Workshops

Session 2  Corpus analysis tools

1- Build an Arabic Part of speech (POS) tagging using Hidden Markov Model (HMM)

It aims to help audience to build a POS tagger for modern standard Arabic using HMM model . It will be hands on for linguists and engineers. The session's target is beginners mainly. It will cover:

• Brief introduction to Part of Speech tagging (POS) .

The talk will illustrate what is part of speech tagging and its challenges in Arabic.

• Introduction to Hidden Markov Model.

This part will introduce the hidden marcov model to the audience.

• A hands-on implementation of the part of speech tagging using HMM

           This part will show audience how to build the pos tagger using python and sklearn.

• Test the built model to find out its accuracy.

2- Build your Parser in ten minutes.

It aims to help audience to build a probabilistic context free grammar parser . It will be hands on for linguists and engineers. The session's target is beginners mainly. It will cover:

• Brief introduction to syntactic parsing and  probabilistic Context free grammar (PCFG).

The talk will illustrate what is syntactic parsing and its challenges in Arabic, and it will give a brief introduction to PCFG.

• Inducing grammar from The Arabic Penn Tree Bank (PATB)

This part will show a hands-on induction of the Arabic syntactic grammar from the gold standard annotated PATB.

• Implementing a Viterbi parser using  the induced grammar and testing the accuracy of the grammar.

           This part will show audience how to use the induced grammar with a parser and test the efficiency of the induced grammar.

It is recommended that you take your own laptop with you and be prepared to do the experiments yourself.

The invited speakers:

Prof. Sameh AlAnsary, Head of the Phonetics and Linguistics department, Faculty of Arts, Alexandria University. Amena Helmy , Assistant lecturer at the Phonetics and linguistics department, Faculty of Arts, Alexandria University. Nihal AlNazli, research assistant at the Phonetics and linguistics department, Faculty of Arts, Alexandria University.    Rehab Arafat, research assistant at the Phonetics and linguistics department, Faculty of Arts, Alexandria University.   

Second Workshop

Deep Learning for Natural Language Processing (NLP)          Arab Academy for Sciences Technology and Maritime Transport , Abu Kir, Alexandria, 29 September 2020

This workshop is concerned, basically, with the deep learning for Arabic NLP. It will be hands on for linguists and engineers who are interested in building NLP models and applications. It will be held at the College of Artificial Intelligence, Arab Academy for Sciences Technology and Maritime Transport (AASTMT), El Alamein new campus, north coast of Egypt.

Also, due to force majeure incurred by limitations set on the hosting capacity of available seminar halls mandated by the authorities to face any potential spread of the Korona Virus, the workshop is dislocated to AASTMT Main Campus in Abu Kir in Alexandria.

The workshop will cover the following topics:

1. Brief introduction to neural networks (Dr. Hanaa Bayoumi).

The talk will illustrate how neural networks learn a certain function by analyzing input data through simple illustrative examples

2. Distributive representation of texts (Word embedding) (Dr. A. Sarah Hassan).

Embedding means words or phrases from the vocabulary are mapped to vectors of real numbers. that words that are used in similar contexts will be given similar numerical vectors. These vectors will be placed close together within the high-dimensional semantic space. They will cluster together, and their distance to each other will be low.

We will describe the different algorithms used to drive theses representations.

3. Language modeling using transformers (Dr. Aly Fahmy).

Transformers are seen as the key breakthrough for the state-of-art performance of deep learning methods on challenging natural language processing problems. They are built around the notion of “Attention” introduced in the paper “Attention is All you Need”. It is considered as one of top Machine Learning papers of the decade

We will describe the structure of the transformers and emphasize what language characteristics are learned through attention.

4. Hands on BERT: A pre-trained model for NLP tasks (Eng. Islam Hassan).

BERT, Bidirectional Encoder Representations from Transformers, is a technique for NLP pre-training developed by Google. BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of NLP tasks such as Question-Answering, Named Entity Recognition, Textual Entailment, Coreference Resolution and many others.

We will provide hands-on fine-tuning pre-trained BERT to NLP tasks involving one or two input sentences such as grammatical checking task and paraphrasing checking task respectively. where the in single sentences.

It is recommended that you take your own laptop with you and be prepared to do the experiments yourself.

The Invited speakers:

Dr. Aly Fahmy, Dean of Artificial Intelligence College, Arab Academy for Sciences Technology and Maritime Transport (AASTMT) and his team. Dr. Fahmy was the Ex-Dean of Faculty of Computers and Artificial Intelligence, Cairo University.

Transportation

Transportation from Bibliotheca Alexandrina to Arab Academy Campus in Abu Kir's workshop will take place at 10:00 a.m. and will leave the Campus at 4:15 p.m back to Alexandrina Bibliotheca and Sidi Gaber train station. The transportation will be provided free by the Arab Academy for Sciences Technology and Maritime Transport.