David Guthrie

Mining web sites using adaptive information extraction

Adaptive Information Extraction systems (IES) are currently used by some Semantic Web (SW) annotation tools as support to annotation (Handschuh et al., 2002; Vargas-Vera et al., 2002). They are generally based on fully supervised methodologies requiring fairly intense domain-specific annotation. Unfortunately, selecting representative examples may be difficult and annotations can be incorrect and require time. In this paper we present a methodology that drastically reduce (or even remove) the amount of manual annotation required when annotating consistent sets of pages. A very limited number of user-defined examples are used to bootstrap learning. Simple, high precision (and possibly high recall) IE patterns are induced using such examples, these patterns will then discover more examples which will in turn discover more patterns, etc.

Methods for Collection and Evaluation of Comparable Documents

Considerable attention is being paid to methods for gathering and evaluating comparable corpora, not only to improve Statistical Machine Translation (SMT) but for other applications as well, e.g. the extraction of paraphrases. The potential value of such corpora requires efficient and effective methods for gathering and evaluating them. Most of these methods have been tested in retrieving document pairs for well resourced languages, however there is a lack of work in areas of less popular (under resourced) languages, or domains. This chapter describes the work in developing methods for automatically gathering comparable corpora from the Web, specifically for under resourced languages. Different online sources are investigated and an evaluation method is developed to assess the quality of the retrieved documents.

Chinese Text Classification without Automatic Word Segmentation

Due to the lack of word boundaries in Asian systems of writing, machine processing of these languages often involves segmenting text into word units. This paper tests the assumption that this segmentation is a necessary step for authorship attribution and topic classification tasks in Chinese, and demonstrates that it is not. We show extensive results for both tasks, considering both single words and short phrases as features, and examining the effect of document length on classification accuracy. Our experiments show that a naïve character bigram model of text performs as well as models generated using a state-of-the-art automatic segmenter.

Maximum entropy modeling in sparse semantic tagging

In this work, we are concerned with a coarse grained semantic analysis over sparse data, which labels all nouns with a set of semantic categories. To get the benefit of unlabeled data, we propose a bootstrapping framework with Maximum Entropy modeling (MaxEnt) as the statistical learning component. During the iterative tagging process, unlabeled data is used not only for better statistical estimation, but also as a medium to integrate non-statistical knowledge into the model training. Two main issues are discussed in this paper. First, Association Rule principles are suggested to guide MaxEnt feature selections. Second, to guarantee the convergence of the boot-strapping process, three adjusting strategies are proposed to soft tag unlabeled data.