Download Advances in Natural Language Processing: 7th International by Jan Hajič (auth.), Hrafn Loftsson, Eiríkur Rögnvaldsson, PDF

By Jan Hajič (auth.), Hrafn Loftsson, Eiríkur Rögnvaldsson, Sigrún Helgadóttir (eds.)

This e-book constitutes the complaints of the seventh foreign convention on Advances in ordinary Language Processing held in Reykjavik, Iceland, in August 2010.

Show description

Read or Download Advances in Natural Language Processing: 7th International Conference on NLP, IceTAL 2010, Reykjavik, Iceland, August 16-18, 2010 PDF

Best international books

Ocean Space Utilization ’85: Proceedings of the International Symposium Nihon University, Tokyo, Japan, June 1985 Volume 2

Ocean improvement has conventionally been precise on the exploitation of typical assets, even if this development is steadily altering: Ocean house has itself become considered as a priceless source. due to the fact difficulties linked to power, meals provide, and inhabitants turns into much more an important over the arriving years, ocean area is being reevaluated as a method for delivering recommendations in lots of of those components.

International Commodity Market Models and Policy Analysis

O. Guvenen, college of Paris IX-Dauphine the purpose of this booklet is to give contemporary advancements in overseas com­ modity marketplace version development and coverage research. This publication is predicated quite often at the study provided on the XlIth foreign convention organised through the utilized Econometric organization (AEA) which was once held on the college of Zaragoza in Spain.

Additional info for Advances in Natural Language Processing: 7th International Conference on NLP, IceTAL 2010, Reykjavik, Iceland, August 16-18, 2010

Sample text

This paper is focused on the statistical analysis. We have used R free software environment10 for statistical computing and graphics of the learners’ results. 1 Item Analysis and Distractor Evaluation The analysis of item responses in a quantitative way provides descriptions of item characteristics and test score properties among others. There are two main theories to address the problem: Classical Test Theory (CTT) and Item Response Theory (IRT). Both statistical theories have been already used in the evaluation of the automatic generation of distractors [3], [5].

It has also been applied in educational applications [8] and in the evaluation of synonym test questions [12]. Our system makes use of Infomap software [13]. This software uses a variant of LSA to learn vectors representing the meanings of words in a vector-space known as WordSpace. In our case, it indexes the documents in the corpora it processes and performs word to word semantic similarity computations based on the resulting model. As a result, the system extracts the words that best match a query according to the model.

Training required a very long time to converge, before introduction of phrase-based features, about 90 hours were necessary to train with the whole in-domain dataset. After the introduction of this kind of features, traning time decreased to about 40 hours. During the experiments, a real risk of local minima was detected. Performance of both rule based (baseline) and CRFs based reranking systems were evaluated in terms of accuracy, F-measure, precision and recall. The Table 1 shows the baseline rule system performance: rules perform well on in-domain sentences, while on out-of-domain sentences, the performance dramatically drops by losing 13,18% on accuracy and 8,89% on F-measure.

Download PDF sample

Rated 4.18 of 5 – based on 8 votes