Hi,

your new defintion "Left wide, right wide" is fine for me. +1

(Symmetry is really important in my opinion, much more important than
the actual specification of predicates for zero-length annotations.)


Best,


Peter


Am 24.10.2020 um 23:05 schrieb Richard Eckart de Castilho:
> Hi,
>
>> On 24. Oct 2020, at 15:12, Peter Klügl <[email protected]> wrote:
>>
>> Yes, there needs to be some additional changes then. In my opinion, it's
>> ok if ADJACENT doe not imply OVERLAPPING. Just by listening to it, it
>> sounds also somwhat reasonable. Afterall, it's just a name for a
>> predicate configuration.
>>
>> For zero-length annotations, OVERLAPPING would still apply due to the
>> inner minimal span that is not ADJACENT.
> Thinking more about it, I am starting to believe that we should not try
> to approach the nature of zero-width spans by observing how a infinitely
> small span would behave but rather be considering their nature of not
> having any width at all.
>
> So what does that mean?
>
> COVERED-BY: we need to decide if a zero-width span at the start/end 
> of another span is covered by the other span or not. For a symmetric
> approach, the same decision should be made irrespective of whether the
> zero-width span is located at the start or end of the other span.
> Let's say we want to query for all annotations on a sentence,
> we'd almost certainly want to include any zero-width annotations at
> the start and end of the sentence - so: let's say a zero-width
> annotation is at the start/end of another annotation is ALWAYS COVERED BY
> this other annotation.
>
> LEFT/RIGHT ADJACENT: in order for a span X to be adjacent to a span Y,
> X should start/end at the same location as Y. However, it seems quite
> contradictory that a span can be simultaneously COVERED-BY and ADJACENT.
> So as a consequence of the choice on COVERED-BY above, a zero-width span
> at the start/end of another span should NOT be considered adjacent. 
> We would say that X is right/left adjacent to Y if it starts/ends at
> the same the size of the intersection between X and Y is zero.
> That implies that zero-width annotations are NEVER ADJACENT.
>
> LEFT/RIGHT OVERLAPPING: if an X annotation is left or right overlapping
> another annotation Y, it implies that some part of X is outside the 
> boundaries of Y. That, however, would contradict the choice we made for
> covered-by above for zero-width annotations. That implies that a
> zero-width annotation at the start/end of another annotation is
> NEVER LEFT/RIGHT OVERLAPPING that other annotation.
>
> LEFT-OF/RIGHT OF: if a zero-width annotation X at the start/end of 
> another annotation Y is not adjacent because it it covered, than
> that implies that is is also not left-of/right-of.
>
> IMHO these interpretations above are symmetric and also avoid the
> odd case of two colocated zero-width spans being simultaneously
> left(overlapping) and right(overlapping) etc. of themselves.
>
> I have added a matrix using the rationales given above in the
> Google Sheets document on sheet "Relation types in detail (WN, WWv2)"
> in the "Left wide, right wide" column group.
>
> https://docs.google.com/spreadsheets/d/1fgMbqVlxwJSBNui7Y_phtRYEhzr_rfXQ5ZUREH1-nwI/edit?usp=sharing
>
> Cheers,
>
> -- Richard

-- 
Dr. Peter Klügl
Head of Text Mining/Machine Learning

Averbis GmbH
Salzstr. 15
79098 Freiburg
Germany

Fon: +49 761 708 394 0
Fax: +49 761 708 394 10
Email: [email protected]
Web: https://averbis.com

Headquarters: Freiburg im Breisgau
Register Court: Amtsgericht Freiburg im Breisgau, HRB 701080
Managing Directors: Dr. med. Philipp Daumke, Dr. Kornél Markó

Reply via email to