PSYC344 Final Exam Notes PDF

Fundamentals of Cognition and Language Organization of Language Hierarchical organization → from the smallest to the largest unit ○ Phoneme ○ Morpheme ○ Word ○ Sentence Rule-based nature → components can be arranged in certain ways, but not othe...

Fundamentals of Cognition and Language Organization of Language Hierarchical organization → from the smallest to the largest unit ○ Phoneme ○ Morpheme ○ Word ○ Sentence Rule-based nature → components can be arranged in certain ways, but not others ○ “My car is over there.” ○ **’There over my car is.’ Phonemes Three different ways to categorize sounds: ○ Place of articulation → where airflow is restricted (p) vs (t) vs (k) ○ Manner of articulation → how the airflow is restricted (s) vs (b) vs (n) ○ Voicing → whether the vocal folds are “buzzing” (f) vs (v) Place, Manner, and Voice All phonemes can be described using combinations of these features ○ (p) = voiceless bilabial stop ○ (z) = voiced alveolar fricative English - 40 different phonemes Some languages have more, some less Morphemes A minimal unit of meaning or grammatical function Word Minimal units of meaning Grammatical function Renewed Re- (“again”) new -ed (past tense) (“recently made”) Tourists Tour (“travel for pleasure”) -s (more than one) -ist (“person who”) Free morphemes → morphemes that can stand by themselves as single words ○ New, tour Bound morphemes → morphemes that can’t stand on their own and typically must be attached to another form ○ Re-, -ist, -ed, -s ○ Affixes (prefixes and suffixes) Free morphemes also fall into categories Lexical morphemes/”content words”: words that carry the content of the messages we convey ○ nouns, verbs, adjectives… Functional morphemes/”function words”: words that serve grammatical functions ○ articles, conjunctions, prepositions… Syntax When we read a sentence, we parse it - assign each word a syntactic role ○ Syntax → : how words are arranged to create well-formed phrases and sentences. “The dog chased the cat” (determiner) (agent noun) (verb) (determiner) (patient noun) “Who is doing what to whom” Recursion → the ability to place a component inside another component of the same type ○ ex. a sentence inside another sentence z = Tom likes beans. y = Susan thinks (x). x = Susan thinks Tom likes beans. x = (y(z)) What effects does recursion have? ○ It allows the expression of very complex concepts.  ○ It removes any upper limit on the length of a sentence. Dave knows Margaret feels Bob believes…Susan thinks Tom likes beans. (a(b(c(d(e(…(y(z))…)))))) Is recursion universal? Perhaps not in Pirahã English sentence with recursion: ”Hand me the nails that Dan bought.” ○ a stack of bowls: two ideas nested within one another ○ (Hand me the nails (Dan bought the nails)) Pirahã equivalent expression without recursion: ○ Hand me the nails. ○ Dan bought those very nails. ○ They are the same Prescriptive vs Descriptive rules Prescriptive rules → a set of “top-down,” arbitrarily assigned rules that govern the “proper” use of language ○ You can’t end a sentence with a proposition Who did you go with? → With whom did you go? Descriptive rules → rules that describe how people actually use language ○ Ending a sentence with a preposition is totally fine! Examples of descriptive rules of English “Each clause can have only one main verb.” My cat Miriel liked ate treats “Verbs go in the middle.” Ate Miriel treats yesterday Hockett’s Design Features Semanticity ○ There are fixed associations between units of language and aspects of the word ○ Language communicates meanings! Arbitrariness ○ No necessary relationship between objects in the world and symbols used to represent them ○ Cat = koschka/gato/chat/neko/mao…lerp? ○ Onomatopoeia? ○ Not that systematic  Discreteness ○ The units of language are separate and distinct from each other rather than being part of a continuous whole ○ Every sound in English belongs to one of about 40 phoneme categories (eg a sound is either /p/ or /b/) Displacement ○ Language can be used to communicate about things that are not present in time and/or space ○ Spatial Displacement: communicating about things/events out of sight of the speaker ○ Temporal Displacement: communicating about things/events in the past/future ○ Degree and scope of displacement varies language to language Productivity ○ Language can be used to say things that have never been said before and yet are understandable to the receiver Duality of patterning ○ Languages have a fixed number of symbols, but potentially infinite number of messages that can be created using those symbols Multiple levels simultaneously – phonemes, words, sentences ○ We “see through” phonemes/word sounds to meaning of speaker– but word sounds can be detected if we concentrate pat/tap/apt Animal Communication Bees communicate: ○ Direction ○ Distance ○ Quality of nectar source Displacement? Duality of patterning? Vervet monkeys (Chlorocebus pygerythrus) ○ Separate leopard/eagle/snake warning calls associated with different behavioral responses Diana monkeys (Cercopithecus diana)  ○ Separate “eagle” and “jaguar” warning calls. Play “eagle” warning call Follow with either real eagle noises, or real jaguar noises Response to “eagle” call is muted following “eagle” warning Response to “jaguar” growl is extreme agitation Ape Language Studies Can language be taught to apes? ○ What kinds of