John Judge

QuestionBank: Creating a Corpus of Parse-Annotated Questions

This paper describes the development of QuestionBank, a corpus of 4000 parse-annotated questions for (i) use in training parsers employed in QA, and (ii) evaluation of question parsing. We present a series of experiments to investigate the effectiveness of QuestionBank as both an exclusive and supplementary training resource for a state-of-the-art parser in parsing both question and non-question test sets. We introduce a new method for recovering empty nodes and their antecedents (capturing long distance dependencies) from parser output in CFG trees using LFG f-structure reentrancies. Our main findings are (i) using QuestionBank training data improves parser performance to 89.75% labelled bracketing f-score, an increase of almost 11% over the baseline; (ii) back-testing experiments on non-question data (Penn-II WSJ Section 23) shows that the retrained parser does not suffer a performance drop on non-question material; (iii) ablation experiments show that the size of training material provided by QuestionBank is sufficient to achieve optimal results; (iv) our method for recovering empty nodes captures long distance dependencies in questions from the ATIS corpus with high precision (96.82%) and low recall (39.38%). In summary, QuestionBank provides a useful new resource in parser-based QA research.

Very High Coercivity Chemically Deposited Co–Ni Films

Films of Co–Ni, varying in composition from 30% Co to 100% Co, and ranging in thickness from 250 to 2500 A, were prepared by chemical deposition. Small amounts of P were also included in the deposit. Their crystallographic and magnetic properties were studied as a function of composition and thickness. X‐ray diffraction revealed a hexagonal structure for the Co films, with increasing presence of a cubic phase as Ni was introduced. The hexagonal crystallites of the all‐Co films were randomly oriented, but in the Co–Ni films there was an increasing orientation of the hexagonal (002) and cubic (111) axes perpendicular to the substrate. The coercivity for a particular thickness increased as the Ni content of the films increased, reaching a peak of about 1300 Oe at 10% to 30% Ni depending on the thickness. Beyond this, the coercivity decreased rapidly to a value of 200 Oe as the Ni content of the films increased to 70%. In the Co films the coercivity decreases markedly with increasing thickness. This dependenc...

Navigating and Annotating Semantically-Enabled Networks of People and Associated Objects

Social spaces such as blogs, wikis and online social networking sites are enabling the formation of online communities where people are linked to each other through direct profile connections and also through the content items that they are creating, sharing and tagging. As these spaces become bigger and more distributed, more intuitive ways of navigating the associated information become necessary. The Semantic Web aims to link identifiable objects to each other and to textual strings via relationships and attributes respectively, and provides a platform for gathering diverse information from heterogeneous sources and performing operations on such linked data. In this paper, we will demonstrate how this linked semantic data can provide an enhanced view of the activity in a social network, and how the Galaxy tool described in this work can augment objects from social spaces, by highlighting related people and objects, and suggesting relevant sources of knowledge.

Dublin city university at CLEF 2004: experiments in monolingual, bilingual and multilingual retrieval

The Dublin City University group participated in the monolingual, bilingual and multilingual retrieval tasks. The main focus of our investigation for CLEF 2004 was extending our information retrieval system to document languages other than English, and completing the multilingual task comprising four languages: English, French, Russian and Finnish. Our retrieval system is based on the City University Okapi BM25 system with document preprocessing using the Snowball stemming software and stopword lists. Our French monolingual experiments compare retrieval using French documents and topics, and documents and topics translated into English. Our results indicate that working directly in French is more effective for retrieval than adopting document and topic translation. A breakdown of our multilingual retrieval results by the individual languages shows that similar overall average precision can be achieved when there is significant underlying variation in performance for individual languages.

Active Learning for Post-Editing Based Incrementally Retrained MT

Machine translation, in particular statistical machine translation (SMT), is making big inroads into the localisation and translation industry. In typical workflows (S)MT output is checked and (where required) manually post-edited by human translators. Recently, a significant amount of research has concentrated on capturing human post-editing outputs as early as possible to incrementally update/modify SMT models to avoid repeat mistakes. Typically in these approaches, MT and post-edits happen sequentially and chronologically, following the way unseen data (the translation job) is presented. In this paper, we add to the existing literature addressing the question whether and if so, to what extent, this process can be improved upon by Active Learning, where input is not presented chronologically but dynamically selected according to criteria that maximise performance with respect to (whatever is) the remaining data. We explore novel (source side-only) selection criteria and show performance increases of 0.67-2.65 points TER absolute on average on typical industry data sets compared to sequential PEbased incrementally retrained SMT.

High‐Coercivity Powders of Partially Reduced Metal Oxides

Recent magnetic recording studies have indicated the following modifications in particulate media to improve their recording performance: increase particle coercivity and magnetic moment density, and decrease coating thickness. In this study we attempted to modify the traditional iron oxide particles used in magnetic recording media by substitution of other ions such as Co2+ and Ni2+, and by partial or full reduction to the metallic state. The alloy powders were prepared by hydroxide precipitation of the mixed oxides with subsequent partial reduction at elevated temperature under hydrogen, and by oxalate precipitation followed by thermal decomposition to the oxides and hydrogen reduction to the metal. The particles thus produced consist of a partially reduced metal oxide enclosed by an outer oxide shell, with attractive magnetic properties, particularly in the Fe–Co–O system.

A linguistic light approach to multilingualism in lexical layers for ontologies

Semantic Web ontologies are being increasingly used in modern text analytics applications and ontology-based information extraction (OBIE) as a means to provide a semantic backbone either for modelling the internal conceptual data structures of the text analytics (TA) engine or to model the knowledge base to drive the analysis of unstructured information in raw text and subsequent Knowledge acquisition and population. creating and targeting language resources (LR)s from a TA to an ontology can be time consuming and costly.The authors describe a user-friendly method for ontology engineers to augment an ontologies with a lexical layer which provides a flexible framework to identify term mentions of ontology concepts in raw text. In this paper we explore multilinguality in these lexical layers using the same framework. We discuss a number of potential issues for the ldquolinguistic lightrdquo lexical extensions for ontologies (LEON) approach when looking at languages more morphologically rich and which have more complex linguistic constraints than English. We show how the LEON approach can cope with these phenomena once the morphological normaliser used in the lexical analysis process is able to generalise sufficiently well for the language concerned.