annotation/features from preconditions   

Hello,

is there a possibility assigning precondition elements to the features or annotate them? For example having a rule for a company name only firing when a location is detected, where the second part is implemented via the precondition. Then one would like to include the location into the features. As far as I understood a rule implies a certain order of the words and a preconditions scopes on different levels (sentence, paragraph,..). I was thinking of examples as:

"Jack and Brothers LTD opened a new cite in California."

"...California... new site was opened by Jack and Brothers LTD."

It would also be nice to be able to specify the scope more detailed. For example: this sentence and the last two or n sentences before and after.

Regards,
Toby

Hi Kevin,

thanks! This totally suffices my requirement :)

Regards,
Toby

Hi Toby, yes these are "loose match" rules. They ignore the majority of Annotations in the CAS, and so they are the opposite of "exact match" rules, where the exact pattern has to exist. I think I covered this topic in more detail in an earlier post, and there should also be some information in the help.

Hi. We are planning to allow scoping of rules under annotation spans other than Sentence, Paragraph or Document in the future. However we have not committed to a specific timeframe for supporting this. Hopefully this would solve your issue, though I'm not 100% sure that it would.

You say that you need the Location to be in a range +/- 1 sentence from the Date. In this case, you probably need slightly different functionality. You can do this test in post-processing right now, while in the future it is again likely that we'll support it when writing rules. Here is what you could do

• When creating Date and Location annotations, link them to the Sentence they are in. You can do this using a Rule - when you create a new Annotation it has a "Features" node below it. Right-clicking on this lets you link to the Sentence, Paragraph or Document covering the matched annotations. Now your Date and Location annotations will have a "sentence" feature, which links to the Sentence Annotation covering them.
• Each Sentence has a "sentenceNumber" feature. So what you need to do is to compare the value of this feature for Dates and Locations
• You can do this in post-processing fairly easily. For customer engagements we provide a number of CAS Consumers that can export to Databases and Triple stores. Once the data is in a triple store or a database table it becomes extremely easy to query for data based on a parameter such as sentence number - a simple SQL query would do the job.
• Alternatively you could write your own custom annotator to do this very easily. The UIMA documentation has plenty of examples which would show how to do this. The next release of the LanguageWare Resource Workbench will let you add in any custom annotator to a pipeline, so you could easily build on top of your range-based annotator.

The ideal solution would be to let you do all this in rules. We should have support for this in the future. We are currently working on support for testing any feature for annotations. We also have plans to let you test ranges on integer features, and combining this would do what you need.

I hope this helps,

Kevin

Hi Matej, can you email us at EMEALAN@ie.ibm.com please? That is the best way to talk about the issues you raised.

Thanks,

Kevin