Table des matières

Pierpaolo Basile, Franco Cutugno, Malvina Nissim et al.

Preface to the EVALITA 2016 Proceedings

• Chairs
• Steering Committee

Part I: Introduction to EVALITA 2016

Pierpaolo Basile, Franco Cutugno, Malvina Nissim et al.

EVALITA 2016: Overview of the 5th Evaluation Campaign of Natural Language Processing and Speech Tools for Italian

• 1 Introduction
• 2 Tasks and Challenge
• 3 Participation
• 4 Making Cross-task Shared Data
• 5 Towards Future Editions

Rachele Sprugnoli, Viviana Patti et Franco Cutugno

Raising Interest and Collecting Suggestions on the EVALITA Evaluation Campaign

• 1 Introduction
• 2 Questionnaire for Industrial Companies
• 3 Questionnaire for EVALITA Participants and Organizers
• 4 Lessons Learnt and Impact on EVALITA 2016

Walter Daelemans

Keynote: Profiling the Personality of Social Media Users

• Bio

Part II: EVALITA 2016: Task overviews and participants reports

Leonardo Badino

The ArtiPhon Task at Evalita 2016

• 1 Introduction
• 2 Data
• 3 Sub-tasks
• 4 Phone recognition systems
• 5 Results
• 5.1 ISCT systems
• 5.2 Baseline systems – acoustic vs. articulatory results
• 6 Conclusions

Piero Cosi

Phone Recognition Experiments on ArtiPhon with KALDI

• 1 Introduction
• 2 Data
• 3 ASR
• 4 The EVALITA 2016 - ArtiPhon Challenge
• 5 Conclusions

Anne-Lyse Minard, Manuela Speranza et Tommaso Caselli

The EVALITA 2016 Event Factuality Annotation Task (FactA)

• 1 Introduction and Motivation
• 2 Task Description
• 2.1 Factuality Attributes
• 2.2 Factuality Value
• 3 Dataset Description
• 4 Evaluation
• 5 System Results
• 5.1 FactPro module
• 5.2 Results
• 5.3 Error Analysis of FactPro
• 6 Conclusion and Future Work

Pierpaolo Basile, Annalina Caputo, Anna Lisa Gentile et al.

Overview of the EVALITA 2016 Named Entity rEcognition and Linking in Italian Tweets (NEEL-IT) Task

• 1 Introduction
• 2 Task Description
• 3 Corpus Description and Annotation Process
• 4 Evaluation Metrics
• 5 Systems Description
• 5.1 UniPI
• 5.2 MicroNeel
• 5.3 FBK-HLT-NLP
• 5.4 Sisinflab
• 5.5 UNIMIB
• 6 Results
• 7 Conclusions

Giuseppe Attardi, Daniele Sartiano, Maria Simi et al.

Using Embeddings for Both Entity Recognition and Linking in Tweets

• 1 Introduction
• 2 Task Description
• 3 Building a larger resource
• 4 Description of the system
• 4.1 Word Embeddings
• 4.2 Bi-LSTM Character-level NER
• 5 Experiments
• 6 Results
• 7 Discussion
• 8 Related Work
• 9 Conclusions

Flavio Massimiliano Cecchini, Elisabetta Fersini, Pikakshi Manchanda et al.

UNIMIB@NEEL-IT : Named Entity Recognition and Linking of Italian Tweets

• Introduction
• 1 Systems Description
• 2.1 Named Entity Recognition
• 2.2 Named Entity Linking
• 2.2.1 Learning2Link
• 2.2.2 Neural-Network Language Model (NNLM) Linking
• 2.3 NIL Clustering
• 3 Results and Discussion
• 3.1 Named Entity Recognition
• 3.2 Named Entity Linking
• 3.3 NIL Clustering
• 3.4 Overall
• 4 Conclusion

Francesco Corcoglioniti, Alessio Palmero Aprosio, Yaroslav Nechaev et al.

