Showing papers by "Mihai Surdeanu published in 2007"

Combination strategies for semantic role labeling

Abstract: This paper introduces and analyzes a battery of inference models for the problem of semantic role labeling: one based on constraint satisfaction, and several strategies that model the inference as a meta-learning problem using discriminative classifiers. These classifiers are developed with a rich set of novel features that encode proposition and sentence-level information. To our knowledge, this is the first work that: (a) performs a thorough analysis of learning-based inference models for semantic role labeling, and (b) compares several inference strategies in this context. We evaluate the proposed inference strategies in the framework of the CoNLL-2005 shared task using only automatically-generated syntactic information. The extensive experimental evaluation and analysis indicates that all the proposed inference strategies are successful - they all outperform the current best results reported in the CoNLL-2005 evaluation exercise - but each of the proposed approaches has its advantages and disadvantages. Several important traits of a state-of-the-art SRL combination strategy emerge from this analysis: (i) individual models should be combined at the granularity of candidate arguments rather than at the granularity of complete solutions; (ii) the best combination strategy uses an inference model based in learning; and (iii) the learning-based inference benefits from max-margin classifiers and global feedback.

Robust Information Extraction with Perceptrons

Abstract: We present a system for the extraction of entity and relation mentions. Our work focused on robustness and simplicity: all system components are modeled using variants of the Perceptron algorithm (Rosemblatt, 1858) and only partial syntactic information is used for feature extraction. Our approach has two novel ideas. First, we define a new large-margin Perceptron algorithm tailored for classunbalanced data which dynamically adjusts its margins, according to the generalization performance of the model. Second, we propose a novel architecture that lets classification ambiguities flow through the system and solves them only at the end. The system achieves competitive accuracy on the ACE English EMD and RMD tasks.

A Comparison of Statistical and Rule-Induction Learners for Automatic Tagging of Time Expressions in English

Abstract: Proper recognition and handling of temporal information contained in a text is key to understanding the flow of events depicted in the text and their accompanying circumstances. Consequently, time expression recognition and representation of the time information they convey in a suitable normalized form is an important task relevant to several problems in Natural Language Processing. In particular, such an analysis is largely significant for Information Extraction (IE), Question Answering (QA) and Automatic Summarization (AS). The most common approach to time expression recognition in the past has been the use of handmade extraction rules (grammars), which also served as the basis for normalization. Our aim is to explore the possibilities afforded by applying machine learning techniques to the recognition of time expressions. We focus on recognizing the appearances of time expressions in text (not normalization) and transform the problem into one of chunking, where the aim is to correctly assign Begin, Inside or Outside (BIO) tags to tokens. In this paper, we explain the knowledge representation used and compare the results obtained in our experiments with two different methods, one statistical (support vector machines) and one of rule induction (FOIL). Our empirical analysis shows that SVMs are superior.

A multi-layer collaborative cache for question answering

Abstract: This paper is the first analysis of caching architectures for Question Answering (QA). We introduce the novel concept of multi-layer collaborative caches, where: (a) each resource intensive QA component is allocated a distinct segment of the cache, and (b) the overall cache is transparently spread across all nodes of the distributed system. We empirically analyze the proposed architecture using a real-world QA system installed on a cluster of 16 nodes. Our analysis indicates that multi-layer collaborative caches induce an almost two fold reduction in QA execution time compared to a QA system with local cache.

UPC: Experiments with Joint Learning within SemEval Task 9

Abstract: This paper describes UPC's participation in the SemEval-2007 task 9 (Marquez et al., 2007). We addressed all four subtasks using supervised learning. The paper introduces several novel issues: (a) for the SRL task, we propose a novel reranking algorithm based on the re-ranking Perceptron of Collins and Duffy (2002); and (b) for the same task we introduce a new set of global features that extract information not only at proposition level but also from the complete set of frame candidates. We show that in the SemEval setting, i.e., small training corpora, this approach outperforms previous work. Additionally, we added NSD and NER information in the global SRL model but this experiment was unsuccessful.

A Bootstrapping Architecture for Time Expression Recognition in Unlabelled Corpora via Syntactic-Semantic Patterns

Abstract: In this paper we describe a semi-supervised approach to the extraction of time expression mentions in large unlabelled corpora based on bootstrapping. Bootstrapping techniques rely on a relatively small amount of initial human-supplied examples (termed “seeds”) of the type of entity or concept to be learned, in order to capture an initial set of patterns or rules from the unlabelled text that extract the supplied data. In turn, the learned patterns are employed to find new potential examples, and the process is repeated to grow the set of patterns and (optionally) the set of examples. In order to prevent the learned pattern set from producing spurious results, it becomes essential to implement a ranking and selection procedure to filter out “bad” patterns and, depending on the case, new candidate examples. Therefore, the type of patterns employed (knowledge representation) as well as the ranking and selection procedure are paramount to the quality of the results. We present a complete bootstrapping algorithm for recognition of time expressions, with a special emphasis on the type of patterns used (a combination of semantic and morpho- syntantic elements) and the ranking and selection criteria. Bootstrap- ping techniques have been previously employed with limited success for several NLP problems, both of recognition and classification, but their application to time expression recognition is, to the best of our knowledge, novel. As of this writing, the described architecture is in the final stages of implementation, with experimention and evalution being already underway.