Mayfest 2024
The science of linguistic diversity.
Research at our top-ranked department spans syntax, semantics, phonology, language acquisition, computational linguistics, psycholinguistics and neurolinguistics.
Connections between our core competencies are strong, with theoretical, experimental and computational work typically pursued in tandem.
A network of collaboration at all levels sustains a research climate that is both vigorous and friendly. Here new ideas develop in conversation, stimulated by the steady activity of our labs and research groups, frequent student meetings with faculty, regular talks by local and invited scholars and collaborations with the broader University of Maryland language science community, the largest and most integrated language science research community in North America.
Current interest in automatic sentiment analysis is motivated by a variety of information requirements. The vast majority of work in sentiment analysis has been specifically targeted at detecting subjective statements and mining opinions. This dissertation focuses on a different but related problem that to date has received relatively little attention in NLP research: detecting implicit sentiment, or spin, in text. This text classification task is distinguished from other sentiment analysis work in that there is no assumption that the documents to be classified with respect to sentiment are necessarily overt expressions of opinion. They rather are documents that might reveal a perspective. This dissertation describes a novel approach to the identification of implicit sentiment, motivated by ideas drawn from the literature on lexical semantics and argument structure, supported and refined through psycholinguistic experimentation. A relationship predictive of sentiment is established for components of meaning that are thought to be drivers of verbal argument selection and linking and to be arbiters of what is foregrounded or backgrounded in discourse. In computational experiments employing targeted lexical selection for verbs and nouns, a set of features reflective of these components of meaning is extracted for the terms. As observable proxies for the underlying semantic components, these features are exploited using machine learning methods for text classification with respect to perspective. After initial experimentation with manually selected lexical resources, the method is generalized to require no manual selection or hand tuning of any kind. The robustness of this linguistically motivated method is demonstrated by successfully applying it to three distinct text domains under a number of different experimental conditions, obtaining the best classification accuracies yet reported for several sentiment classification tasks. A novel graph-based classifier combination method is introduced which further improves classification accuracy by integrating statistical classifiers with models of inter-document relationships.
Linguistics and psycholinguistics differ not in their topic but in their tools, and our choice of tools should be commensurate to the hypotheses we are testing. A case study of long-distance dependencies serves to illustrate the point.
Both the external world and our internal world are full of changing activities , and the question of how these two dynamic systems are linked constitutes the most intriguing and fundamental question in neuroscience and cognitive science. This study specifically investigates the processing and representation of sound dynamic information in human auditory cortex using magnetoencephalography (MEG), a non-invasive brain imaging technique whose high temporal resolution (on the order of ~1ms) makes it an appropriate tool for studying the neural correlates of dynamic auditory information. The other goal of this study is to understand the essence of the macroscopic activities reflected in non-invasive brain imaging experiments, specifically focusing on MEG. Invasive single-cell recordings in animals have yielded a large amount of information about how the brain works at a microscopic level. However, there still exist large gaps in our understanding of the relationship between the activities recorded at the microscopic level in animals and at the macroscopic level in humans, which have yet to be reconciled in terms of their different spatial scales and activities format, making a unified knowledge framework still unsuccessful. In this study, natural speech sentences and sounds containing speech-like temporal dynamic features are employed to probe the human auditory system. The recorded MEG signal is found to be well correlated with the stimulus dynamics via amplitude modulation (AM) and/or phase modulation (PM) mechanisms. Specifically, oscillations at various frequency bands are found to be the main information-carrying elements of the MEG signal, and the two major parameters of these endogenous brain rhythms, amplitude and phase, are modulated by incoming sensory stimulus dynamics, corresponding to AM and PM mechanism, to track sound dynamics. Crucially, such modulation tracking is found to be correlated with human perception and behavior. This study suggests that these two dynamic and complex systems, the external and internal worlds, systematically communicate and are coupled via modulation mechanism, leading to a reverberating flow of information embedded in oscillating waves in human cortex. The results also have implications for brain imaging studies, suggesting that these recorded macroscopic activities reflect brain state, the more close neural correlate of high-level cognitive behavior.
This paper presents three experiments which examine the effect of lexical surface frequency on sentence processing and the interaction between surface frequency and syntactic prediction. The first two experiments make use of the self-paced reading paradigm to show that processing time differences due to surface frequency (e.g., the frequency of cats not including occurrences of cat), which have previously been demonstrated in isolated word tasks like lexical decision, also give rise to reaction time differences in sentence processing tasks, in this case for singular and plural English nouns. The second experiment investigates whether a prediction for the number morpheme triggered by the number-marked determiners this and these might counter the surface frequency effect; however, the small size of the surface frequency effect and baseline differences in reaction times to this and these made the results unclear. Results from a third experiment using lexical decision suggest that the difference in the size of the surface frequency effects between the lexical decision experiments and the self-paced-reading experiments are likely due to differences in task demands. Our results have methodological implications for psycholinguistic experiments that manipulate morphology as a means of examining other questions of interest.