MicroNeel: Combining NLP Tools to Perform Named Entity Detection and Linking on Microposts

• 1 Introduction
• 2 Tools and Resources
• 2.1 The Wiki Machine
• 2.2 Tint
• 2.3 Social Media Toolkit
• 3 Description of the System
• 3.1 Preprocessing
• 3.2 Annotation
• 3.3 Merging
• 4 Results
• 5 Discussion
• 6 The tool
• 7 Conclusion and Future Work

Vittoria Cozza, Wanda La Bruna et Tommaso Di Noia

sisinflab: an ensemble of supervised and unsupervised strategies for the NEEL-IT challenge at Evalita 2016

• 1 Introduction
• 2 Description of the system
• 2.1 Spotlight-based solution
• 2.2 Machine learning based solution
• 2.3 Linking
• 3 Results and Discussion

Anne-Lyse Minard, Mohammed R. H. Qwaider et Bernardo Magnini

FBK-NLP at NEEL-IT: Active Learning for Domain Adaptation

• 1 Introduction
• 2 Domain Adaptation for NER
• 2.1 The TextPro-AL platform
• 2.2 Available Data
• 2.3 Manual Annotation of Training Data with TextPro-AL
• 3 Description of the system
• 3.1 Entity Recognition and Classification
• 3.2 Entity Linking
• 3.3 Clustering
• 4 Results
• 5 Discussion

Cristina Bosco, Fabio Tamburini, Andrea Bolioli et al.

Overview of the EVALITA 2016

Part Of Speech on TWitter for ITAlian Task

• 1 Introduction and motivation
• 2 Data Description
• 2.1 Tokenisation Issues
• 2.2 Tagset
• 2.2.1 Proper Noun Management
• 2.2.2 Foreign words
• 3 Evaluation Metrics
• 4 Teams and Results
• 5 Discussion and Conclusions

Giuseppe Attardi et Maria Simi

Character Embeddings PoS Tagger vs HMM Tagger for Tweets

• 1 Introduction
• 2 Building a larger training resource
• 3 Normalization of URLs, emoticons and emoji’s
• 4 The traditional approach: the TANL tagger
• 5 Character-level Embeddings
• 6 Results
• 7 Discussion
• 7.1 Error analysis
• 8 Conclusions

Andrea Cimino et Felice Dell’Orletta

Building the state-of-the-art in POS tagging of Italian Tweets

• 1 Description of the system
• 1.1 Lexicons
• 1.1.1 Word Embedding lexicons
• 1.1.2 Morpho-syntactic lexicon
• 1.2 The POS tagger architecture
• 1.2.1 Word-based bi-LSTM
• 1.2.2 Bag-of-Character Word-based bi-LSTM
• 2 Results and Discussion
• 3 Conclusion

Tobias Horsmann et Torsten Zesch

Building a Social Media Adapted PoS Tagger Using FlexTag – A Case Study on Italian Tweets

• 1 Introduction
• 2 Experimental Setup
• Results
• 4 Summary

Giulio Paci

Mivoq Evalita 2016 PosTwITA tagger

• 1 Introduction
• 2 Description of the system
• 2.1 Hunpos
• 2.2 Yamcha
• 2.2.1 Baseline
• 2.2.2 Twitter specific elements
• 2.2.3 Normalized string
• 2.2.4 Dictionary tags
• 2.3 CRFSuite
• 2.4 Tagger combination
• 3 Results
• 4 Discussion

Partha Pakray et Goutam Majumder

NLP–NITMZ:Part–of–Speech Tagging on Italian Social Media Text using Hidden Markov Model

• 1 Introduction
• 2 Proposed Method
• 2,1 Conditional Model Approach
• 2.2 Generative Model
• 3 Experiment Results
• 3.1 Dataset
• 3.2 Results
• 3.3 Comparison with other submissions
• 4 Conclusion

Barbara Plank et Malvina Nissim

When silver glitters more than gold: Bootstrapping an Italian part-of-speech tagger for Twitter

• 1 Introduction
• 2 Data selection and bootstrapping
• 3 Experiments and Results
• 3.1 Experimental Setup
• 3.2 Model
• 3.3 Results on test data
• 4 What didn’t work
• 5 Conclusions

Egon W. Stemle

bot.zen @ EVALITA 2016 A minimally-deep learning PoS-tagger (trained for Italian Tweets)

