[Lexicog] FrameNet

[sr_shead]

Yes, there is an imbalance - and I think your "not yet ready for prime
time" comment is about right.

In theory, FrameNet could (should?) be more sophisticated than it
presently is, when it comes to schematic theta-roles vs. frame-specific
roles, and the relationship between them. FrameNet has the capability
for frame-to-frame relations of various kinds; the one relevant to this
problem is called "Inhertance" in FrameNet. Many frames "Inherit" from
more schematic/abstract frames (some of which are labelled
"non-lexical").

Moreover, if frame B inherits from frame A, then all the core "frame
elements" (FEs) from frame A should be mapped to core FEs in frame B -
though they will often have more frame-specific labels in frame B.

To take an example from the current FN data: The "Apply_heat" frame,
which contains verbs like "bake", "cook", and "fry", inherits from the
more schematic "Intentionally_affect" frame (which itself inherits from
both "Intentionally_act" and "Transitive_action").

The "Intentionally_affect" frame has three "core" FEs: Act, Agent, and
Patient. The Agent and Patient FEs (or theta-roles) are mapped to
frame-specific labels in Apply_heat: namely, Cook and Food
(respectively). (I've just discovered that they've done something very
strange with the "Act" FE ... but I'll leave that for now.)

So in a sentence like "Elke fried the eggs", two analyses are possible:
Elke [[Cook]] fried the eggs [[Food]]

or:
Elke [[Agent]] fried the eggs [[Patient]]

Although FN only presents the first analysis, the second is implicitly
present by virtue of the Inheritance relations. Personally, I find this
general approach quite satisfying.

Unfortunately, FN is VERY sporadic and inconsistent in its use of
inheritance relations. In the case of "treat", the sense tied to the
"Cure" frame inherits from "Intentionally_affect" - and Agent and
Patient are mapped to Healer and Patient (presumably in a more specific
sense!). However, the sense of "treat" tied to the
"Processing_materials" frame does not inherit from anything. There are
currently many, many loose strings in the data.

And on your question:
> To follow this to its logical extreme, shouldn't each agentiveverb --
and each sense of each verb -- have its own specific type ofagent? How
about these examples?
>
>     John [[stumbler]] stumbled over a root.
>     John [[drinker]] drank a cup of coffee.

I'd say yes ... and no. If you wanted to "split" your analysis to that
degree, then I suppose you could give each agentive verb (sense) a
specific agent-label. Then again, the premise of Fillmore's Frame
Semantics is that several LUs can evoke the same conceptual structure -
e.g. "drink", "quaff", and "swig". Their aim is to provide labels for
the elements in that conceptual structure (in which case Drinker would
do for all three verbs).

Of course, you can always dispute granularity. FN puts "drink" in a more
general Ingestion frame. Should it be split into "Ingestion_liquid" and
"Ingestion_solid" frames?

Stephen Shead.

P.S. What about the "Act" FE? Frequently, the "Act" FE from
Intentionally_affect is left hanging - it is not mapped to any FE in
inherited frames - which is something of an anomaly, going from their
theory. Sometimes "Act" is mapped, but in bizarre ways. In "Apply_heat",
Act is mapped to "Duration"!!?! There are other similar examples. Go
figure...


--- In lexicographylist at yahoogroups.com, "David Frank" <david_frank at ...>
wrote:
>
> from David Frank:
>
> If Agent is more generic than Healer, then it seems to me that there
is an imbalance in the FrameNet analyses of these sentences:
>
>     John [[Healer]] treated Mary with antibiotics.
>     John [[Agent]] treated the woodwork with creosote.
>
> The second of these sentences should have been given a frame that is
more specific than simply Agent, to match the level of specificity in
the first example. Or else John should be called an Agent in both cases.
>
> To follow this to its logical extreme, shouldn't each agentive verb --
and each sense of each verb -- have its own specific type of agent? How
about these examples?
>
>     John [[stumbler]] stumbled over a root.
>     John [[drinker]] drank a cup of coffee.
>     et cetera
>
> In the first example, of John treating Mary with antibiotics, how
would you distinguish that from the following?
>
>     John healed Mary with antibiotics.
>
> In other words, when John treated Mary with antibiotics, what if that
treatment didn't actually have any positive effect? Is he still a Healer
then? On what basis? I agree with the concerns that I have heard, that
this all seems very ad hoc and inconsistent. Maybe the problem is that
it is not yet "ready for prime time."
>











------------------------ Yahoo! Groups Sponsor --------------------~--> 
You can search right from your browser? It's easy and it's free.  See how.
http://us.click.yahoo.com/_7bhrC/NGxNAA/yQLSAA/HKE4lB/TM
--------------------------------------------------------------------~-> 

 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/lexicographylist/

<*> To unsubscribe from this group, send an email to:
    lexicographylist-unsubscribe at yahoogroups.com

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/