multiple questions Flashcards
what does it mean if a sentence is structurally ambiguous ?
(in a tree diagram) the sentence can be divided in different ways
ex: while susan was dressing (herself? the baby? with what does “dressing” go with ? )
according to garden path model, what syntactical computation has costs ?
a. initial structure building
b. retrieval structural preferences from the lexicon
c. re-analyis and repair
re-analyis and repair
according to garden path, what is initial attachment based on ?
a. the structure which allows a simpler
b. universal heuristics
c. randomly selected among all possible alternative
universal heuristics
Which of the following heuristics of the
garden path model
are questioned on the basis of different attachment
preferences across languages of sentences like “I have seen
the son of colonel who took a train to Rome”?
a. main assertion
b. late closure
c. minimal attachment
late closure
what king of model is the race-based model ?
two stage model
which model does not assume immediate analysis of syntactic info ?
good enough model
hy the distinction between arguments and adjuncts is
relevant within constraint-based models?
stored in the mental lexicon and which is
build on the fly
Construal refuse what ideas of garden path ?
the fact that attachment principle holds all the time independently from what we are analyzing
frequency quency inheritance effects in
homophones is assumed to be an
evidence of:
a. psychological reality of lexeme
b. psychological reality of lamma
c. interaction of lemma and lexeme levels of processing
psychological reality of lexeme
The origin of semantic substitution errors can be:
a. lexeme retrieval
b. conceptual selection
c. phonological encoding
conceptual selection
the exchange of phonemes between
words happens more frequently for
the first phoneme
picture-word interference is strong
when:
a. you name a picture with a semantically related superimposed word
b. you read a word with a semantically related superimposed picture
c. you name a picture with a phonologically similar superimposed word
you name a picture with a semantically related superimposed word
The fact that phonemic restoration is
depends on sentence context shows that :
a. a role of motor representations in phoneme perception
b. the use of co- articulation phonemic feature in phoneme perception
c. top-down effects on phoneme perception
top-down effects on phoneme perception
semantic priming is stronger for:
a. pairs of words frequently associated
b. pairs of words of the same grammatical category
c. pairs of words of the same category
pairs of words frequently associated
One example of mediating priming is:
a. monkey banana
b. lion stripes
c. horse donkey
lion stripes
what is the type of role of unit and the type of search mechanism in morton logogen model ?
the units are active and there is no excplicit search mechanism
what is the flow of activation in the TRACE model
here is feedback from word level representations to letter level representations
in construction integration model what is the text base model of how the sentence is represented ?
a set of interconnected representation of propositions (similar to a network l)
a text lack global coherence when :
a. has non-legal sentence ( violation one level of the language analysis)
b. consequent sentence do not have overlapping argument
c. when is not possible to build a situational model
when is not possible to build a situational model
according to structure building framework what process takes more active and conscious use of mental resources ?
suppression of non related info
which is not a core aspect of text according to event indexing model
a. time
b. causation
c. hierarchy
which is not a core aspect of text according to event indexing model
a. time
b. causation
c. hierarchy
hierarchy
what is the main property of an intensional context
a. its pragmatic makes the meaning of a word in the already given context change
b. its semantic introduces a new concept not explicit in the given context
c. it changes the situational model build on the given context change
its semantic introduces a new concept not explicit in the given context
what are words order according to in FOBS model ?
a. first lexical component
b.first lexical component
c. word class ( category )
first lexical component( root )
Trace introduces the idea that :
a. flow of information mechanism cascades into the whole system
b. flow of information mechanism is discrete
c. mental representations of words are sub symbolic
flow of information mechanism cascades into the whole system
in the early model of COHORT, what type of processi s the contact stage and what is it based on ?
a fully interactive process
non- interactive
non- interactive ( it think it means no lateral connection like inhibitory ) and based on sensory input ( no top down or contextual info )
in the early model of COHORT, what type of processi s the contact stage and what is it based on ?
a. fully interactive process
b. non- interactive and based on sensory input
c. nn interactive and based both on sensory input and word frequency
non- interactive ( it think it means no lateral connection like inhibitory ) and based on sensory input ( no top down or contextual info )
which of the following aspects COHORT model is similar
to FOBS:
a. search is doe only on a subset of lexicon
b unit recede different level of activation bases on their frequency
c. feedback from word representation and phonetic representation levels
search is doe only on a subset of lexicon
what is the common criteria that Hyperspace Analog to Language and Latent Semantic Analysis use to affirm semantic relation between words?
