The paper presents a language model that develops syntactic structure and uses it to extract meaningful information from the word history, thus enabling the use of long distance dependencies. The model assigns probability to every joint sequence of words--binary-parse-structure with headword annotation. The model, its probabilistic parametrization, and a set of experiments meant to evaluate its predictive power are presented. 1 Introduction The main goal of the proposed project is to develop a language model(LM) that uses syntactic structure. The principles that guided this proposal were: ffl the model will develop syntactic knowledge as a built-in feature; it will assign a probability to every joint sequence of words--binary-parse-structure; ffl the model should operate in a left-to-right manner so that it would be possible to decode word lattices provided by an automatic speech recognizer. The model consists of two modules: a next word predictor which makes use of syntactic struc...

this paper, we present a method where more of the tree structure is used in the parsing model. We define a set of features that capture long distance dependency such as parallelism in coordination. These features are then integrated with a Maximum Entropy model into an overall probabilistic model for parsing. We introduce the decision tree parser in Section 2, describe the Maximum Entropy model in Section 3, describe the feature extraction algorithm in Section 4, give experimental results in Section 5, and present our conclusions in Section 6.

This paper presents a new general supervised word sense disambiguation method based on a relatively small syntactically parsed and semantically tagged training corpus. The method exploits a full sentential context and all the explicit semantic relations in a sentence to identify the senses of all of that sen- tence's content words. In spite of a very small training corpus, we report an overall accuracy of 80.3% (85.7, 63.9, 83.6 and 86.5%, for nouns, verbs, adjectives and adverbs, respectively), which exceeds the accuracy of a statistical sense-frequency based se- mantic tagging, the only really applicable general disambiguating technique.

This paper explores the kinds of probabilistic relations that are important in syntactic disambiguation. It proposes that two widely used kinds of relations, lexical dependencies and structural relations, have complementary disambiguation capabilities. It presents a new model based on structural relations, the Tree-gram model, and reports experiments showing that structural relations should benet from enrichment by lexical dependencies. 1 Introduction Head-lexicalization currently pervades in the parsing literature e.g. (Eisner, 1996; Collins, 1997; Charniak, 1999). This method extends every treebank nonterminal with its headword: the model is trained on this head lexicalized treebank. Head lexicalized models extract probabilistic relations between pairs of lexicalized nonterminals (\bilexical dependencies"): every relation is between a parent node and one of its children in a parse-tree. Bilexical dependencies generate parse-trees for input sentences via Markov proces...

fcdoran � beth � anoop � srini � fxiag�unagi.cis.upenn.edu

d my ignorance in the field. He was always patient, and took the time to explain his answers at a level I could understand. iv Dr. Todd Ward, a colleague of mine at IBM, has also "been there" for me. I cannot count the number of times that Todd helped me figure out a solution to a problem, either mathematical or programming. Whenever I was not sure about a solution to a problem, Todd was my sounding board. I'm sure that his individual research efforts were slowed by our meetings, but that never stopped him from helping me. Todd also acted as a counselor, providing insight on how to complete a doctorate! Former IBMer, Dr. Stephen Della Pietra, is without a doubt the brightest mathematician with whom I have ever worked. Like Salim and Todd, he knows statistical modeling at a much greater depth than I do, and he never minded "bringing down" the level of his explanations to one where I could understand and absorb the material. Stephen was my mentor, and without his expert tutelag

Abstract In a multimodal human-machine conversation, user inputs are often abbreviated or imprecise. Simply fusing multimodal inputs together may not be sufficient to derive a complete understanding of the inputs. Aiming to handle a wide variety of multimodal inputs, we are building a context-based multimodal interpretation framework called MIND (Multimodal Interpreter for Natural Dialog). MIND is unique in its use of a variety of contexts, such as domain context and conversation context, to enhance multimodal interpretation. In this chapter, we first describe a fine-grained semantic representation that captures salient information from user inputs and the overall conversation, and then present a context-based interpretation approach that enables MIND to reach a full understanding of user inputs, including those abbreviated or imprecise ones.

With the emergence of e-commerce systems, successful information access on e-commerce websites becomes essential. Menu-driven navigation and keyword search currently provided by most commercial sites have considerable limitations, as they tend to overwhelm and frustrate users with lengthy, rigid and not very effective interactions. To provide an efficient solution for information access, we have built the Natural Language Assistant (NLA), a web-based natural language dialog system to help users find relevant products on e-commerce sites. The system brings together technologies in natural language processing and human computer interaction to create a faster and more intuitive way of interacting with web sites. By combining statistical parsing techniques with traditional AI rule-based technology, we have created a dialog system that accommodates both customer needs and business requirements. The system is currently embedded in an application for recommending laptops and was deployed as a pilot on IBM’s website.

Abstract—This paper proposes two methods to incorporate semantic information into word and concept level confidence measurement. The first method uses tag and extension probabilities obtained from a statistical classer and parser. The second method uses a maximum entropy based semantic structured language model to assign probabilities to each word. Incorporation of semantic features into a lattice posterior probability based confidence measure provides significant improvements compared to posterior probability when used together in an air travel reservation task. At 5% False Alarm (FA) rate relative improvements of 28 % and 61 % in Correct Acceptance (CA) rate are achieved for word level and concept level confidence measurements, respectively. I.

State-of-the art parsers use lexicalized grammars to achieve high accuracy on real-world texts. Most of these systems are based on traditional parsing algorithms that were originally developed for the unlexicalized versions of the adopted grammar formalisms. We show that these parsing algorithms do not perform efficiently when processing lexicalized grammars. We then develop novel parsing algorithms that overcome the computational inefficiencies of the standard algorithms .