It seems possible to detect this kind of error in most cases. The XmiCas only contains feature structures that are indexed or are referenced by other feature structures. For the problem described in this thread, after deserialization there are ByteArray FS that are not indexed and are left unreferenced.
A problem is with out of typesystem data. Imagine that a primitive Array type is referenced by an OOTS feature structure. The array FS will be deserialized, but there is no way to know if it is referenced by something OOTS. Eddie On Mon, Sep 20, 2010 at 6:00 PM, Marshall Schor <[email protected]> wrote: > > > On 9/20/2010 5:13 PM, Eddie Epstein wrote: >> On Mon, Sep 20, 2010 at 9:50 AM, Eddie Epstein <[email protected]> wrote: >>> Hi Diman, >>> >>> Indeed the problem is Xmi deserialization. It looks like ByteArray >>> features must be defined with >>> <multipleReferencesAllowed>true</multipleReferencesAllowed> >>> >>> At first glance this is missing from our documentation. Will double check >>> that. >>> >> In fact UIMA is working as designed. > > Do you think the design could be improved with some kind of error message that > could help the user figure out what to change :-) ? (Perhaps that's not > possible in this case?) > > -Marshall >> It appears that your XmiCas data >> was serialized out with multipleReferencesAllowed=true for the feature >> hypernyms, but then deserialized with a different typesystem where >> multipleReferencesAllowed=false. >> >> Eddie >> >> >
