[developers] Lexical rules changing predicate symbols

Emily M. Bender ebender at uw.edu
Thu Sep 22 00:18:46 CEST 2016


Thanks, Ann.  As for Lushootseed, I think we'd need some more data. Joshua,
does this discussion inspire any questions/ideas?

Emily


On Sat, Sep 17, 2016 at 3:49 AM, Ann Copestake <aac10 at cl.cam.ac.uk> wrote:

> Yes, I would be happy with that, except that I would suggest making use of
> meaningful sense labels, at least as a mnemonic.  So the pattern on the
> rule would be
>
> _P_v_dance (e,x) <- LR ->  _P_n_dance (x)
> The one could apply the rule productively, by specifying the appropriate
> verbs as having predicates of the form _v_dance, if one wanted.  (The
> alternative is to just use the rule as a description and explicitly state
> what it applies to, but since one could control the application via an
> inventory of sense labels if the predicate is decomposed and the senses
> correspond to types, I think this is effectively equivalent.)
>
> What I would ideally want to say about  the _dance_ component of the
> predicate is that one should take it as commiting to the lexeme implying
> dance when making inferences about the real world.  I'm deliberately
> stating this informally because what this actually means is up to the
> grammar writer, but I would take it as allowing that inference to be made
> by a computational system (using a robust notion of inference!).  This
> certainly does not commit one to saying that all lexemes denoting dances
> will have this sense.
>
> For a language like Lushootseed this might not be enough to capture the
> way that nouns and verbs behave, but at least it is a start.
>
> For what it's worth, I think of _tango_v_dance as corresponding to a
> semantic space which is close to _tango_n_dance on most dimensions, but
> shifted corresponding to the verb/noun shift.  (There's been a little
> experimental exploration of this sort of thing in distributional semantics
> - there could be much more - but I'm talking about some sort of idealised
> space now.  What Aurelie and I have called `ideal distributions' would work
> here.)  In this idealized space, these are subspaces of the dance n and v
> spaces.  And, for that matter _tango_*_dance (where * generalises over n/v)
> has _tango_n_dance and _tango_v_dance as subspaces, and so on.
>
> Going back to what I was saying before, if one thinks there's a meaning
> postulate that expresses the relationship between the verbal and nominal
> form - e.g., something of the form:
>
> forall x,e [ _tango_v_dance (e,x) <=> exists y [ _dance(y) &
> PERFORM(e,x,y) ]]]
>
> or whatever, where I'm using PERFORM as a placeholder, then the
> alternative would be to express that in the lexical rule and not have a
> separate predicate for the nominal sense.  But then one would have to work
> out a story about what the precise nature of the relationship PERFORM (or
> whatever) is, and whether this can be expressed in terms of properly
> motivated semantic predicates like Dowty's DO, CAUSE and so on.  Saying
> that there are related semantic spaces is actually compatible with this but
> much more general - obviously it's very vague, unless one starts to
> additionally express constraints on the relationship between the semantic
> spaces. But my own feeling is that doing these semantic relationships by
> hand/symbolically is a dead-end (because there's lots of different ones,
> and they have nuances and exceptions and so on).
> The alternative approach of using systematic naming conventions for the
> predicates allows us to make the connection explicitly visible to some
> distributional component, without commiting to the actual lexical semantics.
>
> All best,
>
> Ann
>
>
> On 17/09/2016 05:29, Emily M. Bender wrote:
>
> Thank you for this follow up, Ann. I think I understand your view of
> things.  Just one piece
> I wanted to follow up on this evening:
>
>>
>> On the other hand, as you said, there are other processes where the
>> pattern is regular but with some exceptions, and where the precise semantic
>> effect is difficult to pin down.  Much of English derivational morphology
>> falls into this class - it is useful to represent the relationship with the
>> stem somehow, without claiming that we're capturing everything there is to
>> say.  e.g., "unkind" definitely has some relationship with (a sense of)
>> "kind", and it's useful to know about that, but there are some
>> idiosyncratic aspects of its meaning.   Similarly, I'd say it's useful to
>> represent the relationship between nouns denoting dances and the
>> corresponding verbs (tango etc), which is productive, but I'm quite content
>> to do that via a predicate changing operation.
>>
>>
> As I currently understand things, the change in the predicate symbol from
> _tango_n_1
> to _tango_v_1 would correlate with:
>
> (1) Change of type of ARG0 from x to e
> (2) Addition of ARG1 (of type x)
>
> ... or in the other direction:
>
> (1) Change of type of ARG0 from e to x
> (2) Removal of ARG1
>
> In other words, I think the predicate changing operation would also have
> concomittant
> changes elsewhere in the EP.  Also okay?  (This has implications, btw, for
> the analysis
> of Lushootseed that Joshua is working on, where there are lots of stems
> that seem to
> be happy to serve as either nouns or verbs.)
>
> Emily
>
>
>
> --
> Emily M. Bender
> Professor, Department of Linguistics
> Check out CLMS on facebook! http://www.facebook.com/uwclma
>
>
>


-- 
Emily M. Bender
Professor, Department of Linguistics
Check out CLMS on facebook! http://www.facebook.com/uwclma
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20160921/c84b4c03/attachment.html>


More information about the developers mailing list