evidence would be convincing? How could we know that an ape understands the symbol that it is using? Kanzi the bonobo (Pan paniscus) ○ Makes different vocalizations in the context of different objects (e.g., juice vs. bananas vs. grapes) Nim Chimpsky the chimpanzee (Pan troglodytes) ○ Never learned to talk; did learn to gesture Washoe the chimpanzee (Pan troglodytes) ○ 1960s ○ Learned gesture language “naturalistically” Learning = use new sign for 14 consecutive days Double-blind vocab tests ○ Generativity? “water” + “bird” = “swan Panbanisha the bonobo (Pan paniscus) ○ Raised for 5 years in an enriched environment ○ “Caregivers communicated to (the apes) with spoken English and visuographic symbols called ‘lexigrams.’” Complex questions, past events, objects out of view Differences between apes and humans ○ As utterances grow longer: In children: utterances become more complex (recursion!) In apes: signs are repeated (eg., e.g., “eat Nim eat Nim,” "banana me eat banana,” “give orange me give eat orange me eat orange give me eat orange give me you”) Humans use words to comment and express intentions; apes use signs as tools to get things  Humans apply grammatical rules consistently; apes apply grammatical rules inconsistently (“more banana” vs “banana more”) Apes interrupt more than humans Evaluate the Evidence Based on the evidence you’ve seen here, do you think great apes (chimps, bonobos, gorillas) are capable of using/understanding language in the way that humans do? ○ If yes: what evidence supports your conclusion? ○ If not: what evidence would you need to see to be convinced? Social Language Apes don’t point. Painting = intended to communicate Joint attention: awareness between individuals that they are paying attention to the same thing ○ Possibly found only in humans ○ Connected with vocab size as toddlers ○ Necessary for agreeing on what words mean! Language Origins Any language abilities present in modern apes likely existed in the common ancestor ~5-8 million years ago. Speech is an adaptation. ○ E.g., our vocal tracts differ from other apes ○ They make us vulnerable to death by choking ○ They are ideally suited to producing speech Two modern human characteristics required to produce speech ○ The right kind of vocal tract Equal distance from the larynx to the top of the throat and from the top of the throat to the mouth opening Rudimentary speech pre-dates the modern physical configuration? (Lieberman, 2000) ○ The ability to control the speech apparatus to produce rapid small changes in air flow  Human ancestors (and apes) lack the neural systems necessary for fine breathing control Therefore, speech does not pre-date modern humans (Homo sapiens) Proto-word hypothesis ○ Word-like units must be present before complex sequences emerge. ○ “Naming insight” Possible proto-word types: ○ Animal calls: “Moo” ○ Exertion noises: “Heave-ho” ○ Lip-smacking: “Yum yum” ○ Greetings: “Hi there” What about gestures? ○ Apes are much better at learning to communicate via gesture than via speech ○ Apes spontaneously communicate with each other via gesture ○ Some researchers posit that the first human language was gestural (Corballis, 1999) Nicaraguan Sign Language (LSN) ○ Homesign: personal communication system invented by a deaf person to communicate with people who do not know sign language ○ 1977: central school for the deaf established ○ “First-wave”: standardized vocabulary, grammatically impoverished ○ “Second-wave”: spontaneous addition of grammatical features The Language Instinct? Nativism: humans are genetically programmed to have a general capacity for language, and particularly aspects of language are also genetically specified ○ Universal Grammar: inborn predisposition to learn certain kinds of linguistic structures and not others ○ Sensitive period: a window of time during which a specific type of learning takes place more easily than during any other time Anti-nativism/empiricism: language is an aspect or by-product of our other cognitive abilities (learning, memory, etc) Speech Perception and Production Empiricism Vs. Nativism Empiricism → children are born as a blank slate Nativism → children have innate comprehension abilities ○ Developed through adaptation and natural selection ○ Visual perception, physical properties, numbers What, if any, language skills are innate? Picture is complicated! We can explain voice, language, and story preferences in infants with learning accounts ○ Don’t need additional innate language mechanism Doesn’t rule out innate accounts entirely Innate Knowledge But learning starts before birth 2-day old infants can discriminate native from non-native language (spoken by the same individual). This is the result of prenatal exposure, not instinct Prenatal Learning Fetuses respond to environmental sounds in the 3rd trimester. The loudest environmental sound is the pregnant person’s voice. ○ High frequency sounds: blocked ○ Low frequency sounds: less affected; transmitted to fetus Babies can learn prosodic features of their native language in utero. ○ Patterns of stress and intonation High-Amplitude Sucking (HAS) Nutritive vs non-nutritive sucking Babies are trained to suck on a pacifier connected to a pressure transducer ○ Preference: how long babies are willing to continue sucking ○ Differentiation: amplitude increases when babies notice new stimulus ○ Dishabituation HAS