AlphaNET Semantic Net
Versione italiana di questa pagina
AlphaNET: an Italian Semantic Nework, extracted from a dictionary through automatic NLP procedures
Is it possible to write a NLP system that retourns the syntactic representation of dictionary glosses?
Is it possible to formalize and implement procedures which translate the result of the syntactic analysis into a frame of semantic relations?
AlphaNET is the answer to these questions!
SemNet DEMO
The purpose of AlphaNET, a project started in 2010, is
the definition and implementation of criteria and procedures for full
automatic extraction of semantic information
from a monolingual dictionary (in this case Italian). From about 54'000 words of the source-dictionary, 105'000 meanings have been isolated,
each with its morphological and syntactic features (such as argument structure, subcategorization frame etc.).
The hypertext version, that one may browse starting from this page, counts about 139'000 linked html files.
This semantic net, which becomes progressively an ontology, has been generated from the dictionary glosses,
previously parsed by a hybrid constraint-based and
data-driven dependency parser. Syntactic relations have then been
translated into semantic relations,
by a Syntax-to-Semantic-Interpreter (SynToSem module). No editing or
human supervising has been involved in this process, since the project
aims to formalize computational procedures for this task.
Therefore one will find errors and gaps, due to parsing errors in most
cases:
a new parsing session is forthcoming. Meanwhile I
provide to make the semantic interpretation increasingly rich and
precise, enhancing the SynToSem module.
Each meaning-unit (Semantic Frame) is provided by morphological and syntactic information, which is not shown in this on-line version.
Here an example: mangiare (trans.) mangiare (intrans.).
Read more about: Rule-based Semantic Tagging and Word Sense Disambiguation. An Application Undergoing Dictionary Glosses(paper 2013)
Mind that what You see is only the surface... :-) But here you can see some examples of the inverse relations: bianco or scimmia .
Graph "pianta">"albero">"cacao"
One of AlphaNET applications is supporting a parser in
disambiguation tasks. But it may be used in Information Retrieval,
Knowledge Management, etc. as well.
This version is a prototype resulting from a first step of the AlphaNET project, completed in February 2013 (although continuously being enhanced).
Second step (ongoing) adresses word sense disambiguation within the paraphrases.
The real aim of this project is not the semantic net in itself, but rather the procedures that allows its generation.