Semafor 3.0 from Github vs online demo

[Сергій Дахній]

Hello,

I'm using Semafor 3.0 with Malt parser model and my results differs from online demo.

For example, for sentence "I need large bbq" I've got two frames:

1) Required_event

2) Dimension

In online demo (http://demo.ark.cs.cmu.edu/parse?sentence=I%20need%20large%20bbq)  there is Needing frame instead of Required_event.

My CoNLL input is the same as in online demo.

Could anybody give me an idea about the reason of such difference?

Thanks,

Sergiy Dakhniy 

[jonathanwi...@gmail.com]

I'm getting different results too. The ones from the online demo seem to be much better. I thought it might be the dependency parser (online uses Turbo and Github uses Malt), but if your CoNLL input is the same that must not be it.

[Sam Thomson]

Hi all,

The online demo uses version v3.0-alpha-03. I forget if it was retrained with Turbo parsed training data. I believe it was, which might explain the difference. If you're using v3.0-alpha-03 and the exact same conll input, then the only explanation would be differences in the learned parameters. With these machine-learned models it can be quite hard to explain specific differences...

Cheers,

-Sam

[Sam Thomson]

I rechecked, and yes, it's using a model specifically trained for Turbo parses. The model is available here: http://www.ark.cs.cmu.edu/SEMAFOR/turbo_20130606.tar.gz

Cheers,

-Sam

[Jonathan Mugan]

Thanks! Do we need to change 

bash ${MY_DIR}/runMalt.sh ${INPUT_FILE} ${TEMP_DIR}

in bin/config.sh? There doesn't appear to a runTurbo.sh.

[Jonathan Mugan]

I couldn't figure how to run Semafor with the turbo model over a file, but I did get it running as a server, but I got the same results. Still different from the demo, even with CoNLL copied from the demo.  Any other ideas?

[data academy]

Hello, 

I am beginner in semafor, can you explain how can I parsing the text with semafor. I cant find ant training ....