likelihood to just coexist in the same text
According to structure mapping and focus theory: why are people shallow in comprehension ?
cause the language processing system makes them focus on specific aspects , and other are signaled as not to be processed throughly
what is semanticity
ability to use sound to refer to things that are different from a sound , the fact that sound carry meaning
what is referentiality
language use signs to refere to things that are not in ny other way related to that sign, the fact that word carry a specific meaning
what are the properties of language
- semanticity
- referentiality
- discreteness
- arbitrariness
- duality of pattern
- productivity
- compositionality
- recursitvity
- temporale and spatial displacement
what is discreteness
basically we can tell where one begins and the other ends
what is arbitrariness
there is nothing that connects the sound to the word other than our agreement that it does
duality of pattern
meaningless sounds can be combined into meaningful morphemes and words, which themselves could be combined further
temporal and spatial displacement
we can refer to thing that are distant in time and space
what is productivity
anguage allow to create new sentence for the fist time ( allow creativity )
what is compositionality
total meaning is defined by the meaning of the single basic units
what is recursivity
ability to place one sentence inside another of the same type ( which is besicallly infinite
what is phonetic
tudy of speech sounds that have a function ( phones ) and that don’t (phonemes) how they are produced, their properties
what is phonology
study criteria by which sounds are assembled to make words such as intonation and how they are different across langauges
what is morphology
tudy the rules b which different speech sound convey different meaning adn why they are put at certain part of a word ( including word derivation , composition , inflection etc..)
what is lexicon
study of communicative logic rule that if not respected , even if everything else is correct, can still give no meaning to the sentence
what is syntax
study which are the rules because words are put in a certain order, which orders are allowed and which not
what is x bar theory
X-bar theory makes the claim that every single phrase in every single sentence in the mental grammar of every single human language
state the difference between phones and phonemes
phonemes are the equivalent class of phones of the same letter
each letter , depending on the words , is pronounced slightly differently—-> each of those pronunciations is a phone
state the difference between syllables and morphemes
Morpheme is related to the meaning and structure of a word while syllable is mainly related to the pronunciation of a word
state the difference between phrases and sentences
Both are a group of words BUT in phrase the subject and verb are not seen together, so only expresses a simple meaning VS in a sentence, both the subject and verb can be seen, so it can convey a complete meaning.
list the main methods in psycholinguistic experiments :
reaction time: usually implement inference priming, masked priming, additional task( lie picture naming )
eye movement,
reading time
MRI ad fMRI
EEG and MEG —> problem: measure of activity cannot be localized or defined ( activation ? inhibition? )
===> secondary indexes, derived from the raw data lie ERP
what re the three universal stages of speech production ?
conceptualization
formalization
articulation
describe conceptualization in brief
define the concept to want to express and how it should be perceived
describe formalization in brief
choose what word you want to use to convey the right message ( sapir word theory has something to say about this) and the phonetic properties of them
describe articulation in brief
activate the set of movement that correspond to the phoneme of the words chosen
what is coarticulation
phenomenon by which the gestural score of the phoneme that follow influence the gestural score of the phoneme we are currently articulating , modifying also the auditory perception of it ( the coarticulated phones, despite being the same, will sound differently other than being physically produced differently )
pro of coarticulation an it’s evidence
pro: helps identifing the correct words when the auditory input is not precise ( because we lean on the other letter to deduce the damaged phoneme)
evidence from cross splice d stimuli : we take a coarticulated phoneme, and splice the first version of the coarticulation n a word that should have the second version of the coarticulation —> people will misperceive the word where the word where the
pro of coarticulation an it’s evidence
pro: helps identifing the correct words when the auditory input is not precise ( because we lean on the other letter to deduce the damaged phoneme)
evidence from cross splice d stimuli : we take a coarticulated phoneme, and splice the first version of the coarticulation n a word that should have the second version of the coarticulation —> people will misperceive the word as the one where the coarticulation is taken from ( jog will be percieve d as job if “jo” is taken form job
cons of coarticulation
cannot analyze phonetic properties because they are highly dependent on the word
what are the main models of production
Sentence Production Model by Levelt ( both at word level and sentence level)
spreading nding activation model by Dell
Independent Network model by Caramazza
main point s of levelt 1994 sentence production model
type: modular and lexicalist approach
stages:
- message ( here you have already gone through conceptualization, so you know what you want to say and have some kind of syntactic structure )
