Comments and questions for Levin intro
	======================================


- Goals

For Levin, the key theoretical idea is meaning components. Verb
classes are primarily a tool for uncovering meaning components.

Unfortunately, we never see any meaning components? What are we
to assume that they are like? The largest example is the
analysis of "touch","hit","cut" and "break", leading to the
conclusion that

- "touch" is a pure verb of contact
- "hit"   is a verb of contact by motion
- "cut"   is a verb of change of state involving contact and motion
- "break" is a pure verb of change of state

The middle construction

    "This bread cuts easily" is associated with change of state

as is the nominal sense of "cut" and "break" in which we are
dealing with the result of the action.

The conative construction

    "Kim cut at the bread" is associated with contact and motion


The Body-Part Possessor alternation

Kim cut Sandy's arm 
Kim cut Bill on the arm

is associated with the contact aspect.


This is all well and good. These verbs are however ones in which
our physical intuitions work fine. Moving to the larger scale, what
do you think about the practicality of extending to less concrete
verbs? How feasible is the program of identifying meaning 
components then building lexical representations outlined on
p 19.
 

- Methodology

There are a number of traditional ways of classifying verbs. By
intuition alone it is hard to tell how to phrase the description
of meaning. See for example the discussion of "whistle" on p15. I
am concerned about the assumption that one of

"make a shrill clear sound, esp. by rapid movement"

or

"to move with a whistling sound caused by rapid passage through the
air"

is "correct", and the other not. 


The use of syntactic cues in the
sequel seems to me totally plausible, but what does this tell us
about the verbs? 

Maybe the point should just be that the syntactic behaviour of "whistled
through the window" is good evidence for the motion sense.

- Generality

The discussion of "arrossire" = blush on pp 13-14 is interesting? What
if anything does it tell us about the cross-linguistic relevance
of these ideas? Why should we believe (or not) in "bodily processes"
as relevant?

- Prospects for CL

It is now more practical to carry out "exhaustive hypothesis checking"
than it was when Levin wrote. 



Assume that we have a large quantity of
unlabelled text. Suppose that there are indeed meaning components
and that their behaviour is reflected in diathesis alternations, which
are again reflected in patterns in the corpus.

We will want to abstract away from details of the corpus. How should
we do this?

Let's be precise about what we expect to see in the corpus if this
hypothesis is true (so we'll also have to consider what would be
seen if the hypothesis is false). 


Following standard CL practice, let's try to phrase this as a problem
with hidden variables. Can you draw a diagram of the hidden variables
and their relationships?