• 1 Introduction
• 2 Design
• 2.1 Word Embeddings
• 2.2 Character-Level Sub-Word Information
• 2.3 Recurrent Neural Network Layer
• 3 Implementation
• 3.1 Word Embeddings
• 3.2 Character-Level Sub-Word Information
• 3.3 Recurrent Neural Network Layer
• 4 Results
• 4.1 Training Data for w2v and PoS Tagging
• 4.1.1 DiDi-IT (PoS, w2v)
• 4.1.2 Italian UD (PoS, w2v)
• 4.1.3 PoSTWITA (PoS and w2v)
• 4.1.4 C4Corpus (w2v)
• 4.1.5 PAISÀ (w2v)
• 4.2 PoSTWITA shared task
• 5 Conclusion & Outlook

Fabio Tamburini

A BiLSTM-CRF PoS-tagger for Italian tweets using morphological information

• 1 Introduction
• 2 Input features
• 2.1 Word Embeddings
• 2.2 Morphological features, Unknown words handling and Sentence padding
• 3 (Deep) Learning Blocks
• 4 Experiments
• 4.1 Hyper-Parameters
• 4.2 The Early Stopping procedure
• 4.3 Results

Annalina Caputo, Marco De Gemmis, Pasquale Lops et al.

Overview of the EVALITA 2016 Question Answering for Frequently Asked Questions (QA4FAQ) Task

• 1 Motivation
• 2 Task Description: Dataset, Evaluation Protocol and Measures
• 2.1 Data Format
• 2.2 Baseline
• 2.3 Evaluation
• 3 Systems
• 4 Results
• 5 Conclusions

Divyanshu Bhardwaj, Partha Pakray, Jereemi Bentham et al.

Question Answering System for Frequently Asked Questions

• 1 Introduction
• 2 Corpus and Preprocessing
• 2.1 Parsing
• 2.2 Stopword Removal
• 3 System Architecture
• 4 Experiments
• 5 Performance
• 6 Discussion
• 7 Conclusion and Future Direction

Erick R. Fonseca, Simone Magnolini, Anna Feltracco et al.

Tweaking Word Embeddings for FAQ Ranking

• 1 Introduction
• 2 System Description
• 3 Evaluating our system
• 3.1 Development Set
• 3.2 Baseline
• 4 Results
• 5 Error Analysis
• Conclusion

Arianna Pipitone, Giuseppe Tirone et Roberto Pirrone

ChiLab4It System in the QA4FAQ Competition

• 1 Introduction
• 2 The QuASIt System
• 2.1 Searching in the support text
• 3 ChiLab4It
• 3.1 The architecture
• 3.2 A toy example
• 4 Evaluations
• 5 Discussion and Future Works

Francesco Barbieri, Valerio Basile, Danilo Croce et al.

Overview of the Evalita 2016 SENTIment POLarity Classification Task

• 1 Introduction
• 2 Task description
• 3 Development and Test Data
• 3.1 Corpora Description
• 3.2 Annotation Scheme
• 3.3 Annotation procedure
• 3.4 Format and Distribution
• 4 Evaluation
• 5 Participants and Results
• 5.1 Task1: subjectivity classification
• 5.2 Task2: polarity classification
• 5.3 Task3: trony detection
• 6 Discussion
• 7 Closing Remarks

Giuseppe Attardi, Daniele Sartiano, Chiara Alzetta et al.

Convolutional Neural Networks for Sentiment Analysis on Italian Tweets

• 1 Introduction
• 2 Description of the Systm
• 2.1 Data Preprocessing
• 2.2 Word Embeddings and Sentiment Specific Word Embeddings
• 2.3 Distant supervision
• 3 Experiments
• 4 Results
• 5 Discussion

Davide Buscaldi et Delia Irazù Hernandez-Farias

IRADABE2: Lexicon Merging and Positional Features for Sentiment Analysis in Italian

• 1 Introduction
• 2 Building a unified dictionary
• 3 Positional Features
• 3.1 Other features
• 4 Results and Discussion
• 5 Conclusions