- functional processing : lexical selection (search in the mental lexicon the lemma)+ function assignment ( grammatical function is retrieved too)
- positional processing : inflection ( decide the lexeme for each lemma) + constituent assembly ( put the lexeme in the right spot of the structure)
phonological encoding : select phoneme , the corresponding gestural score and stimulate it
main potino of levelt 1999 word proaction model
type : again, modular and syntactical approach but senate level this time
stages:
- conceptual preparation; kinda like conceptualization but for a single word
-lexical selection: —> lemma
morphological encoding —> morpheme
phonological encoding–> syllabification–> retrieve the sound for syllables ( phonological words )
phonetic encoding : retrieve the actual phonemes for the syllable –_> gestural score
articulation
explain the frequency inheritance effects
speed at which words are produced is not dependent only on frequency–> it depends on how many time we use the collection of sound s even if they mean different things —> infrequent words are still read fast if they have a frequent homophone
what phenomenon are explained by self monitoring and self repair
- lexical bias in speech errors: when making error we still produce real words, cause they are not detected as real error since they are actual words
- limited resource availability : errors are more easily detected dat the end of sentences because we don’t give many resources to sentence planning
- automatic and fast
what are the rules of phonological errors in speech production and the type of errors that derive from them :
positional constrain: exchange of syllables happens at between syllable in the same position (ex first syllables of a word with the first sillableof another words )
phonotactic rules : wrong word still present combination of sounds that are allowed in the language
what are the rules of word exchange in speech production and the type of errors that derive from them :
lexiacl bais
word category contrain: wrong word tend to be of the sam ecategory of the original one
main points of dell reading activation model
mechanism to chose the outcome at each stage of any model
-cascade activation : ex: once lemma at lexical selection stage receives activation, it immediately send it to the phonological encoding
- gradual activation: is not all or none
- feedback activation : whatever final product is more fitting , it send back positive reinforcement for its own component s
main points of Independent network model by camarazza
due to its criticism ,camarazza model see to there be multiple independent representation for each word, one for spoken and one for written and one for structural info ( syntactical and grammatical ) that is accessed only when needed
what are the criticism camarazza make to lemma theory
given the stages in both levelt snf dell models, if grammatical level info is disrupted then so would be word retrieval
BUT
in neurological patient with brain lesions, words that have grammatical meaning are retrieved only in spoken modality but not in written , and viceversa—> meaning that grammatical info is not necessary for word/sentence production
what are the main point of motor theory of speech perception ?
-words are not recognized by categorizing them in sounds patterns , but in the set of movements we use to produce those sounds ( formant structures and formant transitions )
-principle of preemption : there is a specialized system to deal with linguistic auditory stimuli , different from the rest of auditory sister–_> when incoming stimulus is recognized as speech, the system prevents the general auditory system from acting
- info overlap in time–> difficult categorization But good for damaged input ( also explain mcgurk effect )
- mirror neuron involvement : motor cortex are activate when thearing the words ( meaning we associate movement with sounds ( different motor areas for different words, same areas for same word both imagined or heard )
what are the problems with mirrr neurons involvement in motor theory of speech perception?
- babies can percieve difference in sounds even if don’t have gestural scores for them (the latter must then not be necessary)
- aphasic patient with speech production disruption at a phisical level can still perceive different words
- mirror neurons where found in macqques and only assumed to there be a correspondent network in humans
what are the main point of general auditory approach to speech perception ?
we distinguish word based on specific characteristics in the sound , so we kinda have a prototype for each speech sound and recognize stimuli as linguistic if they match the prototype themselves
main model is the Fuzzy Logical Model of Perception:
- we use bottom up gradual activation of possible matching phonetic representations
- top down processes using stored memory to see which of these representation are an actual thing
what s the main advance of general auditory approach?
phonemic restoration effect )
ex: For example, when the /n/ phoneme precedes the /b/ sound (as in lean bacon), often times coarticulation makes the /n/ phoneme come out sounding more like an /m/.
bottom up activates both phoneem
but top doww knows that leam bacon is not a thing→ activate N
what are the main model of lexical access
first generation :
-logogen
-fobs
secondo generation:
-trace
-cohort
third generation :
-simple recurrent networks
-distribuito cohort
main points of logogen
active units : auditory input , visual input and context relatedness enhance activation of stored memories
passive mechanisms
bottom up
all or one activation ( threshold)
kind of refractory period —> explain priming between layer activation
main point FOBS :
passive units
active mechanism–>self terminating
bottom up
explain :
what phenomena are explained by fobs?
