Textbook: Chapter 11

garden path sentence

-cog system builds a phrase structure but then you realize the interpretation couldn't have been right --later in the sentence becomes obvious something is wrong with it -suggest that sentence parser assigns words to phrase structures as we read

texts

-3 levels of representations in text processing 1. surface code 2. textbase representation 3. situation model

experiment on solving speech stream

-8 month olds listen to 2 minutes of synthesized speech with 4 fake words said in a way that sounds continuous -then exposed to several isolated sounds --1/2 fake words and 1/2 crossed word boundaries -showed preference for 2nd (words that crossed word boundaries) suggesting that these words were novel to them -also appears that highly familiar words may play a special role in helping babies figure out word segregation boundaries -other research suggests this ability to distinguish statistical regularities extends to pictures, maybe not specific to language

letter to phoneme rules

-a set of rules that we know that convert letters and groups of letters into phonemes --cannot completely account for reading because we can also read words whose pronunciations aren't in line with the letter to phoneme rules ---example: colonel, pint

visual lexical decision task

-a string of letters appear on screen and participant must say if it's a word --open to semantic priming

what about words and adults?

-already have mental representations for words -most believe that people recognize words through a matching process --spoken word compared to a mental dictionary (lexicon)

dual route models account for:

-being able to read aloud fake words --IOW words that we have no lexical entry for ---we know a set of rules that convert letters and groups of letters into phonemes

writing and segregation?

-clear spaces between words so there is no segregation problem

what does the lexicon contain?

-representations of all known words to them -contains pronunciation, spelling, and part of speech -meaning -the lexical entry points to where meaning is stored

experiment on principle of minimal attachment

-shown 2 similar sentences that differed in phrase structure --one required a new node to be generated to represent the noun phrase results: -reading times were longer when minimal attachment was violated, the increased time observed at locations of violation, their eyes dwelled on there longer

solving speech stream (words)

-stream of speech solved by babies by sensitivity to the statistical regularities of language --example: hearing "pretty" followed by many different sounds suggest pretty" is a unit

how does the parser decide what fits where?

-syntatic and semantic clues 1. key word 2. word order 3. principle of minimal attachment -but semantic context also matters: can overturn principle of minimal attachment

understanding letters

-2 additional tasks to phoneme perception

experiment on voice onset time

-computer generated speech to vary voice onset time --all sounds within "b" category sounded like good examples of "b" until they were fine examples of "p" --answer is multiple choice, not blurry spectrum

experiment on surrounding context and phonemes

-cough during phoneme in a sentence --found that no one perceived that any of the parts were missing and were able to interpret correctly in a later experiment, replaced with cough when ends of sentences varied -heard different phonemes depending on the rest of the sentence -most not consciously trying to figure it out though --believed that they heard the completed word ---called the phoneme restoration effect

How are ambiguities resolved: phonemes

-difficult because of accents and coarticulation -3 strategies to overcome 1. categorizing phonemes 2. using surrounding context 3. including visual info

pinpointing difference between "bat" and "pat"

-doesn't come naturally to children --hearing individual speech sounds in words must be learned

what about reading?

-dual route models of reading say there are 2 mechanisms

mondegreens

-errors from misheard lyrics

understanding words issues continued

-even if you hear the word right, words have multiple meanings

are the two routes of the dual route model really separate?

-evidence for yes: acquired dyslexia (through brain damage) 1. surface dyslexia: reading of non words and regular words preserved but can't read irregular words -extreme case: K.T. --shows us that within dual route, damage was to spelling lookup while letter to phoneme was still intact 2. phonological dyslexia: can read irregular words but non nonword -evidence for yes: brain imaging data --two separate pathways for reading --letter to sound associated with left superior temporal lobe --direct lookup associated with junction of occipital and temporal lobes in the left hemisphere (visual word form area) AND lateral and anterior parts of temporal lobe

matching process

-first few phonemes that are perceived cause words consistent with it to become active --as more phonemes are perceived, some words become inconsistent with input and lost activation --remaining candidates gain activation

