[Corpora-List] looking for FRED tools/source code

Valerio Basile valerio.basile at inria.fr
Sun May 22 19:43:27 CEST 2016


Hi Parisa,


> I am master student of NLP. In part of my thesis, i used semantic role
labelling by mate plus (https://github.com/microth/mateplus). Now i need to employ the FRED tool from STLab(http://stlab.istc.cnr.it/stlab/FRED) for testing the outputs.

to my knowledge, FRED internally runs Boxer for semantic parsing, and this includes role labeling. If that is the only functionality you need you could try to use just the C&C tools + Boxer: http://svn.ask.it.usyd.edu.au/trac/candc (right now the link seems down to me). I also maintain an installation of an online API to access the pipeline, including Boxer, via HTTP. It's experimental, please notify me if you plan to use it extensively. The instructions are here: http://valeriobasile.github.io/candcapi/

Note that the semantic role labeling made by Boxer is at the discourse referent level. Aligning it with word dependencies may not be always straightforward.


> Could anybody help me to access the FRED tool?
> another question, does STLab site allow accessing the FRED's source code?

As far as I know the FRED source code has never been released to the public. I wanted to use FRED to parse a collection of documents, but it has been offline at least since last March (that's when I started to check, anyway). So I wrote my own tool that has some of the same functionalities, that is, it runs the C&C pipeline and Boxer on a text, then disambiguates the word senses and the named entities, and it aligns it all together to produce RDF triples. It can run locally or use online services for parsing, word sense disambiguation and entity linking.

Parsing the sentence "A boy is driving his car", for instance, you get triples like:

<http://wordnet-rdf.princeton.edu/wn31/02961779-n> < http://ns.inria.fr/aloof/boxer/relation#Theme> < http://wordnet-rdf.princeton.edu/wn31/01934845-v> . <http://wordnet-rdf.princeton.edu/wn31/09647338-n> < http://ns.inria.fr/aloof/boxer/relation#Agent> < http://wordnet-rdf.princeton.edu/wn31/01934845-v> . <http://wordnet-rdf.princeton.edu/wn31/10305010-n> < http://ns.inria.fr/aloof/boxer/relation#Agent> < http://wordnet-rdf.princeton.edu/wn31/01934845-v> .

The role inventory is VerbNet, as implemented in Boxer, but I'm implementing the mapping with FrameNet from SemLinks.

As I'm actively working on it, it's unpolished and the output changes often. However, I'm happy to provide help to install and run it. and of course, it's open source, so contributions are very welcome: https://github.com/valeriobasile/learningbyreading

Best, Valerio -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: text/html Size: 3929 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20160522/cfed529a/attachment.txt>



More information about the Corpora mailing list