This paper describes a system and set of algorithms for automatically inducing stand-alone monolingual part-of-speech taggers, base noun-phrase bracketers, named-entity taggers and morphological analyzers for an arbitrary foreign language. Case studies include French, Chinese, Czech and Spanish.Existing text analysis tools for English are applied to bilingual text corpora and their output projected onto the second language via statistically derived word alignments. Simple direct annotation projection is quite noisy, however, even with optimal alignments. Thus this paper presents noise-robust tagger, bracketer and lemmatizer training procedures capable of accurate system bootstrapping from noisy and incomplete initial projections.Performance of the induced stand-alone part-of-speech tagger applied to French achieves 96% core part-of-speech (POS) tag accuracy, and the corresponding induced noun-phrase bracketer exceeds 91% F-measure. The induced morphological analyzer achieves over 99% lemmatization accuracy on the complete French verbal system. This achievement is particularly noteworthy in that it required absolutely no hand-annotated training data in the given language, and virtually no language-specific knowledge or resources beyond raw text. Performance also significantly exceeds that obtained by direct annotation projection.

This paper presents a corpus-based algorithm capable of inducing inflectional morphological analyses of both regular and highly irregular forms (such as brought→bring) from distributional patterns in large monolingual text with no direct supervision. The algorithm combines four original alignment models based on relative corpus frequency, contextual similarity, weighted string similarity and incrementally retrained inflectional transduction probabilities. Starting with no paired examples for training and no prior seeding of legal morphological transformations, accuracy of the induced analyses of 3888 past-tense test cases in English exceeds 99.2% for the set, with currently over 80% accuracy on the most highly irregular forms and 99.7% accuracy on forms exhibiting non-concatenative suffixation.

This paper presents the WordFrame model, a noiserobust supervised algorithm capable of inducing morphological analyses for languages which exhibit prefixation, suffixation, and internal vowel shifts. In combination with a näive approach to suffix-based morphology, this algorithm is shown to be remarkably effective across a broad range of languages, including those exhibiting infixation and partial reduplication. Results are presented for over 30 languages with a median accuracy of 97.5% on test sets including both regular and irregular verbal inflections. Because the proposed method trains extremely well under conditions of high noise, it is an ideal candidate for use in co-training with unsupervised algorithms.

An ensemble of supervised maximum entropy classifiers can accurately detect and identify sentiments expressed in suicide notes. Using lexical and syntactic features extracted from a training set of externally annotated suicide notes, we trained separate classifiers for each of fifteen pre-specified emotions. This formed part of the 2011 i2b2 NLP Shared Task, Track 2. The precision and recall of these classifiers related strongly with the number of occurrences of each emotion in the training data. Evaluating on previously unseen test data, our best system achieved an F1 score of 0.534.

As part of the 2006 i2b2 NLP Shared Task, we explored two methods for determining the smoking status of patients from their hospital discharge summaries when explicit smoking terms were present and when those same terms were removed. We developed a simple keyword-based classifier to determine smoking status from de-identified hospital discharge summaries. We then developed a Naïve Bayes classifier to determine smoking status from the same records after all smoking-related words had been manually removed (the smoke-blind dataset). The performance of the Naïve Bayes classifier was compared with the performance of three human annotators on a subset of the same training dataset (n = 54) and against the evaluation dataset (n = 104 records). The rule-based classifier was able to accurately extract smoking status from hospital discharge summaries when they contained explicit smoking words. On the smoke-blind dataset, where explicit smoking cues are not available, two Naïve Bayes systems performed less well than the rule-based classifier, but similarly to three expert human annotators.