Ulf Hermjakob

Parsing and question classification for question answering

This paper describes machine learning based parsing and question classification for question answering. We demonstrate that for this type of application, parse trees have to be semantically richer and structurally more oriented towards semantics than what most treebanks offer. We empirically show how question parsing dramatically improves when augmenting a semantically enriched Penn treebank training corpus with an additional question treebank.

Toward semantics-based answer pinpointing

We describe the treatment of questions (Question-Answer Typology, question parsing, and results) in the Weblcopedia question answering system.

Learning Parse and Translation Decisions from Examples with Rich Context

We present a knowledge and context-based system for parsing and translating natural language and evaluate it on sentences from the Wall Street Journal. Applying machine learning techniques, the system uses parse action examples acquired under supervision to generate a deterministic shift-reduce parser in the form of a decision structure. It relies heavily on context, as encoded in features which describe the morphological, syntactic, semantic and other aspects of a given parse state.

Unsupervised Entity Linking with Abstract Meaning Representation

Most successful Entity Linking (EL) methods aim to link mentions to their referent entities in a structured Knowledge Base (KB) by comparing their respective contexts, often using similarity measures. While the KB structure is given, current methods have suffered from impoverished information representations on the mention side. In this paper, we demonstrate the effectiveness of Abstract Meaning Representation (AMR) (Banarescu et al., 2013) to select high quality sets of entity “collaborators” to feed a simple similarity measure (Jaccard) to link entity mentions. Experimental results show that AMR captures contextual properties discriminative enough to make linking decisions, without the need for EL training data, and that system with AMR parsing output outperforms hand labeled traditional semantic roles as context representation for EL. Finally, we show promising preliminary results for using AMR to select sets of “coherent” entity mentions for collective entity linking 1 .

Using knowledge to facilitate factoid answer pinpointing

In order to answer factoid questions, the Webclopedia QA system employs a range of knowledge resources. These include a QA Typology with answer patterns, WordNet, information about typical numerical answer ranges, and semantic relations identified by a robust parser, to filter out likely-looking but wrong candidate answers. This paper describes the knowledge resources and their impact on system performance.

Improved Word Alignment with Statistics and Linguistic Heuristics

We present a method to align words in a bitext that combines elements of a traditional statistical approach with linguistic knowledge. We demonstrate this approach for Arabic-English, using an alignment lexicon produced by a statistical word aligner, as well as linguistic resources ranging from an English parser to heuristic alignment rules for function words. These linguistic heuristics have been generalized from a development corpus of 100 parallel sentences. Our aligner, Ualign, outperforms both the commonly used GIZA++ aligner and the state-of-the-art LEAF aligner on F-measure and produces superior scores in end-to-end statistical machine translation, +1.3 Bleu points over GIZA++, and +0.7 over LEAF.

Aligning English Strings with Abstract Meaning Representation Graphs

We align pairs of English sentences and corresponding Abstract Meaning Representations (AMR), at the token level. Such alignments will be useful for downstream extraction of semantic interpretation and generation rules. Our method involves linearizing AMR structures and performing symmetrized EM training. We obtain 86.5% and 83.1% alignment F score on development and test sets.

Parsing English into Abstract Meaning Representation Using Syntax-Based Machine Translation

We present a parser for Abstract Meaning Representation (AMR). We treat Englishto-AMR conversion within the framework of string-to-tree, syntax-based machine translation (SBMT). To make this work, we transform the AMR structure into a form suitable for the mechanics of SBMT and useful for modeling. We introduce an AMR-specific language model and add data and features drawn from semantic resources. Our resulting AMR parser significantly improves upon state-of-the-art results.

How To Select An Answer String

Given a question Q and a sentence/paragraph SP that is likely to contain the answer to Q, an answer selection module is supposed to select the “exact” answer sub-string A ⊂ SP. We study three distinct approaches to solving this problem: one approach uses algorithms that rely on rich knowledge bases and sophisticated syntactic/semantic processing; one approach uses patterns that are learned in an unsupervised manner from the web, using computational biology-inspired alignment algorithms; and one approach uses statistical noisy-channel algorithms similar to those used in machine translation. We assess the strengths and weaknesses of these three approaches and show how they can be combined using a maximum entropy-based framework.

Cross-Language Question Answering at the USC Information Sciences Institute

The TextMap-TMT cross-language question answering system at USC-ISI was designed to answer Spanish questions from English documents. The system is fully automatic, including question translation from Spanish to English, question type determination, rewriting to generate expected answer structures, search in the target collection and on the Web as a side collection, and answer selection from among the plausible candidates that were found. A development test collection with answer patterns for 100 questions in English and Spanish was used to assess the effect of question translation on each processing stage, and some adjustments were made to the question translation process to minimize these effects. Two runs were submitted, both of which sought to return exact answers. For the better of the two runs (which omitted an additional Web-based answer validation stage), the top-ranked answer was scored as correct in 56 of 200 cases, 53 of which were judged to be supported by the content of the target collection.