Integrated semantic processing of complex pictures and spoken sentences - Evidence from event-related potentials

It has been suggested that semantic information processing is modularized according to the input form (e.g., visual, verbal, non-verbal sound). A great deal of research has concentrated on detecting a separate verbal module. Also, it has traditionally been assumed in linguistics that the meaning of a single clause is computed before integration to a wider context. Recent research has called these views into question. The present study explored whether it is reasonable to assume separate verbal and nonverbal semantic systems in the light of the evidence from event-related potentials (ERPs). The study also provided information on whether the context influences processing of a single clause before the local meaning is computed. The focus was on an ERP called N400. Its amplitude is assumed to reflect the effort required to integrate an item to the preceding context. For instance, if a word is anomalous in its context, it will elicit a larger N400. N400 has been observed in experiments using both verbal and nonverbal stimuli. Contents of a single sentence were not hypothesized to influence the N400 amplitude. Only the combined contents of the sentence and the picture were hypothesized to influence the N400. The subjects (n = 17) viewed pictures on a computer screen while hearing sentences through headphones. Their task was to judge the congruency of the picture and the sentence. There were four conditions: 1) the picture and the sentence were congruent and sensible, 2) the sentence and the picture were congruent, but the sentence ended anomalously, 3) the picture and the sentence were incongruent but sensible, 4) the picture and the sentence were incongruent and anomalous. Stimuli from the four conditions were presented in a semi-randomized sequence. Their electroencephalography was simultaneously recorded. ERPs were computed for the four conditions. The amplitude of the N400 effect was largest in the incongruent sentence-picture -pairs. The anomalously ending sentences did not elicit a larger N400 than the sensible sentences. The results suggest that there is no separate verbal semantic system, and that the meaning of a single clause is not processed independent of the context.

Semantic Classes in Topic Detection and Tracking

Topic detection and tracking (TDT) is an area of information retrieval research the focus of which revolves around news events. The problems TDT deals with relate to segmenting news text into cohesive stories, detecting something new, previously unreported, tracking the development of a previously reported event, and grouping together news that discuss the same event. The performance of the traditional information retrieval techniques based on full-text similarity has remained inadequate for online production systems. It has been difficult to make the distinction between same and similar events. In this work, we explore ways of representing and comparing news documents in order to detect new events and track their development. First, however, we put forward a conceptual analysis of the notions of topic and event. The purpose is to clarify the terminology and align it with the process of news-making and the tradition of story-telling. Second, we present a framework for document similarity that is based on semantic classes, i.e., groups of words with similar meaning. We adopt people, organizations, and locations as semantic classes in addition to general terms. As each semantic class can be assigned its own similarity measure, document similarity can make use of ontologies, e.g., geographical taxonomies. The documents are compared class-wise, and the outcome is a weighted combination of class-wise similarities. Third, we incorporate temporal information into document similarity. We formalize the natural language temporal expressions occurring in the text, and use them to anchor the rest of the terms onto the time-line. Upon comparing documents for event-based similarity, we look not only at matching terms, but also how near their anchors are on the time-line. Fourth, we experiment with an adaptive variant of the semantic class similarity system. The news reflect changes in the real world, and in order to keep up, the system has to change its behavior based on the contents of the news stream. We put forward two strategies for rebuilding the topic representations and report experiment results. We run experiments with three annotated TDT corpora. The use of semantic classes increased the effectiveness of topic tracking by 10-30\% depending on the experimental setup. The gain in spotting new events remained lower, around 3-4\%. The anchoring the text to a time-line based on the temporal expressions gave a further 10\% increase the effectiveness of topic tracking. The gains in detecting new events, again, remained smaller. The adaptive systems did not improve the tracking results.