writing ambiguities

-homographs: 2 words with with identical spellings but different pronunciations and meanings -homonyms: 2 words pronounced and spelled the same way, but have different meanings --example: "right"

matching process experiment

-if you hear "capt" both "capture" and "captain" should become active but only captain should be active if whole word is said --test: listen to words and simultaneously perform visual lexical decision task ---if you have just heard "captain" will be faster to affirm ship is a word results -fast on either words when only said "capt" and fask on captain words given full word

coarticulation

-making one movement in a way that anticipates future movements or is influenced by past movements --example: tulip ---don't utter each phoneme in the order they appear -coarticulation makes phoneme production sloppy, and makes phonemes vary from word to word --so why don't we hear it as irregular

sentence parser

-mechanism that derives phrase striations from sentences --how does it work? concluded more like a conveyer belt -we don't really have the mental floor space to spread out all the words before deciding what word belongs in a phrase -auditory WM limits the number of items our sentence parser can simultaneously evaluate -interference also makes sentence comprehension across long string of words difficult -most times the gambles are good but sometimes the sentence parser makes a mistake

phoneme restoration effect

-missing phoneme restored by context and never consciously identified it as missing -this is hard for computers --so use to distinguish bots (CAPTCHA)

proposition (used by text base representation)

-most basic unit of meaning that has truth value -usually represents causes, events, objects -verbal descriptions but intended to be independent of any particular language -have syntax relation --read(car), gave (boy, girl, ball) -but a proposition must be a the most basic unit of meaning --wet, red ball would not be represented by that by by 2 ---wet(ball), red(ball) -propositions don't persevere particular wording and syntax so 3 different sentences with same words represent the same proposition

Sentences

-once we identify words, the problem of comprehension are mostly the same whether they are written or spoken

dual route model accounts for (summarized)

-our ability to read non words (example: slint) through the letter to phoneme rules -our ability to read irregular words (example pint) through spelling look up route -our ability to read regular words (either route)

3. principle of minimal attachment

-parser is biased to add new words and phrases to a node that already exists on the hierarchy rather than creating a new node

texts

-people make inferences that go beyond what the speaker/writer explicitly communicate -ambiguity about "why" --not ambiguity about the phonemes, word, or sentences

but it might not be that simple (about 5 dimensions that guide inferences)

-rather what goes in depends on your goals and expectations as a reader -example: when told to keep track of where MC was in the story, noticed contradictory info about space -also depends on whether the story is written from that character's perspective --1st person promote empathy for MC

experiment addressing the ambiguity of homonyms (same sound, different meaning)

-read sentences out loud --in which same word is used in all but only first time provide context for meaning --as soon as they finished one sentence, had to quickly read the next ---last word was the key word -it should be easier to read a word if a semantically related meaning was active in the lexicon --unrelated word provides a baseline about how quickly they can read a word -results --context affects lexical access --when biasing context matched final word, participants read faster ---if it didn't match, no advantage in reading time and presumably no lexical access for that meaning of the word

what would dual route say about "koat" or "phocks"?

-these are pronounced like other real words -would predict response time to nonword should be slower to nonword whose pronunciations aren't like real words because the two routes will conflict about the right answer --this has been verified

why do we make inferences about certain things and when?

-to resolve incongruity --we make it so that the text does make sense if it didn't ---quick constellation with LTM provides the necessary info -and 5 dimensions of the text that guide inferences

voice onset time and categorizing phonemes

-voice onset time only difference between "p" and "b" --short = "p" --long = "b" ---but we never hear a "medium, mushy" sound that's a cross between the two

3. using vision (third strategy to overcome ambiguities in phonemes)

