Saturday, December 8, 2007

Grove and Hale, 3rd draft.

CUNY Abstract

Thanks for all yr help, guys!

1 comment:

John Hale said...

OK, this is improved but still not ready to go out.

There are some ridiculously long sentences that obscure the argument being made, e.g the "within the continuum" sentence.

- it needs to fit on one page. get rid of this double spacing stuff.
- don't label your Figure as appendix material. becomes an excuse not to look at it
- explain what each of the symbols Crel, CIT, CPass, V_CAUSE etc mean. If you can't explain it succinctly, take it out.
put in a note saying that the numbers on the arcs have something to do with BNC
- no need to re-cite S & M after the data after they are already cited in the first para
- the thing in Figure 1 is not obviously a PCFG, so probably better to avoid the acronym altogether and call it a probabilistic grammar
- put your awesome title back in
- verify that there is a real transition between the antepenultimate para and the penultimate para, i.e with the word "scale" as suggested below
- use gb4e to make the examples into just 1a and 1b and put the UNACC marker underneath the verb

\begin{exe}
\ex \begin{xlist}
\ex \gll The butter melted in the \ldots \\
{} {} UNACC {} {} {} \\

\ex \gll The horse raced past the barn fell \\
{} {} UNERG {} {} {} {}
\end{xlist}
\end{exe}



split up this sentence:

Inspired by Dowty 91 we argue that butter is uniformly Theme
but that "the horse" is prematurely realized as Agent **PERIOD.**
After reanalysis, the parser is left with an unintegrated rept in which....


In our account, the effect follows from the information-processing difficulty of
transitioning between these parses. The account uses information theory to relate
the markedness of particular verbal morphemes to reanalysis difficulty.

This is made precise in a probabilistic grammar (Fig 1) whose parameters reflect
alternation ratios in the British National Corpus [provide citation giving credit for the corpus].


[ I would then move the sentence about animacy requirements into the separate paragraph where you talk about animacy requirements ]

By recognizing degrees of proto-Agentivity, the grammar locates different verbs'
garden-pathing difficulty on a scale where unacc and unerg are extremes.

The results of an acceptability study confirm the location of causative unergatives on this scale.
On a five-point scale, participants rated transitive unergatives with *animate* direct~objects
as good as unaccusatives. (2.27 vs 2.38 non significant, if indeed that is true).

....not as a difference in grammaticality but as a difference between alternative sequences of grammatical but unequally costly morphemes. The grammar in our model employs only head-movement thus avoiding any special appeal to a distinction between syntactic vs. lexical operations.