473 research outputs found
Ontology Learning from Text: An Overview
No full textBuitelaar P, Cimiano P, Magnini B. Ontology Learning from Text: An Overview. In: Buitelaar P, Cimiano P, Magnini B, eds. Ontology Learning from Text: Methods, Evaluation and Applications. Frontiers in Artificial Intelligence and Applications. Vol 123. Amsterdam: IOS Press; 2005: 3-12
Use of a Lexical Knowledge Base for Information Access Systems
No full textThe role of generic lexical resources as well as specialized terminology is crucial in the design of complex dialogue systems, where a human interacts with the computer using Natural Language. Lexicon and terminology are supposed to store information for several purposes, including lexical discrimination of semantically inconsistent interpretations, the use of lexical variations, the compositional construction of a semantic representation for a complex sentence and the ability to access equivalencies across different languages. For these purposes it is necessary to rely on representational tools that are both theoretically motivated and operationally well defined. In this paper we propose a solution to lexical and terminology representation which is based on the combination of a linguistically motivated upper model and a multilingual wordnet. The upper model accounts for the linguistic analysis at the sentence level, while the multilingual WordNet accounts for lexical and conceptual relations at the word leve
Experiments in Word Domain Disambiguation for Parallel Texts
No full textThis paper describes some preliminary results about Word Domain Disambiguation, a variant of Word Sense Disambiguation where words in a text are tagged with a domain label in place of a sense label. The Englis Wordnet and its aligned Italian version, MultiWordNet, both augmented with domain labels, are used as the main information repositories. A baseline algorithm for Word Domain Disambiguation is presented and then compared with a mutual help disambiguation strategy, which takes advantages of the shared senses of parallel bilingual text
Merging Global and Specialized Linguistic Ontologies
No full textThere is an increasing interest in linguistic ontologies (e.g. WordNet) for a variety of content-based tasks, including conceptual indexing, word sense disambiguation and cross-language information retrieval. A relevant contribution in this direction is represented by linguistic ontologies with domain specific coverage, which are a crucial topic for the development of concrete application systems. This paper tries to go a step further in the direction of the interoperability of specialized linguistic ontologies, by addressing the problem of their integration with global ontologies. This scenario poses some simplifications with respect to the general problem of merging ontologies, since it enables to define a strong precedence criterion so that terminological information overshadows generic information whenever conflicts arise. We assume the EuroWordNet model and propose a methodology to `plug` specialized linguistic ontologies into global ontologies. Experimental data related to an implemented algorithm, which has been tested on a global and a specialized linguistic ontology for the Italian language, are provide
Optimizing textual entailment recognition using particle swarm optimization
No full textThis paper introduces a new method to improve
tree edit distance approach to textual
entailment recognition, using particle
swarm optimization. Currently, one of the
main constraints of recognizing textual entailment
using tree edit distance is to tune
the cost of edit operations, which is a difficult
and challenging task in dealing with
the entailment problem and datasets. We
tried to estimate the cost of edit operations
in tree edit distance algorithm automatically,
in order to improve the results for
textual entailment. Automatically estimating
the optimal values of the cost operations
over all RTE development datasets,
we proved a significant enhancement in
accuracy obtained on the test sets
Scalable Neural Dialogue State Tracking
No full textA Dialogue State Tracker (DST) is a key component in a dialogue system aiming at estimating the beliefs of possible user goals at each dialogue turn. Most of the current DST trackers make use of recurrent neural networks and are based on complex architectures that manage several aspects of a dialogue, including the user utterance, the system actions, and the slot-value pairs defined in a domain ontology. However, the complexity of such neural architectures incurs into a considerable latency in the dialogue state prediction, which limits the deployments of the models in real-world applications, particularly when task scalability (i.e. amount of slots) is a crucial factor. In this paper, we propose an innovative neural model for dialogue state tracking, named Global encoder and Slot-Attentive decoders (G-SAT), which can predict the dialogue state with a very low latency time, while maintaining high-level performance. We report experiments on three different languages (English, Italian, and German) of the WOZ2.0 dataset, and show that the proposed approach provides competitive advantages over state-of-art DST systems, both in terms of accuracy and in terms of time complexity for predictions, being over 15 times faster than the other systems
Simple Data Augmentation for Multilingual NLU in Task Oriented Dialogue Systems
No full textData augmentation has shown potential in alleviating data scarcity for Natural Language Understanding (e.g. slot filling and intent classification) in task-oriented dialogue systems. As prior work has been mostly experimented on English datasets, we focus on five different languages, and consider a setting where limited data are available. We investigate the effectiveness of non-gradient based augmentation methods, involving simple text span substitutions and syntactic manipulations. Our experiments show that (i) augmentation is effective in all cases, particularly for slot filling; and (ii) it is beneficial for a joint intent-slot model based on multilingual BERT, both for limited data settings and when full training data is used
- …
