Kfir Bar

Automatically Classifying Documents by Ideological and Organizational Affiliation

We show how an Arabic language religious-political document can be automatically classified according to the ideological stream and organizational affiliation that it represents. Tests show that our methods achieve near-perfect accuracy.

SLS at SemEval-2016 Task 3: Neural-based Approaches for Ranking in Community Question Answering.

Community question answering platforms need to automatically rank answers and questions with respect to a given question. In this paper, we present the approaches for the Answer Selection and Question Retrieval tasks of SemEval-2016 (task 3). We develop a bag-of-vectors approach with various vectorand text-based features, and different neural network approaches including CNNs and LSTMs to capture the semantic similarity between questions and answers for ranking purpose. Our evaluation demonstrates that our approaches significantly outperform the baselines.

The Tel Aviv University System for the Code-Switching Workshop Shared Task

We describe our entry in the EMNLP 2014 code-switching shared task. Our system is based on a sequential classifier, trained on the shared training set using various characterand word-level features, some calculated using a large monolingual corpora. We participated in the Twitter-genre Spanish-English track, obtaining an accuracy of 0.868 when measured on the tweet level and 0.858 on the word level.

Arabic Multiword Expressions

In this work we address the problem of automatic multiword expression identification and classification in Arabic running text. We propose a supervised machine learning approach using a relatively small manually annotated data augmented with an increasing size of automatically tagged data, labeled using a deterministic pattern-matching algorithm. In particular, in this chapter, we show the impact of explicitly modeling morpho-syntactic features calculated on the detection task. Moreover, we present the first work to address the problem of handling gapped verb-noun constructions in running text. We show that using the syntactic construction classes as labels improves identification results for verb-noun and verb-particle constructions. Our best identification algorithm yields an F-measure of 61.4%, which is a significant improvement over our baseline of 48.8%.

Inferring Paraphrases for a Highly Inflected Language from a Monolingual Corpus

We suggest a new technique for deriving paraphrases from a monolingual corpus, supported by a relatively small set of comparable documents. Two somewhat similar phrases that each occur in one of a pair of documents dealing with the same incident are taken as potential paraphrases, which are evaluated based on the contexts in which they appear in the larger monolingual corpus. We apply this technique to Arabic, a highly inflected language, for improving an Arabic-to-English statistical translation system. The paraphrases are provided to the translation system formatted as a word lattice, each assigned with a score reflecting its equivalence level. We experiment with the system on different configurations, resulting in encouraging results: our best system shows an increase of 1.73 5.49% in BLEU.

Matching Phrases for Arabic-to-English Example-Based Translation System

An implementation of a non-structural example-based translation system that translates sentences from Arabic to English, using a bilingual parallel corpus, is described. Each new input sentence is fragmented into phrases, and those phrases are matched to example patterns, using various levels of morphological data. We study the effect of forcing the system to match only fragments that do not break base phrases in the middle, and the results for small corpora are encouraging.