why pseudo roots take longer to retrieve
why root frequency is better at predicting frequency effect
why morphological priming is stronger at short onset in masked priming ( we rely more on morphological similarities if we are not completely aware of the semantic meaning of words due to masking ) and why this effect decline with time ( we then become aware of the semantic unrelatedeness
main points of trace
active units
passive mechanism ?? ( there are between layer connection)
both bottom up and top down (back activation)–_> good for damaged stimuli
cascade activation
parallel
between layer excitatory connection
what is the main phenomenon explained by trace ?
word superiority effect
main points of cohort
passive units
active mechanism?? ( inhibitory connection )
incremental
cascade actiavation
parallel at the beginning ( cohort ) and serial after ( selection)
inhibitory connection within layer ( option actively compete with each other but only after selection stage )
word activation goes directly to phonetic encoding without ( no homophone problem
what are the three stages of cohort
activation: multiple possible option ( that could match with auditory stimulus) are activated —> parallel and bottom up ( NO CONTEXT INFLUENCE )
selection : context help inhibit the not related candidates –> competition start ( serial )
integration ation: higher level of analysis ( grammatical , syntactical etc..—-> =incremental ) is done to see what candidates fit better
this process is continuously updated letter by letter of each point, untile recognition point is reached where one candidate wins
what are the evidences of cohort ?
cross modal priming : at short SOA, a word A can prime both a related wordB and a and unrelated word C if word B and C have similar onset , until recognition point is reached ( ex: ship both prime captain and captive until we reach the recognition point, then in only times for captain
why mismatch in words at the beginning lead to faster recognition of the correct word than if mismatch is at the end ( in trace does not matter where mismatch is )
why pseudo words with early recognition points are spotted earlier than psudowords that hav later rec point ( in fobs it would be the same )
why recognition point is a better prediction of reaction times than frequency ( generally against fobs )
what was the modification of updated cohort and its evidence
cold not explain the frequency effect as we ahem seen —> units are made active and added higher reactivity for frequent words
evidence : Dahan, Magnuson, & Tanenhaus (2001) used presented a auditory stimulus with ambiguous onset ( ex: bell, bed, bet, belt…),together with e with a number of objects is displayed
eye tracking showed look more often and more quickly at a picture that goes with a high-frequency word than a picture that goes with a lower frequency word
main point s of simple recurrent network
parallel
like trace But plus context unit ( effect of previous processe , like context is taken into account)
words are represented day the vector of units activated—> easily comparable
how are frequency effect implemented in SRN ?
Frequency effect are simulated by presenting more frequent words during training.
what is the training process for SRN?
Before training, the network’s connection weights were randomized,
then it processed a set of sentences one word at a time: as each word was encountered, the network tried to predict what the next word would be. When it made errors, the connection weights throughout the network were changed so that its output would more closely match the desired output the next time around. I
main points of distributed cohort
like cohort but with the structure of SRN :
- activation stage is changes from phonetic-word match and then to its meaning and properties , here the phonetic info is connected directly to phonemes and semantic components ( skipping the word stage )
- selection stage consist in the assestamento of a single pattern of activation
- influence of context in section stage is transposed in a hidden layer of context units
- less importance to recognition point oitn
what are the theories of semantic access
semantic network theory
associative accounts of meaning ( HAL and LSA)
indexical hypothesis
briefly describe semantic network
a word meaning is represented by all the other things that comes to your mind related to its (nodes) and the tipe of relation between these and the word ( link—> “is a “ it’s a type of link
spreading activation con tanto di decay function
what type of priming does semantic network spreading activation explain ?
semantic priming
mediated priming
non immediate priming of unrelated words ( given that the task explicitly asks so)
proves of separateness of associative priming
aster RT to associative relation tan to semantic relation
alzeheimer patient show associative bt not semantic priming
terre are different N400 waves in ERP for associative+ semantic priming than for bth only associative and only semantic priming
what is the problemof associative accounts of semantic access that originated the indexical hypothesis?
asscoitives models did not have grounding for the symbols
what are the steps of indexical hypothesis
activation of perceptual symbol ( = actual defined representation o the concept )
definition of affordance
integration of affordances
describe why we think of implementation of motor neuron in indexical hypothesis
subjects’ responses were faster when the motion undertaken to answer the question matched the motion indicated by the answer sentence
words related to actions led to increased neural activity in the same parts of the brain that became active when subjects actually moved the corresponding body part.
that he parts of the brain that become active when a person views an action also become active when the same person reads a sentence describing the corresponding actio
what are the two general strategies we use when parsing
immediacy principle
incremental processing
hwich are the main model of parsing?
