Cristina Giannone

Data-Driven Dialogue for Interactive Question Answering

In this paper, a light framework for dialogue based interactive question answering is presented. The resulting architecture is called REQUIRE(Robust Empirical QUestion answering for Intelligent Retrieval), and represents a flexible and adaptive platform for domain specific dialogue. REQUIRE characterizes as a domain-driven dialogue system, whose aim is to support the specific tasks evoked by interactive question answering scenarios. Among its benefits it should be mentioned its modularityand portabilityacross different domains, its robustnessthrough adaptive models of speech act recognition and planning and its adherence of knowledge representation standard. The framework will be exemplified through its application within a sexual health information service tailored to young people.

Ontological modeling for interactive question answering

This paper proposes a model for ontological representation supporting task-oriented dialog. The adoption of our ontology representation allows to map an interactive Question Answering (iQA) task into a knowledge based process. It supports dialog control, speech act recognition, planning and natural language generation through a unified knowledge model. A platform for developing iQA systems in specific domains, called REQUIRE(Robust Empirical QUestion answering for Intelligent Retrieval), has been entirely developed over this model. The first prototype developed for medical consulting in the sexual health domain has been recently deployed and is currently under testing. This will serve as a basis for exemplifying the model and discussing its benefits.

Supervised semantic relation mining from linguistically noisy text documents

In this paper, we present models for mining text relations between named entities, which can deal with data highly affected by linguistic noise. Our models are made robust by: (a) the exploitation of state-of-the-art statistical algorithms such as support vector machines (SVMs) along with effective and versatile pattern mining methods, e.g. word sequence kernels; (b) the design of specific features capable of capturing long distance relationships; and (c) the use of domain prior knowledge in the form of ontological constraints, e.g. bounds on the type of relation arguments given by the semantic categories of the involved entities. This property allows for keeping small the training data required by SVMs and consequently lowering the system design costs. We empirically tested our hybrid model in the very complex domain of business intelligence, where the textual data are constituted by reports on investigations into criminal enterprises based on police interrogatory reports, electronic eavesdropping and wiretaps. The target relations are typically established between entities, as they are mentioned in these information sources. The experiments on mining such relations show that our approach with small training data is robust to non-conventional languages as dialects, jargon expressions or coded words typically contained in such text.

Learning Semantic Roles for Ontology Patterns

An ontology learning method, based on large scale linguistic ontologies, such as FrameNet and WordNet, is here discussed. A robust learning method is defined to assign semantic roles to domain specific grammatical patterns according to distributional models of lexical semantics. Large scale experimental results over an IE task show an accuracy around 85-90%.

Kernel-based relation extraction from investigative data

In a specific process of business intelligence, i.e. investigation on organized crime, empirical language processing technologies can play a crucial role. In the data used on investigative activities, such as police interrogatory or electronic eavesdropping and wiretap, it is customary to find out expressions in non conventional languages as dialects, slangs or coded words. The recognition and storage of complex relations among subjects mentioned in these sources is a very difficult and time consuming task, ultimately based on pools of experts. We discuss here an inductive relation extraction platform that opens the way to much cheaper and consistent workflows. SVMs here are employed to produce a set of possible interpretations for domain relevant concepts. An ontology population process is here realized, where further reasoning can be applied to proof the overall consistency of the extracted information. The empirical investigation presented here shows that accurate results, comparable to the expert teams, can be achieved, and parametrization allows to fine tune the system behavior for fitting the specific domain requirements.

Kernel-Based Learning for Domain-Specific Relation Extraction

In a specific process of business intelligence, i.e. investigation on organized crime, empirical language processing technologies can play a crucial role. The analysis of transcriptions on investigative activities, such as police interrogatories, for the recognition and storage of complex relations among people and locations is a very difficult and time consuming task, ultimately based on pools of experts. We discuss here an inductive relation extraction platform that opens the way to much cheaper and consistent workflows. The presented empirical investigation shows that accurate results, comparable to the expert teams, can be achieved, and parametrization allows to fine tune the system behavior for fitting domain-specific requirements.

Latent topic models of surface syntactic information

Topic Models like Latent Dirichlet Allocation have been widely used for their robustness in estimating text models through mixtures of latent topics. Although LDA has been mostly used as a strictly lexicalized approach, it can be effectively applicable to a much richer set of linguistic structures. A novel application of LDA is here presented that acquires suitable grammatical generalizations for semantic tasks tightly dependent on NL syntax. We show how the resulting topics represent suitable generalizations over syntactic structures and lexical information as well. The evaluation on two different classification tasks, such as predicate recognition and question classification, shows that state of the art results are obtained.

Acquiring IE patterns through distributional lexical semantic models

Techniques for the automatic acquisition of Information Extraction Pattern are still a crucial issue in knowledge engineering. A semi supervised learning method, based on large scale linguistic resources, such as FrameNet and WordNet, is discussed. In particular, a robust method for assigning conceptual relations (i.e. roles) to relevant grammatical structures is defined according to distributional models of lexical semantics over a large scale corpus. Experimental results show that the use of the resulting knowledge base provide significant results, i.e. correct interpretations for about 90% of the covered sentences. This confirms the impact of the proposed approach on the quality and development time of large scale IE systems.

Semantic Word Spaces for Robust Role Labeling

Semantic role labeling systems are often designed as inductive processes over annotated resources. Supervised algorithms based on complex grammatical information achieve state-of-the-art accuracy. However, their generalization on the argument classification task is poorer, as large performance drops in out-of-domain tests showed. In this paper, a robust method based on a minimal set of grammatical features and a distributional model of lexical semantic information is proposed. The achievable generalization ability is studied in several training conditions where negligible performance drops are observed.

A Robust Geometric Model for Argument Classification

Argument classification is the task of assigning semantic roles to syntactic structures in natural language sentences. Supervised learning techniques for frame semantics have been recently shown to benefit from rich sets of syntactic features. However argument classification is also highly dependent on the semantics of the involved lexicals. Empirical studies have shown that domain dependence of lexical information causes large performance drops in outside domain tests. In this paper a distributional approach is proposed to improve the robustness of the learning model against out-of-domain lexical phenomena.