Re: Changes in synsem and vartrans

Hi Philipp, group

> The general comments, which given that this Friday's is the last Telco and
> Philipp wants to freeze the model, are more discussion points in the case
> that there is time and willingness:
>
>  I think having an OntoMap class is a better solution than having the two
> different sets of arguments, semantic and syntactic, if you're going to be
> strict about semantics by reference that is and about having all the
> semantic information in the ontology (though personally speaking I don't
> think you can be too strict here without making numerous theoretical
> assumptions about natural language semantics or pissing off the ontology
> modelling guys who're going to want the ontology to be as language
> independent as possible, but that's a discussion for another day...).  That
> being said, I did notice that in the VerbNet lemon-encoding, semantic
> arguments are explicitly mentioned (using the semArg mapping) -- even if
> they are identified with syntactic arguments -- and semantic predicates are
> subclasses of Lexical Sense. With the OntoLex model I guess you would
> expect semantic predicates to belong to an external ontology or for users
> to construct their own, but I don't know if users will always want to do
> this (the lemon version of VerbNet being a case in point).
>
>   I am not sure about the VerbNet lemon encoding. But the position of
> ontolex is clear: semantics is in the ontology. What is not in the ontology
> does not exist.
> If things exist that can be refered to in natural language, then they
> should be part of the ontology. If the ontology has limitations then this
> is not an ontolex problem, but a problem of the ontology...
>

Fair enough.

>
>  Secondly, I don't get why OntoMap necessarily needs to be a subclass of
> Lexical Sense. Indeed I think that intuitively speaking such a
> syntactic-semantic mapping seems to play a different role from
> LexicalSense. Also it may well be the case that two different senses of a
> word share the same synsem mapping (depending on how fine grained your
> ontology is), e.g., subject -> Agent, object->Patient; making ontomaps
> separate senses muddies the water in this case. Additionally synsem/Example
> 7 seems to bundle lots of different things together and is complicated and
> hard to understand (for me anyway) largely because :giving_semframe is both
> mapping the entry to an extension (the class of giving events) and
> describing its (meaning-influenced) syntagmatic behaviour.  I think it
> would be easier  (obviously at the cost of concision) at least conceptually
> if :give had a sense that pointed to <http://example.org/giving> and each
> of the synBehaviours had a separate OntoMap to the ontology, or even one
> mapping with submappings describing its argument structure.
>
>
> The LexicalSense implements the lexicon-ontology interface. A sense is a
> pair of lexical entry and corresponding concept. An OntoMap is a subclass
> that relates a lexical entry with internal structure (e.g. with arguments)
> to a structure in the ontology. Thus, it is clearly a subclass of a
> LexicalSense.
> Lexical Senses can not share the same mapping because the arguments of a
> syntactic behaviour are unique for this behaviour, so the mapping to
> ontological arguments needs to be specified for each syntactic behaviour...
> so two senses can not share the same mapping...
>
> I hope this clarifies. Otherwise, we can continue discussing tomorrow.
> Thanks for your input.
>
> I always understood senses as tying a lexical entry to an extension, but
not necessarily matching up one to one with a verb's argument structures.
So for a verb like "to eat", there are at least two syntactic behaviours,
i.e., transitive and instransitive, as in 'Fahad ate an apple', and 'Fahad
ate', and with two different argument structures (or maybe one, with the
Object being optional). But both of these sentences use 'eat ' in the same
sense. On the other hand it's possible that I can understand  the instance
of 'to eat' in "The ATM ate my card" as a different sense of the verb but
with an argument structure <Agent, Patient> that matches one of the
argument structures of the first sense. IMO it all depends on the
theoretical framework you're operating in.

On example 7: certainly not the most compact one, but I added this example
> because you challenged us to represent the alternation for X gave Y Z"
> and "X gave Z to Y
> I have tried to completely spell out how to do this? What exactly would
> you propose to remove from the example?
>
> Kind regards,
>
> Philipp.
>



I don't propose to remove anything from the example per se. I still think
it's a good example and illustrates the kind of issues you face frequently
when working on the representation of verb syntax -- and that it is
therefore important to have such an example in the specifications. I was
just pointing out that the way of dealing with it using the OntoMap class
is a bit hard to follow -- at least for me: it took me a while to get to
grips with.

Cheers,
Fahad


>