garden path
multiple constrain
constural
race based
good enough
which are the two methods to parse long distance dependencies
gap and traces
active filler
main ponte of garden path
serial
automatic
incremental
uses heuristics
two staged :
- categorization
- meaning assignment
main points of multiple constrain model
incremental
parallel on multiple levels of analysis
context taken into account
competition for activation
what are the evidence of the use of multiple constrain over garden path
-context and visual context effect: implementation fo context help reduce Rt which should not happen according to garden path
- subcategory effect : option coherent with subcategory rules show faster rt
- no longer rt for more compelx structure are recorded ( agains garden path ) if they respect the subcategory rules
- people infer a missing argument in cases where a verb requires an argument.:
- semantic effect: reduced rt are showed if RR ( usually more costly) have words that she coherent semantic structure with each other ( “the defendant examined” and the “evidence examined”
what are the criticism to multiple constrain ?
sometime simpler structure is still preferred over more likely
multiple constati is hard to test
in some sentences the context effect does not verify
main point of construal model
garden path for primary relation and multiple constarin for non primary relations ( so context can influence and multiple structures are constructed only for non primary relations)
main points of race based model and evidence
limited resources
two stage :
- simultaneous activation of multiple plausible grammatical structures ( parallel )
- race for activation
choice between temporal structural ambiguity where only one choice is correct will take longer according to race becausein half of the case the wrong structure will win and reanalysis will be needed , while ambiguous structural ambiguities will be easier cause whatever result will be ok
main reasons for good enough parsing
inability to correctly parse
reluctant to let go fo a structure only because parsing is incorrect
which are the model of discourse processing
construction integration processing model
structure building framework
Event Indexing Model ( EIM)
structure mapping and focus theory ,
Current state strategy model
min points of construction integration model
three level of representation : surface , text based, situational
two stages ( chunks of text per time, not single words ) :
- construction: once surface and text based level of rep are made, all associated concept activated ( automatic )
-integration: if related proposition are active in the working memory we link them into a situational model
if not in the woken memory, we search in LTM,—–> if to there either, we discard the proposition
what are the evidences of all three le level of representation of contraction integration model ?
if asked to remember two surface level ( so word by word) , are mistakes are made if they share situational model
alike , if asked to say whether a word was present in the story, words that belonged to the situational model were more likely to be stated as present in the long term , text bases in the short term
found that words that came from the same proposition were much better retrieval cues than words from different propositions, even when distance in the verbatim form was shorter
Errors increase geometrically as the number of propositions to be remembered increases,
If any part of the proposition is recalled it is very highly likely that the entire proposition will be recalled
reading time increased the more proposition in the text
a free association task most likely responses are word from the same proposition
main points of structure building
three stages:
- laying the foundation —> concept are activated in an automatic way—> once activated they send either enhancement or suppression
- mapping
-shifting
what is one evidence of enhancement being automatic and suppression lmroe effortful:
saying to to the wrong meaning of a a word with multiple meaning takes time in short time frame, but this phenomenon disappears as time goes by
what are the main point of current state energy model a
says that comprehension is achieved when understanding causality
so each sentence needs to be the consequence for th previous ,AND the antecedent for the next, otherwise etc e chain is interrupted
the easier we find the antecedent , the faster we understand
what are the effect of coherence on rt and memory
sentences with higher coherence have faster RT but middle coherence are the easiest to remember ( due to depth of processing )
what are the proof that IA ( so bridging inference ) is more costly than DA?
longer RT for DA ( even when controlled for word repetition)
NAI
bigger N 400 ( related to comprehension effort) and P 600( related to semantic integration costs= situational model ) for AI —> those indexes indicate that our brain is doing something rather so it is understanding more in AI compared to other considitons
what does NAI theory explains ?
why we don’t have cost for bridging inference if there are verbs needing enriched composition : the “extra step “ we usually do when we we look for an antecedent in IA cases, is compensated with the fact that enriched composition mental question ( called QUD —> ex. want a beer = want what ? to drink) serve as a salient antecedent ( to drink will become an antecedent even if not explicated)
what are the evidences of motor theory
coarticulation = two formant transitions , both of the same letter but from two words with different coarticulation o the transition., are perceived as two different sound –> this is because , despite being the sam phoneme , the gestural score is different
duplex perception : when earring two superimposed syllables but with same formant structure, we still hear a sound ( either of the two syllables ) because due to principle of preemption, e still recgognise it as speech
VOT: some letter differs only by time onset are still perceived as iifffeenet due to gestural score
