This paper describes our approach to the development of a Proposition Bank, which involves the addition of semantic information to the Penn English Treebank. Our primary goal is the labeling of syntactic nodes with specific argument labels that preserve the similarity of roles such as the window in John broke the window and the window broke. After motivating the need for explicit predicate argument structure labels, we briefly discuss the theoretical considerations of predicate argument structure and the need to maintain consistency across syntactic alternations. The issues of consistency of argument structure across both polysemous and synonymous verbs are also discussed and we present our actual guidelines for these types of phenomena, along with numerous examples of tagged sentences and verb frames. Metaframes are introduced as a technique for handling similar frames among near− synonymous verbs. We conclude with a summary of the current status of annotation process.

/pdf/from-treebank-to-propbank-3v7y64dt8d.pdf

From treebank to propbank

The popularity of the Web and the large number of documents available in electronic form has motivated the search for hidden knowledge in text collections. Consequently, there is growing research interest in the general topic of text mining. In this dissertation, we develop a text-mining system by integrating methods from Information Extraction (IE) and Data Mining (Knowledge Discovery from Databases or KDD). By utilizing existing IE and KDD techniques, text-mining systems can be developed relatively rapidly and evaluated on existing text corpora for testing IE systems. 
We present a general text-mining framework called DISCOTEX which employs an IE module for transforming natural-language documents into structured data and a KDD module for discovering prediction rules from the extracted data. When discovering patterns in extracted text, strict matching of strings is inadequate because textual database entries generally exhibit variations due to typographical errors, misspellings, abbreviations, and other sources. We introduce the notion of discovering “soft-matching” rules from text and present two new learning algorithms. TEXTRISE is an inductive method for learning soft-matching prediction rules that integrates rule-based and instance-based learning methods. Simple, interpretable rules are discovered using rule induction, while a nearest-neighbor algorithm provides soft matching. SOFTAPRIORI is a text-mining algorithm for discovering association rules from texts that uses a similarity measure to allow flexible matching to variable database items. We present experimental results on inducing prediction and association rules from natural-language texts demonstrating that TEXTRISE and SOFTA PRIORI learn more accurate rules than previous methods for these tasks. We also present an approach to using rules mined from extracted data to improve the accuracy of information extraction. Experimental results demonstrate that such discovered patterns can be used to effectively improve the underlying IE method.

/pdf/text-mining-with-information-extraction-7g2sbtyl7d.pdf

Text mining with information extraction

Recently, statistical machine translation models have begun to take advantage of higher level linguistic structures such as syntactic dependencies. Underlying these models is an assumption about the directness of translational correspondence between sentences in the two languages; however, the extent to which this assumption is valid and useful is not well understood. In this paper, we present an empirical study that quantifies the degree to which syntactic dependencies are preserved when parses are projected directly from English to Chinese. Our results show that although the direct correspondence assumption is often too restrictive, a small set of principled, elementary linguistic transformations can boost the quality of the projected Chinese parses by 76% relative to the unimproved baseline.

http://www.umiacs.umd.edu/~weinberg/acl02.pdf

Evaluating Translational Correspondence using Annotation Projection

The exploitation of information extraction (IE), a technology aiming to provide instances of structured representations from free-form text, has been rapidly growing within the molecular biology (MB) research community to keep track of the latest results reported in literature. IE systems have traditionally used shallow syntactic patterns for matching facts in sentences but such approaches appear inadequate to achieve high accuracy in MB event extraction due to complex sentence structure. A consensus in the IE community is emerging on the necessity for exploiting deeper knowledge structures such as through the relations between a verb and its arguments shown by predicate-argument structure (PAS). PAS is of interest as structures typically correspond to events of interest and their participating entities. For this to be realized within IE a key knowledge component is the definition of PAS frames. PAS frames for non-technical domains such as newswire are already being constructed in several projects such as PropBank, VerbNet, and FrameNet. Knowledge from PAS should enable more accurate applications in several areas where sentence understanding is required like machine translation and text summarization. In this article, we explore the need to adapt PAS for the MB domain and specify PAS frames to support IE, as well as outlining the major issues that require consideration in their construction. We introduce PASBio by extending a model based on PropBank to the MB domain. The hypothesis we explore is that PAS holds the key for understanding relationships describing the roles of genes and gene products in mediating their biological functions. We chose predicates describing gene expression, molecular interactions and signal transduction events with the aim of covering a number of research areas in MB. Analysis was performed on sentences containing a set of verbal predicates from MEDLINE and full text journals. Results confirm the necessity to analyze PAS specifically for MB domain. At present PASBio contains the analyzed PAS of over 30 verbs, publicly available on the Internet for use in advanced applications. In the future we aim to expand the knowledge base to cover more verbs and the nominal form of each predicate.

https://link.springer.com/content/pdf/10.1186%2F1471-2105-5-155.pdf

PASBio: predicate-argument structures for event extraction in molecular biology.

