[Lexicog] Digest Number 1050

[Ronald Moe]

Mike Maxwell wrote:

"Paint me skeptical..."

 

Ah, a man after my own heart. On the positive side the DDP word collection
process works because people are able to start with (what seems to me like)
a hub and think of lots of words related to it. So there must be mental
links that we can use to recall a set of semantically related words. On the
other hand, Mike is correct that  you can take any word and come up with
links to other words. 'High chair' is a good example and Mike points out
some obvious links. It is used by small children (old enough to sit up and
too young to sit on a regular chair). It is used to feed the child, usually
be placing food on a tray that attaches to the arms of the chair. So 'high
chair' must be linked to 'child' and 'eat' (or 'feed') and 'arm (of a
chair)', etc. Or is it? Would it be more accurate to say that 'high chair'
activates a scenario that includes a parent placing the child in the high
chair and feeding the child? In which case the links are to elements of the
scenario rather than to particular lexical items.

 

Theoretical models are, by necessity, over-simplifications of reality,
otherwise we could not conceptualize them away from the mass of detail in
the real world. A model such as the one we are discussing may enable us to
make some observations that we couldn't without the over-simplification. If
we attempt to enrich the model with more detail, eventually we become
overwhelmed with the detail. Mike points out that there are different kinds
of links. In fact Steyvers and Tenenbaum's article talks about directional
links. In free association 'cat' leads to 'dog' in a single direction. 'Dog'
may also lead to 'cat', but not necessarily. Some lexical relations are
reflexive (e.g. synonym) but others are not. The part-whole relationship is
a one-to-many relationship.

 

Steyvers and Tenenbaum's article state that there are an average of four or
five links between any two words. But the examples they give are flawed. One
of the series they give is "volcano > mountain > sea > saw > tooth > ache".
But the meaning of 'sea' in the 'mountain > sea' link is obviously different
than the meaning in the 'sea > saw' link. In fact the 'sea > saw' link could
be interpreted two ways-as an inflectional link between 'see' and its past
tense 'saw', or as a compound link 'see' + 'saw' = 'seesaw'. In either case
the linkage makes a transcontinental leap from 'sea' to 'see'. Another
series is "volcano > erupt > blow > hair > head > ache". In this series it
is 'blow' that has two meanings. So they are tracing links between
phonological or orthographic forms and ignoring the semantics. Yet they
claim they are investigating the "semantics network". If so, these links
would have to specify which meaning of the word is being linked. So.

 

Paint me skeptical too.

 

Ron Moe

 

  _____  

From: lexicographylist at yahoogroups.com
[mailto:lexicographylist at yahoogroups.com] On Behalf Of Mike Maxwell
Sent: Tuesday, November 18, 2008 10:03 PM
To: lexicographylist at yahoogroups.com
Subject: Re: [Lexicog] Digest Number 1050

 

Ronald Moe wrote:
> Mike Maxwell wrote:
> "what does it mean for one word to be a 'hub', rather than another?"
> 
> This is a metaphor used in a theory of semantic networks. 
> ...
> Let's imagine that 'armchair' is only related to 'chair', but that 
> 'chair' was related to 'high chair' 'sofa' 'furniture' and 'stool'.
> In this case 'armchair' would be a node with only a single link to
> 'chair':
> 
> armchair-----chair
> 
> But 'chair' would be a hub (and a node) with lots of spokes:
> 
> furniture
> |
> armchair-----chair----sofa
> / \
> stool high chair
> 
> It turns out that 'furniture' is also a hub...

Right, but to go from a metaphor to a theory, there has to be some way 
to accurately measure (count) these links. For example, how do we know 
that 'stool' is connected to 'chair' instead of directly to 'furniture'?

And what about multiple inheritance? Is 'high chair' linked to both 
'chair' and 'table'? If not, why not? And is 'high chair' perhaps 
linked to 'baby' or 'infant' as well, or do only 'is-a' links count (as 
opposed in this example to 'for the use of' links)? How would we know 
which kinds of links count and which kinds don't? If we count links 
other than 'is-a' links, I suspect that every word might turn into a 
hub, because words which are specific have lots of links to other 
concepts--namely, the concepts that help define their specificity. And 
at a guess, there won't be much space between clusters.

Also, do incoming 'is-a' links (the one from 'furniture' to 'chair' 
count differently from outgoing 'is-a' links (the ones from 'chair' to 
'armchair', 'stool' and 'high chair')? And what kind of link is it 
between 'chair' and 'sofa'? If the two are instead linked directly from 
'furniture' (and linked to each other only by way of the 'furniture' 
node), then the count of links from 'chair' goes down; so if we're going 
to determine whether s.t. is a hub by counting the links, this is important.

I suspect there's also an issue of the gradiency of links. For example, 
if I'm camping, a stump or rock may make a good chair. Is there a weak 
link between 'stump' or 'rock' and 'chair'? What about a barstool that 
has a back to it: is it linked to 'stool' or to 'chair', and how would 
we know? Is a 'wheelchair' a 'chair', or is it linked more to 'bicycle' 
or 'tricycle' (or to some node above them)? Is a 'seat' in a car a 
chair? Does it matter whether it's a bucket seat, a bench seat, or a 
child's car seat? What about the seat on a motorcycle or bicycle?

It reminds me a bit of Fillmore's Case Grammar; very attractive, until 
you try to pin down exactly what it means. (Your mileage may vary :-).)

Paint me skeptical...
-- 
Mike Maxwell
maxwell at ldc. <mailto:maxwell%40ldc.upenn.edu> upenn.edu

 

No virus found in this incoming message.
Checked by AVG - http://www.avg.com
Version: 8.0.175 / Virus Database: 270.9.7/1799 - Release Date: 11/19/2008
8:58 AM


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lexicography/attachments/20081119/46fe2e49/attachment.htm>