Frazier and Fodor's Sausage Machine 1978

Question 1

what does this sausage machine do

Answer 1

attempts to explain phenomena identified by Kimball without the principles - why some sentences are to process and why structural ambiguities are missed.
less principles, describes how use of linguistic knowledge interacts with working memory limitations more precisely.
explains RA and exceptions to it
explains mispredictions

Question 2

explain the sausage machine, 2 stage top-down parser

Answer 2

1. preliminary phrase packager, ppp. attaches words to sentences within 6 word processing window (WM)
2. sentence structure supervisor (SSS): assembles structures put together by PPP (sausages) into overall phrase marker (string of sausages) but doesn’t undo work of PPP.

Question 3

the principle of minimal attachment (one principle not 7 like Kimball)

Answer 3

MA, means parser always tries to make simplest attachment possible, when build structures, builds structure that requires you to create fewest nodes.
build simplest structure tree- so don’t see the more complex interpretation.
parser built with sausage machine structure.

Question 4

why does MA happen?

Answer 4

arises as natural outcome of race between alternative set of rules, creation of fewest nodes wins because its simplest and tasks least time to assemble.

Question 5

explain multiply centre embedded

Answer 5

difficulty of multiply centre embedded sentences explained in terms of lim capacity of WM (6 word viewing window) can’t process 2 centre embedded sentence as has left word viewing window, can’t make attachment b/c two parts of sentence never together in WM

Question 6

RA by default

Answer 6

explains exceptions to RA in short sentences and apparent use of RA in longer ones.
Accord to sausage machine, does this because start of the sentence has left WM therefore don’t have minimal attachment possibility so therefore use RA.

Question 7

advantages of Sausage machine (3)

Answer 7

accounts for more data than Kimball in only 2 stage model of parsing.
provides much more explicit account of how parsers use of linguistic knowledge interacts w WM
makes strong predictions about difficulty of different kinds of sentences and how ambiguities are solved in short and longer sentences.

Question 8

empirical problems with the Sausage Model (3)

Answer 8

Wanner 1980- pointed out that strong predictions of model are actually incorrect
multiply centre embedded sentences with only 6 words should be easy to understand but just as difficult
claims to capture RA by default, fits in 6 word viewing window, both inters use same number nodes therefore no minimal attachment, following RA not by default- you are making the choice therefore sausage machine doesn’t capture it.

Question 9

evaluation of sausage machine

Answer 9

better model- clearly specified, clearer predictions.
starting point for important current model of parsing GPM (Frazier and Rayner)
good theory should explain lots of data with small no principles.