[CaCL] cacl this week

[Nathan Rasmussen]

On Mon, Nov 30, 2015 at 9:07 AM, William Schuler <schuler at ling.osu.edu>
wrote:

> hi all,
>
> i got more interest in the lin wang work, so let's do that this week, and
> some interest in multiword vectors / morpheme RAEs / bob papers, so let's
> do that next week.
>

 Friday next week is the first day of finals. Are we really meeting then?



Ideas for whenever we do meet next:

In connection with multiword expressions, I've been eyeballing a paper by
Peter Turney (J of Artificial Intelligence Research 44 (2012) 533-585,
Domain and function: A dual-space model of semantic relations and
compositions), which provides a single, capable model for both relational
and compositional tasks (previously, each task fared better with a
different vector-space model). Among its interesting properties is that it
doesn't actually calculate a vector representation for 'dog house', but it
still allows calculating how similar it is to 'kennel'. Anyone else want to
read more?

The introductory chapter of Simon Charlow's dissertation is a 15-page
overview at high level. I'd suggest we read that far, and then decide if
we're going to pursue it further.

Quick summaries of three of the processing-constraint papers:
- Hofmeister and Sag 2010 shows that reading time and acceptability vary
inversely in filler-gap constructions.
- Hofmeister et al 2013 argues that discourse-linking of wh-phrases eases
island violations not because of anything special in their syntax but
because it eases a cognitive load with widespread effects.
- Hofmeister et al 2014 nails down the connection between processing costs
and acceptability judgements more precisely by showing how violations of
processing constraints compound one another and how the
processing/acceptability curve varies with the reader's working memory.

Given this, I'd probably spring for the 2014 paper as the most inclusive
and concrete.

Finally, I've forgotten what we were talking about non-statistical learning
of. What was it?

Nathan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osu.edu/pipermail/cacl/attachments/20151130/ff6be6a4/attachment.html>