Most existing systems for subcategorization frame (SCF) acquisition rely on supervised parsing and infer SCF distributions at type, rather than instance level. These systems suffer from poor portability across domains and their benefit for NLP tasks that involve sentence-level processing is limited. We propose a new unsupervised, Markov Random Field-based model for SCF acquisition which is designed to address these problems. The system relies on supervised POS tagging rather than parsing, and is capable of learning SCFs at instance level. We perform evaluation against gold standard data which shows that our system outperforms several supervised and type-level SCF baselines. We also conduct task-based evaluation in the context of verb similarity prediction, demonstrating that a vector space model based on our SCFs substantially outperforms a lexical model and a model based on a supervised parser 1 .

/pdf/an-unsupervised-model-for-instance-level-subcategorization-4n26wyocpu.pdf

An Unsupervised Model for Instance Level Subcategorization Acquisition

This paper describes an approach for handling structural divergences and recovering dropped arguments in an implemented Korean to English machine translation system. The approach relies on canonical predicate-argument structures (or dependency structures), which provide a suitable pivot representation for the handling of structural divergences and the recovery of dropped arguments. It can also be converted to and from the interface representations of many off-the-shelf parsers and generators.

/pdf/handling-stuctural-divergences-and-recovering-dropped-1wlf6cfqey.pdf

Handling Stuctural Divergences and Recovering Dropped Arguments in a Korean/English Machine Translation System

This document describes an on-going project of developing a grammar of Korean, the Korean XTAG grammar, written in the TAG formalism and implemented for use with the XTAG system enriched with a Korean morphological analyzer The Korean XTAG grammar described in this report is based on the TAG formalism (Joshi et al (1975)), which has been extended to include lexicalization (Schabes et al (1988)), and unification-based feature structures (Vijay-Shanker and Joshi (1991)) The document first describes the modifications that we have made to the XTAG system (The XTAG-Group (1998)) to handle rich inflectional morphology in Korean Then various syntactic phenomena that can be currently handled are described, including adverb modification, relative clauses, complex noun phrases, auxiliary verb constructions, gerunds and adjunct clauses The work reported here is a first step towards the development of an implemented TAG grammar for Korean, which is continuously updated with the addition of new analyses and modification of old ones

