Caroline Sabty


Enhancing Deep Learning with Embedded Features for Arabic Named Entity Recognition
Ali L. Hatab | Caroline Sabty | Slim Abdennadher
Proceedings of the Thirteenth Language Resources and Evaluation Conference

The introduction of word embedding models has remarkably changed many Natural Language Processing tasks. Word embeddings can automatically capture the semantics of words and other hidden features. Nonetheless, the Arabic language is highly complex, which results in the loss of important information. This paper uses Madamira, an external knowledge source, to generate additional word features. We evaluate the utility of adding these features to conventional word and character embeddings to perform the Named Entity Recognition (NER) task on Modern Standard Arabic (MSA). Our NER model is implemented using Bidirectional Long Short Term Memory and Conditional Random Fields (BiLSTM-CRF). We add morphological and syntactical features to different word embeddings to train the model. The added features improve the performance by different values depending on the used embedding model. The best performance is achieved by using Bert embeddings. Moreover, our best model outperforms the previous systems to the best of our knowledge.


Contextual Embeddings for Arabic-English Code-Switched Data
Caroline Sabty | Mohamed Islam | Slim Abdennadher
Proceedings of the Fifth Arabic Natural Language Processing Workshop

Globalization has caused the rise of the code-switching phenomenon among multilingual societies. In Arab countries, code-switching between Arabic and English has become frequent, especially through social media platforms. Consequently, research in Natural Language Processing (NLP) systems increased to tackle such a phenomenon. One of the significant challenges of developing code-switched NLP systems is the lack of data itself. In this paper, we propose an open source trained bilingual contextual word embedding models of FLAIR, BERT, and ELECTRA. We also propose a novel contextual word embedding model called KERMIT, which can efficiently map Arabic and English words inside one vector space in terms of data usage. We applied intrinsic and extrinsic evaluation methods to compare the performance of the models. Our results show that FLAIR and FastText achieve the highest results in the sentiment analysis task. However, KERMIT is the best-achieving model on the intrinsic evaluation and named entity recognition. Also, it outperforms the other transformer-based models on question answering task.