Hierarchical LMs Based on Classes of Phrases

Raquel Justo

Natural Language and Dialogue System lab, UC Santa Cruz

Tuesday, August 13, 2013
12:30 p.m., Conference Room 5A

Abstract:

Hierarchical LMs based on classes of phrases are presented as an alternative to classical word- or phrase-based LMs when considering applications that suffer from data sparseness. The definition of such kind of complex models, including different knowledge sources, was carried out within the Finite State Machine framework. The use of this framework assists the composition of different models in an ASR system. Different experiments were performed showing that these models could improve the results provided by baseline LMs. Additionally, some preliminary experiments were carried out using these models in speech translation systems and some promising results were also obtained.

On the other hand, we propose to use this kind of models when facing a very different problem like the designing of classifiers for emotion detection on online dialogues. A bootstrapping method that considers different classifiers based on NLP techniques was used as a baseline to detect sarcastic/not sarcastic and nice/nasty posts picked up from Internet forums. The idea is to explore whether the use of Hierarchical LMs to learn the structure of sarcastic sentences can improve the results obtained when considering handcrafted patterns.

Bio:

Raquel Justo received her Electronics Engineering Degree from the University of the Basque Country  (UPV/EHU) in 2001, and her Bachelor's degree in Physics from the University of Cantabria in 2004. She obtained her Master's degree and Ph.D. in Language and Information Systems, also from UPV/EHU, in 2005 and 2009 respectively. When finishing her graduate studies she spent two years at IK4-Ikerlan research centre working in the Advanced Sensors Department. Since 2003, she is with the Pattern Recognition and Speech Technology (PR&ST) research group at UPV/EHU, where she has been working in automatic speech recognition and understanding, dialog management and information retrieval. She has been a lecturer at University of the Basque Country since 2007. She carried out a research stay at the Technical University of Valencia in 2006 and she is currently a visitor researcher in the Natural Language and Dialogue System (NLDS) lab at University of California, Santa Cruz. She has participated in several national and local research projects and currently leads one of them related to dialog systems. She has also been involved in different research contracts granted by private companies and has led some of them.