/pdf/a-feature-based-lexicalized-tree-adjoining-grammar-for-30eeudbkc4.pdf

A Feature-Based Lexicalized Tree Adjoining Grammar for Korean

This paper addresses linguistic and implementation problems for a practical LTAG parser raised by rich morphology in Korean. We propose a way of representing the Korean inftectional system asfeature structures in lexicalized elementary trees, and describe our implemented modijicarions on the XTAG system for a more efficient grammar development for Korean. 1. Issues Korean is an agglutinative language with a very productive infiectional system. lnfiections include postpositions on nouns; tense morphemes and endings that indicate sentence types on verbs and adjectives; among others. Furthermore, these inftections can combine with each other to form compound infiections. (1) Noun (2) Verb a. hakkyo-ka a. ka-ss-ta school-Nom go-Past-Decl b. hakkyo-eyse-ka b. ka-si-ess-ta school-from-Nom go-Honor-Past-Decl c. hakkyo-eyse-num c. ka-ki-ka school-from-only · go-Nominalizer-Nom d. hakkyo·eyse·man-un d. ka·si-ess-ki-ey-nun school-from-only-Topic go-Honor-Past-Nominalizer-to-Topic This implies that a word in Korean can have a very !arge number of morphological variants. For example, verbs can be followed by honorific and tense mo1phemes which can then be followed by endings indicating clause-type which then can be followed by case postpositions. Similarly, adverbial postpositions which correspond to English prepositions, can be followed by other case postpositions such as nominative or accusative case markers and auxiliary postpositions such as to ('also') and man ('only'), which then can be followed by a topic marker. Accordingly, 'We thank Martha Palmer, Aravind Joshi, Anoop Sarkar and the XTAG Group at Penn for their support and discussion. Wc also acknowledge the two anonymous rcviewers. This projcct has bcen partially funded by the Anny Research Lab via a subcontract from CoGenTex, Inc. and NSF Grant SBR 8920230. The third author's contribution to this work was made while she was n visiting researchcr at m.CS. 222 J. Yoon, C. Ran, N. Kim and M. Kim the number of possible morphological variants of a word can in principle be in the tens of thousands. This property of Korean raises two issues within the context of developing and implementing a Feature Based Lexicalized Tree Adjoining Grammar (FB-LTAG) for Korean using the XTAG system (The XTAG-Group, 1998): (1) adequate linguistic description of the inflections and (2) efficient Jexicon development. From a linguistic point of view, describing a grammar of a Janguage is to construct rules that generate sentences in the language at a formal level. From an implementational point of view, the grammar should be described in a consistent and efficient way. The XTAG system helps US to pursue both these goals, but the complicated inflection system mentioned above leads to difficulties in building a grammar for Korean. In this paper, we provide our solution to the linguistic and implementational issues raised by these morphological properties of Korean. We first provide a way of handling the Korean inflectional system using feature structures in lexicalized elementary trees in section 2. We impose a hierarchy on various types of inflections in order to handle all possible ways of combining inflections, and we represent this by assigning different feature attributes to different types of inflections. In section 3, we then point out that the current XTAG system as it is forces us to construct a lexicon (i.e., syntactic database) that lists all possible morphological variants of words. A lexicon must contain all possible eojeols, where an eojeol is a terrn in Korean for denoting a spacing unit which consists of a content word and associated functional words. However, this is highly impractical and inefficient given the rich inflectional system in Korean. We would end up with a very !arge (even unbounded) Jexicon. Therefore, we found it necessary to develop an alternative method for constructing the lexicon in order to continue to use the XTAG parser for developing a Korean grammar . One possible solution to the problem is to incorporate morphological rules in the grammar that regulate the generation of eojeols with several morphemes combined. However, doing so will mix up morphological generative rules with syntactic rules, complicating the TAG grammar tremendously. Instead, we have chosen to pursue an approach in which morphological regularities are handled by a separate morphological component using a morphological analyzer (Yoon et al„ 1999). The output of this analysis then interacts with our Korean TAG grammar which handles syntactic regularities. As a way of implementing this approach, we modified the XTAG system by dividing up the syntactic database into elementary syntactic database.(ESDB) and local syntactic database (LSDB). ESDB is a general lexicon that contains stems with the elementary trees associated with them. LSDB is a partial lexicon dynamically generated for each input sentence using infonnation from ESDB and the output of a morphological analyzer. That is, it contains only entries for eojeols occurring in the input sentence. The morphological analyzer produces the morphological analysis of each eojeol in the input sentence identifying its stem and inflections. Then, the stem of each eojeol is associated with elementary trees or tree families by looking up the ESDB and stored in the LSDB. The inftections of each eojeol are converted into features and are also stored in the LSDB. This modification to the XTAG system allows us to build a Jexicon efficiently and develop a grammar for Korean that is compatible with the XTAG system. 2. Handling inflectional morphology In our current Korean grammar, the inflectional morphology on an eojeol that are relevant for syntactic analysis is represented as features on the tree node. For instance, a noun with a nominative case marker is associated with the feature <case:nom> and when this Jexical item is anchored by an NP tree, the feature <case:nom> is passed up to the NP node. In Korean, combining inftections is a highly productive process with some restrictions. For

Customizing the XTAG system for efficient grammar development for Korean

We report on a small study undertaken to demonstrate the feasibility of combining portable information extraction with MT in order to support translingual information access. After describing the proposed system's usage scenario and system design, we describe our investigation of transferring information extraction techniques developed for English to Korean. We conclude with a brief discussion of related MT issues we plan to investigate in future work.

/pdf/towards-translingual-information-access-using-portable-3wofyk5wbd.pdf

Nari Kim

Papers

Handling Stuctural Divergences and Recovering Dropped Arguments in a Korean/English Machine Translation System

A Feature-Based Lexicalized Tree Adjoining Grammar for Korean

Customizing the XTAG system for efficient grammar development for Korean

Towards translingual information access using portable information extraction