neural network learned representations and is shown to achieve substantial topic spotting performance gains on two conversational speech tasks, Switchboard and AT&T HMIHY initial collection. V. Non-uniform sequential discriminative training (DT) of DNNs with LSRKs is proposed which directly links the information of the proposed LSRK framework to the objective function of the DT. The experimental results on the subset of Switchboard show the proposed method can lead the acoustic modeling to a more robust system with respect to the semantic decoder.

There is significant interest in using gene expression measurements to aid in the personalization of medical treatment. The presence of significant normal tissue contamination in tumor samples makes it difficult to use tumor expression measurements to predict clinical variables and treatment response. I present a probabilistic method, TMMpure, to infer the expression profile of the cancerous tissue using a modified topic model that contains a hierarchical Dirichlet process prior on the cancer profiles. I demonstrate that TMMpure is able to infer the expression profile of cancerous tissue and improves the power of predictive models for clinical variables using expression profiles.