[Lexicog] FrameNet

Wed May 31 11:17:21 UTC 2006

[from  lexicographylist]

Rudy, thanks for the reminder. Fillmore's Frame Semantics is an inspiration
to all of us working in this area. It is certainly one of the most important
theoretical developments affecting lexicography.  It assigns semantic roles
to valences, and that is a major step forward.

However, I do have reservations about FrameNet (the applied version of Frame
Semantics), and it does not quite do what you say it does. Some of my
reservations are mentioned in an article that I wrote with James
Pustejovsky, published last year in Revue francaise de la lingistique
appliquée (10:2), in whihc we outline our own, as yet unpublished, work.

My main reservation is methodological. FrameNet proceeds frame by frame, not
word by word. This may seem a trivial point, but it isn't. Although FrameNet
uses empirical data, it does not use an empirical methodology. The
methodology is that FrameNetters first think up a frame, then try to think
of all the words that might fit into that frame, then go and annotate some
of the corpus evidence for those words. This methodology creates two
problems. Firstly, FN occasionally misses an important member of a frame
(because the annotators did not think of it). Secondly, in quite a high
proportion of cases FN currently does not cover all senses of a word
(because "we haven't got to that Frame yet") and in some cases ("spoil", v.,
is one example) all main major senses of the word and covers only minor
senses. So FN cannot be used for tasks like word sense disambiguation until
all frames are complete, and we don't know when that will be. Worse still,
there may be senses that will never fit into any frame.  Even when FN is
"complete", it will require a completely different process, with a different
methodology, to check that all normal senses of all common words have been
covered.

A second reservation is that FN has no pre-defined inventory of semantic
tags. I believe they make them up as they go along. This may be a good
thing, as it is at least empirically well founded, but there is a danger of
an uncontrolled explosion of tags, and, as far as I know, FN's semantic tags
are not systematically related to an ontology or inventory of lexical items.
So how is a user to know which words in a text realize which tags?

Thirdly, the relationship between semantic types and semantic roles needs
attention. To take an example:

John treated Mary with antibiotics = [[Healer]] treated [[Patient]] with
[[Medication]]
John treated the woodwork with creosote = [[Agent]] treated [[Material]]
with [[Alterant]]

Does FN provide sufficient information for the interpretation that in one
case John is a Healer and in the other an Agent? The semantic type of "John"
is [[Human]]; there is nothing explicit in the text that says that he is a
Healer or an Agent. This is a semantic role assigned to John (semantic type
= Human) by the context. This is a systematic problem with FN, which we are
trying to address in the Verb Pattern Dictionary (watch this space).

These are a complex issues, and I won't go on. It was what I had in mind
when I said that Agent and Patient (in the grammatical sense) don't get us
very far.  But my quibbles should not detract from the fact that FS and FN
represent tremendous advances in our theoretical understanding of semantics.

I will now put my tin hat on (a British idiom dating from WW2, I believe) -- 
and await a counterblast from Berkeley.  :-)

Patrick

----- Original Message ----- 
From: <rtroike at email.arizona.edu>
To: <lexicographylist at yahoogroups.com>
Sent: Wednesday, May 31, 2006 10:01 AM
Subject: [Lexicog] Semantic parsing

>
> Patrick's excellent and very clear exposition of the problem of semantic
> parsing is exactly what is being addressed in Charles Fillmore's seemingly
> little known Framenet project, which is using frame semantics to set up an
> automatic semantic annotation system. The url for the project is:
>
>         http://framenet.icsi.berkeley.edu/
>
> A description of the project at the website is as follows:
>
> The Berkeley FrameNet project is creating an on-line lexical resource for
> English, based on frame semantics and supported by corpus evidence. The
aim is
> to document the range of semantic and syntactic combinatory possibilities
> (valences) of each word in each of its senses, through computer-assisted
> annotation of example sentences and automatic tabulation and display of
the
> annotation results. The major product of this work, the FrameNet lexical
> database, currently contains more than 8,900 lexical units (defined
below),
> more than 6,100 of which are fully annotated, in more than 625 semantic
> frames,
> exemplified in more than 135,000 annotated sentences. It has gone
> through three
> releases, and is now in use by hundreds of researchers, teachers, and
students
> around the world (see FrameNet Users). Active research projects are now
> seeking
> to produce comparable frame-semantic lexicons for other languages and
> to devise
> means of automatically labeling running text with semantic frame
information.
>
>      A Google search shows that there is a Spanish Framenet project as
well.
> This is exactly the sort of detailed
> co-occurrence/collocation/subcategorization
> information that lexicographers need to be aware of, and that Patrick was
> calling for. Fillmore has been working on this for well over a decade,
> although
> for some reason his work seems much less well known than Adele Goldberg's,
> perhaps because he has not published a book on it.
>
>       Rudy Troike
>
>
>
>
>
>
> Yahoo! Groups Links
>
>
>
>
>
>
>

------------------------ Yahoo! Groups Sponsor --------------------~--> 
Everything you need is one click away.  Make Yahoo! your home page now.
http://us.click.yahoo.com/AHchtC/4FxNAA/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/