Research
Career
I am a post-doctoral researcher at Center for Language and Speech Technologies and Applications (TALP), an inter-departmental research lab of Universitat Politècnica de Catalunya (UPC). I'm mostly working on statistical machine translation, with a special focus on confidence estimation, robust multi-lingual linguistic processing and user feedback incorporation.
Until May 2006 I was a PhD student at the Department of Information Engineering and Computer Science (DISI) at the University of Trento, Italy, funded by a grant from Fondazione Bruno Kessler (FBK). My PhD advisors were Marcello Federico and Alessandro Moschitti.
I got both my Bachelor and Master degrees from the Faculty of Computer Engineering of the University of Rome "Tor Vergata", advised by Roberto Basili and Alessandro Moschitti.
Research interests
As an engineer I'm mostly interested in software models and architectures involving statistical and machine learning methods. So far, for the most part I have been dealing with natural language processing tasks and applications, with a special attention to shallow semantic parsing, relation extraction and automatic translation. My main areas of expertise are support vector machines and kernel methods for NLP. My PhD thesis was centered around a linearization/feature selection framework for high-dimensional kernel function, with a focus on tree kernels.
Publications
Click here to get to my list of publications.
Software and data
FLinK
FLinK is a Framework for the Linearization of Kernel functions. More precisely, it allows to reverse-engineer models learned using tree kernel functions, and to isolate the most relevant structured features identified in the huge fragment space. It allows to perform very aggressive feature selection directly in the kernel space, and to transform a tree-kernel learning problem into a very efficient linear-space learning problem. Click here for more details.
ExRel
ExRel is a generic framework for specifying relation extraction architecture, and a set of modules that implement a 2-Stage Semantic Role Labeling system. You can find more information here.
Mixed Features for Semantic Role Labeling
Datafiles with structured (AST1m) and linear features for semantic role labeling (based on the data provided for the CoNLL 2005 shared task on semantic role labeling) is available here. (Oct 2009)
Relevant Tree Fragments for Question Classification
The relevant fragments obtained by reverse-engineering tree kernel models for the Question Classification task are available here. Linguists and experts are invited to study and employ them to discover new and relevant syntactic features for this task. (June 2009)
MapNet
Version 1.1 of the FrameNet to WordNet mapping MapNet is available here. (May 2009)
Structured Features for Semantic Role Labeling
Datafiles with structured features for semantic role labeling (based on the data provided for the CoNLL 2005 shared task on semantic role labeling) is available here. (May 2009)
Datafiles with structured features for semantic role labeling (based on the data provided for the CoNLL 2005 shared task on semantic role labeling) is available here. (May 2009)