-watch mouth -called the McGurk Effect --video "bababbaba" --dubbed with sound played "gagaggagga" --participants fuse 2 differing sources of info into a perception that best fits the visual and auditory pattern -McGurk effect doesn't just change auditory perception into something else but also auditory perception is better when you have supporting visual info

words and speech stream

-we perceive discrete words with small bits of dead air between --but not actually what strikes our ears --when we speak, we produce a continuous stream of phonemes ---there are pauses in speech but they are relatively rare ----why we call it a speech stream (emphasizes how continuous it is) -continuous nature would seem to be a special problem for children learning speech -even for adults, segregation of phonemes into words has the occasional error -pauses are more likely to occur within than in between words

words like colonel and pint

-we use the spelling of these words to establish they're in our lexicons

experiment on lexicon

-when someone pronounces a string of words, compares string with pronunciations in lexicon --if phoneme string matches an entry, the word sis identified and cog system gains access to other properties of the word (spelling, part of speech, meaning)

sentences

-word order to deriving meaning, but not the whole story --meaning can be unambiguous but at least a couple grammatically correct interpretations from a single order of words ---despite the different possible interpretations, few people perceive ambiguity and perceive intended meaning

3. situation model (third level of representation in text processing)

3. situation model -we not only interpret words and sentences but also construct a coherent representation of the world in which the text occurs -describes entire situations in which they take place -adds content from LTM that makes sense given the context -other info from the text that is judged to be peripheral to over all situation is cut -can be thought of as a summary -longest lasting rep in memory -requires far more prior knowledge in LTM than either of the 2 other reps --because most texts provide less info than the author intends for you to understand

Why are perception of phonemes difficult? reason #1

1. individual speakers produce phonemes differently, especially English speakers from different countries or even different regions -variation even more extreme among non native speakers --despite subtle variations, we able still able to correctly understand ---but it's true that native English speakers make more errors in perceiving speech of non native speakers ----the extent of the errors depends on strength of the accent

1. key word (parser deciding what fits where)

1. key word - a word that reliably provides a cue to a bit of phrase structure organization -eg: "a" indicates a noun phrase follows --evidence when key words omitted from sentences ---results: faster and more accurate at paraphrasing when contained relative pronouns

5 dimensions of the text that guide inferences

1. main characters 2. where they are 3. the time that things happen 4. what causes the events 5. the goals of the main character -these answer what foes into situation model and what is omitted)

additional task to understanding letters (first)

1. need to recognize each letter and distinguish it from others -which is a purely visual task -orthographic processing -recognition is not too difficult (there are only 26 to distinguish) --although fonts will have a set of critical features to make the letter "A"

dual route models

1. one route just matches written word to spelling entries in lexicon -with access contains pointer to meaning 2. uses a translation procedure that converts written letters into sound and then matches the sound to an auditory entry in lexicon -then recognition of the written word is similar to recognizing a spoken word

1. categorizing phonemes (first strategy to overcome ambiguities in phonemes)

1. phonemes are compared to distinct categorical boundaries -this categorical perception means we don't perceive slight differences in phonemes just the ones that them in a different category --not sensitive to variation within phonemic categories ---voice onset time only difference between "p" and "b"

1. surface code (first level of representation in text processing)

1. surface code -a representation of the exact words in the order that they appear --shallow, short lived reps --makes sense (we wouldn't remember exact way something was phrased for long) -typically remember meaning and discard exact phrasing

additional task to understanding letters (second)

2. figuring out what phoneme is associated with a letter or set of letters -this is more difficult -learning association requires being able to consciously separate the speech sounds composing a word/hear them individually -example: how do we know how to pronounce pronouns "ea" combo --this is particularly hard in English --complicated mapping from letters to phrases --it's harder to learn how to spell in English -pinpointing difference between bat and pat

why are perception of phonemes difficult? reason #2

2. phoneme production varies not only between speakers but within an individual -individual phonemes are affected by surrounding phonemes --coarticulation

2. surrounding context (second strategy to overcome ambiguities in phonemes)

2. surrounding context for ambiguous phonemes or missing phonemes

2. textbase representation (second level of representation in text processing)

2. textbase representation -semantically deeper and captures explicitly stated logical and causal relations between the objects and the events in the text -meanings and how they are related to each other -uses a format called a proposition

2. word order (parser deciding what fits where)

2. word order - parser assumes that sentences will be active -people faster at understanding active over passive voice

what makes language processing difficult?

there are ambiguities at each level