SSLI-LAB : Signal, Speech, and Language Interpretation Seminar

Sentiment classification is a very domain-specific problem: classifiers trained in one domain do not perform well in others. At the same time, large amounts of labeled data for fully-supervised learning approaches are not available for some domains, and a sentiment classifier needs to be customizable to new domains in order to be useful in practice. In this paper we survey four different approaches to customizing a sentiment classification system to a new target domain in the absence of large amounts of labeled data. We base our experiments on data from four different domains. After establishing that naïve cross-domain classification results in poor classification accuracy, we compare results obtained by using each of the four approaches and discuss their advantages, disadvantages and performance.

We describe our recent work on the development, implementation, and evaluation of the structured speech model with statistically characterized hidden trajectories. Bi-directional filtering (forward as well as backward in the temporal dimension) is developed on the hidden vocal tract resonance domain for all classes of speech sounds (including consonantal closure/constriction), offering strong power in parsimonious modeling of long-span speech co-articulation and capturing fine acoustic cues of CV and VC formant transitions. This statistical model, when appropriately implemented, also simultaneously exhibits the property of contextually assimilated phonetic reduction or phonetic target undershooting that is prevalent in casual, fluent speech (e.g., conversational speech). Experiments on large-scale N-best rescoring (N=1000) have demonstrated substantially lower TIMIT phone recognition errors achieved by the model compared with a context-dependent (triphone) HMM system built with HTK. When the ``error propagation'' effect of this long-span model is artificially removed in the N-best rescoring paradigm, the error bound is further cut down in a dramatic manner.

We present a part-of-speech tagger which introduces two new concepts: virtual evidence in the form of an .observed child. which is used to link its parents, and the use of negative training data in learning the conditional probabilities for the observed child. This model remains within the framework of Dynamic Bayesian Networks (DBNs) and is a conditionallystructured model which resolves certain drawbacks inherent in the conditional Markov model (CMM).

Reading proficiency is a fundamental component of language competency. However, finding topical texts at an appropriate reading level for foreign and second language learners is a challenge for teachers. This task can be addressed with natural language processing technology to assess reading level. Existing measures of reading level are not well suited to this task, but previous work and our own pilot experiments have shown the benefit of using statistical language models. In this paper, we also use support vector machines to combine features from traditional reading level measures, statistical language models, and other language processing tools to produce a better method of assessing reading level.

• Fall Quarter, 2004
• Spring Quarter, 2004
• Winter Quarter, 2004
• Fall Quarter, 2003
• Summer Quarter, 2003
• Spring Quarter, 2003
• Winter Quarter, 2003
• Spring Quarter, 2002
• Win Quarter, 2002
• Fall Quarter, 2001
• Spring Quarter, 2001
• Fall Quarter, 2000
• Spring Quarter, 2000