Giuseppe Castellucci, Danilo Croce et Roberto Basili

Context-aware Convolutional Neural Networks for Twitter Sentiment Analysis in Italian

• 1 Introduction
• 2 A Sentiment and Context aware Convolutional Neural Networks
• 2.1 Injecting Sentiment Information through Polarity Lexicons
• 2.2 Context-aware model for SA in Twitter
• 3 The Unitor Classification Workflow
• 4 Experimental Results
• 5 Conclusions

Andrea Cimino et Felice Dell’Orletta

Tandem LSTM-SVM Approach for Sentiment Analysis

• 1 Description of the system
• 1.1 Lexical resources
• 1.1.1 Sentiment Polarity Lexicons
• Existing Sentiment Polarity Lexicons
• Automatically translated Sentiment Polarity Lexicons
• Automatically created Sentiment Polarity Lexicons
• 1.1.2 Word Embedding Lexicons
• 1.2 The LSTM-SVM tandem system
• 1.2.1 The LSTM network
• 1.2.2 The SVM classifier
• Raw and Lexical Text Features
• Morpho-syntactic Features
• Lexicon features
• 2 Results and Discussion
• 3 Conclusion

Vito Vincenzo Covella, Berardina De Carolis, Stefano Ferilli et al.

Lacam&Int@UNIBA at the EVALITA 2016-SENTIPOLC Task

• 1 Introduction
• 2 Systems Description
• 2.1 SentimentWS
• 2.2 SentiPy
• 3 Experiments
• 3.1 SentimentWS Results
• 3.2 SentiPy Results
• 4 Conclusions

Jan Deriu et Mark Cieliebak

Sentiment Analysis using Convolutional Neural Networks with Multi-Task Training and Distant Supervision on Italian Tweets

• 1 Introduction
• 2 Convolutional Neural Networks
• 3 Training
• 3.1 Three-Phase Training
• 3.2 Data
• 3.3 Experiments & Results
• 4 Conclusion

Emanuele Di Rosa et Alberto Durante

Tweet2Check evaluation at Evalita Sentipolc 2016

• 1 Introduction
• 2 Tweet2Check description
• 3 Experimental Analysis
• 4 Conclusion

Simona Frenda

Computational rule-based model for Irony Detection in Italian Tweets

• 1 Introduction
• 2 Related Work
• 3 Methodology
• 4 Analysis of corpus
• 4.1 Resources
• 4.2 Data Processing
• 5 Features
• 5.1 Positive Interjections
• 5.2 Expressions with “che”
• 5.3 Pronoun “tu” and Verb Morphology
• 5.4 Disjunctive Conjunction
• 5.5 Onomatopoeic Expressions for laughter
• 5.6 Ironic Emoticons
• 5.7 Hashtag
• 5.8 Regional Expressions
• 5.9 Quotation Marks
• 5.10 Heavy Punctuation
• 6 Results
• 7 Conclusion

Daniela Moctezuma, Eric S. Tellez, Mario Graff et al.

On the performance of B4MSA on SENTIPOLC’16

• 1 Introduction
• 2 Our participation
• 2.1 Text modeling with B4MSA
• 2.2 Set of Features
• 2.3 Model Selection
• 2.4 B4MSA + EvoDAG
• 3 Results and Discussion
• 4 Conclusions

Lucia C. Passaro, Alessandro Bondielli et Alessandro Lenci

Exploiting Emotive Features for the Sentiment Polarity Classification of tweets

• 1 Introduction
• 2 Description of the system
• 2.1 Preprocessing
• 2.2 Feature extraction
• 2.2.1 Lexical Features
• 2.2.2 Negation
• 2.2.3 Morphological features
• 2.2.4 Shallow features
• 2.2.5 Twitter features
• 3 Introducing emotive and LDA features
• 4 Classification
• 5 Results
• 6 Conclusions

Irene Russo et Monica Monachini

Samskara

Minimal structural features for detecting subjectivity and polarity in Italian tweets

• 1 Introduction
• 2 System description
• 2.1 Data analysis, data preprocessing and feature selection
• 2.2 Results and Discussion