Description and contextualization of the subject

The objective of the subject is to obtain the ability to implement applications based on language technologies/natural language processing. During the course we will learn the basic applications of Natural Language Processing that are currently used in the industry of language technology industry..

The content will focus on the following tasks:



i) Information extraction: We present advanced techniques of lexical disambiguation of multiple various linguistic levels. Disambiguation techniques include word sense disambiguation algorithms, entity linking, and recognition and classification of named entities (NERC). We will learn and implement structured information extraction algorithms, as well as semantic relation and event extraction. For this, the student will be able to use advanced techniques of Deep Learning (embeddings, transfer learning, LSTM, CNN, etc.), sequence labeling (inference, beam search, viterbi, etc.) and distant supervision.



ii) Question Answering: We present unsupervised learning techniques based on semantic textual similarity (embeddings, graph theory), and techniques based on supervised algorithms that include end-to-end methods, information retrieval, and knowledge acquisition . Language generation techniques will also be studied (e.g. language models, seq2seq). The latest advances in multimodal tasks will be studied (e.g. visual question answering)



iii) Conversational systems: We will learn the modules that define a conversational systems, and the algorithms that control the interaction between human and machine. Special emphasis will be given to the natural language comprehension module (NLU) as well as the language generation.

Learning outcomes of the subject

Ability to implement applications based on language technologies.



Knowledge about basic tools of natural language processing.



Ability to use and code applications that use advanced methods such as Deep Learning and Machine Learning.



Knowledge to evaluate NLP applications.

Ordinary call: orientations and renunciation

1) 25% of the grading is based on class assignments: these will be typically performed in class hours, and will be handed in using e-gela. The student who hands-in approximately correct solutions to all assignments will be assigned full credit.



2) 75% of the grading is on three projects, each related to main tasks introduced in the course: each student will select or propose a subject for the project to one of the lecturers, depending on his/her interests. If a student does not propose any subject, the lecturers will assign a final project subject to him/her. The final project will be graded based on an oral presentation and written report., with the following percentages:



- write-up 30%, including features like clarity, structure,background, references, discussion

- technical 40%, including features like correctness and depth ofthe work

- presentation 30%, including clarity, structure, discussion

Extraordinary call: orientations and renunciation

In case the class assignments are missing or clearly deficient, the student will be evaluated depending on the three projects.

Temary

1. Introduction to NLP applications.



2. Information Extraction and Disambiguation techniques.



3. Question Answering systems.



4. Conversational systems.

Bibliography

Basic bibliography

Chris Manning and Hinrich Schütze. Foundations of Statistical Natural Language Processing, MIT Press. Cambridge, MA: May 1999.

Daniel Jurafsky, James H. Martin. Speech and Language Processing (2nd Edition), Pearson. May, 2008.

Eneko Agirre, Phillip Edmonds (Eds.). Word Sense Disambiguation: Algorithms and applications, Springer. 2007

Yoav Goldberg, Graeme Hirst. Neural Network Methods in Natural Language Processing (Synthesis Lectures on Human Language Technologies). 2017. Primer. http://u.cs.biu.ac.il/~yogo/nnlp.pdf

ACL: https://aclweb.org/anthology/