Hi,
At the upcoming ParGram Meeting in Warsaw, we would like to experiment with a new way to do our usual structure comparison: We would like to use INESS (http://iness.uib.no/) and ParGramBank directly.
There are a couple of advantages to this. One obvious advantage is that ParGramBank will grow. But we can also easily switch between languages/sentences, and there is no need to create a humongous PDF file with structures in it (last year, our structure handout had 234 pages).
Paul Meurer (www.uib.no/persons/Paul.Meurer) has kindly implemented a couple of new features in INESS. I attach an email from Paul below. Please follow his instructions when adding to ParGramBank.
To be able to upload structures to INESS (in Prolog format), you need to have an account there. Please follow the instructions on the INESS homepage to create your account. Once the account is created, Paul can give you the necessary user rights to upload files.
Lastly, I attach the sentences in a text file to this email. As Paul says in his email, please use a consistent naming scheme for the structures; you should use something like urd-051-fs (i.e., ISO 639-3 language code - sentence number - fs). Start the sentence number at 51 (there are already 50 sentences in ParGramBank).
If you have more questions, please don't hesitate to send email to Paul, or myself. Also, Agnieszka/Paul and others, if I forgot anything, feel free to jump in.
Best, Jani
-------- Weitergeleitete Nachricht -------- Betreff: Re: ParGram sentences Datum: Thu, 15 Jan 2015 11:48:57 +0100 Von: Paul Meurer paul.meurer@uni.no An: Agnieszka Patejuk agnieszka.patejuk@googlemail.com Kopie (CC): Sebastian Sulger sebastian.sulger@uni-konstanz.de, Adam Przepiórkowski adamp@ipipan.waw.pl, Miriam Butt Miriam.Butt@uni-konstanz.de, Victoria Rosén victoria@uib.no
Hi,
BTW: I went through previous correspondence related to the meeting and have gathered the following points to cover:
PARGRAM MEETING (3 days) • structure comparison: – how: traditional or using INESS?
I think now everything needed to make INESS suitable for structure comparison is in place. Here is what is new:
* I have implemented uploading of prolog files (one by one, or as a gzipped archive) * Once the sentences (not structures) are aligned, it is easy to switch between treebanks/languages. It is enough for the sentences to be aligned to a pivot language (e.g., Urdu). You can test this in the ParGram treebanks.
What people should do:
* Parse their sentences in XLE * Use a consistent naming scheme for the sentences (e.g., deu-050-fs), where the number corresponds to the running sentence number. We should not start with 1, but continue where we stopped last time. I think new sentences should start at 50. Alternative translations could be called deu-050a-fs etc. * Upload the sentences using the _upload files_ link on the Treebank overview page (either one-by-one, or as a gzipped archive, no other archiving program will work). * Disambiguate the sentences in INESS. * Add glosses, as described in the documentation. * Align the sentences, with English as a minimum, using the Alignment tool. This, Jani or I could do for them.
Alternatively, for languages whose grammar is in INESS, the sentences could be parsed in INESS directly.
Does this sound feasible?
I think the sentences should be ready quite soon, for people to be able to do all this before Pargram.
— Best wishes, Paul