Computational processing of written and spoken natural language (Natural Language Processing - NLP) refers to the analysis, interpretation, and production of natural language clauses. NLP is a growing, interdisciplinary research field, highly interesting from both the theoretical and the practical point of view.
Technical innovation is shifting both speculative and applicative attention towards heterogeneous language forms, such as verbal, vocal, iconic, and gestural.
After decades of theoretical and applied research a vast collection of models exist (symbolic, stochastic, neural networks and deep learning based, …). Such models enable the development of applications in several fields: human-machine interaction;  document analysis, search; multimodal and multilingual linguistic authoring; speech analysis and vocal interaction; sentiment analysis, conversational agents; ... 
Problems, models, and methodologies faced by NLP are also quite interesting for the study of communication, expression, and interaction processes among human beings, conversation and dialogue analyses and augmentation;  sentiment analysis and language rehabilitation  as such topics are common to several disciplines (for example, think to communication, instruction,  cognitive sciences, psychology, or medicine).
History of NLP - full of huge efforts, intense discussions, and many failures - shows the complexity of the topic. Symbolic models, initially prevalent in the area, turned out to be unable to capture the intrinsic complexity of natural language. Today, such models are often augmented by means of stochastic and deep learning models, especially about morphology, lexicon, syntax, semantics, pragmatics, prosody, conversation and dialogue. Moreover, stochastic and deep learning based models are also useful for management, search and retrieval of knowledge, whenever it is expressed by means of linguistic clues.
Current research directions in the NLP field, as in modern linguistics, tend to emphasize relationships between production and interpretation of written and  spoken communication where emotions, sentiments and affect play an important role.

Introduce students to problems and suitable solution methodologies (with their strengths, and weaknesses) related to the analysis and production of natural language clauses, both written and spoken.
Present current role of stochastic and deep learning based models; enlighten new opportunities of combining traditional, formal analysis based models with stochastic and neural models, for morphology, syntax, semantics, pragmatics, voice, prosody, discourse and dialog analysis, sentiment analysis.
Provide hands-on, tutored practice sessions, where students can test models and techniques presented during classes. Applications will include: analysis of language based, human-machine and human-human interaction (for written and spoken dialogues); pattern search and recognition for sentiment analysis in critical interaction; complexity analysis for both texts and generic communication events; definition of user profiles including preferences about expression modalities (in forensic, educative and clinical context). 

Introduction.

Mind models and linguistic / expressive / interactive competencies

Computational linguistics as a representation of human linguistic competencies, as a model, and as a solution to specific and well defined problems.

Roles of symbolic, stochastic and deep learning based models in: morphologic, syntactic, semantic, and pragmatic analysis; sentiment analysis; spoken language, phonologic, and prosodic analysis and generation; linguistic prediction; complexity evaluation; pattern recognition.

Models and techniques for natural language processing.

Morphologic analysis and ambiguity resolution: lexicons, corpora and dictionaries.

Syntactic and structural analysis:

• Symbolic approaches
• Stochastic approaches
• Neural Networks and Deep Learing based approaches
• Hybrid approaches

Semantic and discourse analysis: using integrated approaches; analysis of different representation levels.

Models and techniques for spoken natural language processing.

Components and characteristics of vocal expression and interaction: feature extraction, classification of vocal characteristics, voice profile definition, vocal expression and interaction model.

Models for the description of: tone and prosody, time scheduling, forms, interactions, and complex dialogues, expressivity.

High quality text-to-speech (TTS) and speech recognition (ASR). Analysis strategies and models for emotional and affective components in both TTS and ASR.

Models and tools supporting human-machine and human-human interaction.

Sentiment analysis

Conversational Agents

Linguistic user profiles.

NLP for language rehabilitation

PRACTICES

Hands-on sessions about applications and tools. – 3 sessions (presence will contribute to the final mark)

Useful, but not required, machine learning and formal languages.

Useful, but not required, basic knowledge of the Python language. 

II edition (2008) - III edition to appear in 2022 (?)