search

SIMPATICOProject / SimpaticoTAEServer

This is the official repository of the Text Adaptation Engine server of Simpatico.
GNU Lesser General Public License v3.0
7 stars 3 forks source link

Syntactic simplification test output doubtful #7

Open islam-nassar opened 6 years ago

islam-nassar commented 6 years ago

Hey guys, I am working on a text simplification topic and stumbled upon your awesome repo. I followed your conclusive documentation and finally after about 5 hours of troubleshooting my own system, i managed to get both the lexical and syntactical servers running.

Now I am trying to use the test sentences to test the syntactic simplification and I get weird results: Some of the sentences work perfectly fine such as: Sending... If you later have a problem and need to use your contingency money, you can contact our Customer Accounts Team to explain why you need this money. Receiving... Suppose you later have a problem and need to use your contingency money . Then you can contact our customer accounts team to explain why you need this money .

But others (most of the sentences actually) remain unchanged, some of them are relatively straight forward to simplify (such as example below) so I am suspecting that I might be doing something wrong.

Sending... Though he tried hard, nothing changed. Receiving... Though he tried hard, nothing changed.

Sending... Because he's sick, he can't come. Receiving... Because he's sick, he can't come.

Could you please help shed some light on what is the problem, if any.

I have attached the complete list of output that I get when I run the syntactic simplification for your reference.

Much obliged and thanks again for the awesome tool.

Cheers, INassar test_simplification_out.txt

carolscarton commented 6 years ago

Hi,

Are you using any of the extra modules (complex sentence classifier or confidence model)? If so, either the original sentences may not be selected for simplification (considered as simple by the sentence classifier) or the simplified sentences returned by the system may be considered as inadequate by the confidence model.

If you are not using these modules, one reason may be that the parser may not be accurate for these sentences. If the dependency parser tags a word with the wrong tag, the system will not be able to make the simplification.

Another possible reason is that we made our SS system very conservative, to work as accurate as possible on our real-world case. Therefore, some rules may be very restrict. You can edit this in the code and make them less restrictive.

Hope this helps.

Cheers, Carol