Greg Detre
@12 on 22/11/00
Presentation
grammar as conventions that allow us to get much more out of words and their combinations
the PoS ≈ roles � case-marking/word-order
so you could see �understanding a sentence� as understanding the role played by the individual words, and the relationships between them as modulated by the other little words
however, this syntactic processing is bnound up with semantic processing
e.g. in resolving ambiguities = where the tree could branch in different ways or at different places
however, we get the input word-by-word, and we don't wait for it all before going to work
hence �garden path� sentences, where we guess, wrongly
what affects how we guess?
preference for particular grammatical structures?
least �complex� grammar?
most frequently occurring (≈ simplest) structures
context (�)
constraint-satisfaction of all these factors � if one is missing, rely on the others (e.g. old experiments without context)
spoken vs written. lack of prosody, which adds meaning or disambiguates (Tannenhaus timing experiment)
mental representations
EC(???) = a gap which can be filled by a trace
trace = a silent fille category
Chomsky and deep/surface structure � inflectin + case-marking vs position
locality � the EC is not local to the word� - need keep in memory
generative grammar � produce novel sentences using finite rules/input etc.
competence < performance
universal grammar = the blueprint of innateness, common to all human language
parameters differ, but share deep structure
levels: Deep + Surface structure
surface structure = what you hear
e.g. J is eager to please (actor)
����������� J is easy to please (it is easy to please J)
�wanna� contractions
which team do you want to/wanna beat
linguistics is descriptive
importance of native speaker
which team do you want ___ to win?
supposedly cannot here contract to �wanna� because there�s a gap inside(???)
who do you think ___ will win the game?
but you can contract to �think�ll� � against Chomsky
Derivational theory of complexity
least complex deviation of a sentence is the most accessible/easiest to interpret
easy transformation
minimal attachment � minimising the number of phrases
active sentences processed faster than passive � yes
positive processed faster than negative � yes
Slobin implausibility affects the ambiguity and reading
implausible sentences showed all the same effects, so semantics had to be involved
Rapid Serial Visual Processing � following eye movements
eye goes back when reading garden path sentences
but this problem goes away if you provide context
Gain(???) + the multiple elephans and relative clauses
Grice � don�t add irrelevant info
2)�� Separate processing of syntax
usually talk of modularity on a larger scale
which do we process first?
Fodothal(???) � both processed at the end of each clause
Marslen Wilson � both processed together���� ???
2 main models
serial � Forster (autonomous) � process words, then syntax
not information flowback, yet syntax does affect word recognition
if 2-way connection with mental lexicon + the GPS
interactive � Marslen Wilson � the different modules interact + swap information with each other
Swinney � access both meanings of a sentence
showed that it primes both, until you reach the end of the sentence
Rainer � eyes go back to main verb with ambiguity unless the semantics cleared things up
discounted full interactive/completely parallel models
ERPs � different reaction to semantic/syntactic implausibility
Gibson 98 � TiCs 2, 7 � constraints on sentence comprehension
lexical
contextual � visiting relatives are/is fun
computaitonal resources � locality
frequency � phrase-level
believe (EC)(believe that������� (???)