Experiments Can prosodic information learned prenatally be “remembered” after birth? Pregnant mothers read “Cat in the Hat” 2x/Day for 6 weeks in 3rd trimester. @ 44-60 hours old: testing on familiar vs. unfamiliar story ○ Babies work (suck) harder to hear the familiar story What if we test different speakers? ○ ½ recorded by mom; ½ recorded by different female Infant Speech Perception Phonemes → the basic sounds of a language ○ Some contrasts are meaningful (p/b) and some aren’t (p h/p) (in English!) Categorical perception: signals with different physical properties are placed in the same ‘category’ Infants are born with the ability to detect most phonological contrasts ○ Including contrasts that don’t exist in their native languages ○ Any contrast that is important can be preserved Innate broad categories: help make languages “learnable” by infants” How? ○ Perhaps by developing “prototype” representations. E.g., English and Swedish “ee” are pronounced differently. English infants respond to minor deviations from English “ee,” but not equally small deviations from Swedish “ee.” Speech Segmentation Segmentation starts around 6-7.5 months old Conditioned head turn procedure ○ Training: infants are exposed to a word ○ Test: infants hear familiar word or unfamiliar control word in sentences ○ Young infants tend to prefer familiar stimuli → longer looking times for familiar words 6 months old: can’t segment speech ○ No preference for sentences containing familiar word 7.5 months old: can segment speech ○ Preference for sentences containing familiar word Suggests that segmentation ability is learned, not innate  Precursors → pre-existing skills that infants draw on to develop segmentation abilities ○ Prosodic cues ○ Phonotactic constraints Prosody: patterns of stress and intonation Prosodic Bootstrapping: infants use prosody to identify important units (syllables and words). ○ Prosodic features correlate with word boundaries ○ Infants can use prosodic cues to discriminate between native and non-native languages ○ Prosodic patterns are discriminated against at 2 months old (or younger). Metrical Segmentation Strategy: alternation of strong and weak stressed syllables as a cue for word segmentation (type of prosodic bootstrapping) ○ Most English words use trochaic stress: COOKie, BAby, BOTtle ○ Babies could ID beginnings of words by looking for stressed syllables ○ 5 months old: can distinguish between trochaic and iambic stress patterns Phonotactic constraints: language-specific constraints that govern how the sounds of a language may be combined to form words or syllables ○ /gd/ at beginning: bad in English, fine in Polish ○ /spl/: can begin words in English but can’t end them Noticing where different phoneme combos are and aren’t allowed can help segment speech ○ bigdog /= bi|gdog Sensitivity emerges around 7-9 months ○ Children don’t prefer native language over prosodically similar languages until 9 months Statistical Learning: infants notice patterns in speech stimuli and analyze those patterns to identify subcomponents Transitional probability: the probability that a particular unit will occur, given the previous occurrence of another particular unit ○ Pretty: Probability of / ti/ after / pri/ in Infant-Directed Speech ~ 80% ○ Between-word transitions are lower probability than within-word transitions Maybe syllables that co-occur frequently make a word Are infants able to learn “words” from statistical information?  ○ Training phase: Infants listen to extended sequences of nonsense syllables lum jik pel rud neb jik pel lum neb jik pel neb Pel always follows jik; jik always precedes pel: transition probability = 100% rud follows pel 30% of the time; pel precedes rud 30% of the time: transition probability = 30% Stimuli are synthesized, so no prosodic cues. Test phase: present ‘jikpel’ (word) or ‘pelrud’ (nonword) ○ Infants respond to ‘words’ differently than ‘nonwords’ Infants are able to learn “words” from statistical information! Infant Speech Perception Words in isolation? Yes Prosody? Yes Statistical regularities? Yes Children seem to use whatever info is available to solve the segmentation problem Articulation Articulation perturbs air flow to create different patterns of sound waves. Articulatory phonology theory: ○ Articulation consists of contrastive gestures ○ Each contrastive gesture creates a noticeable change in the speech signal (pattern of sound waves) Articulatory phonology theory (cont.): ○ Speech planning creates a gestural score that tells the articulators how to move Move a particular set of articulators Toward a location in the vocal tract where a constriction occurs With a specific degree of constriction (place/manner/voice) Occurring in a characteristic dynamic manner Speech production involves co-articulation: ○ The gestures for adjacent phonemes overlap in time ○ Gestures are influenced by preceding and following phonemes. E.g., “pool” vs. “pan”  Perceptual invariance: perceiving sounds that have variable acoustic manifestations as members of the same sound category ○ Co-articulation E.g., / di/ vs. / du/ Speech consists of steady-state, stable vibration patterns, and rapid changes in energy Coarticulation: produces redundancy in the speech signal. ○ Individual segments provide clues about preceding and following segments. E.g., “silent center vowels” “bag” ➔ “b_g” ➔ perceived as “bag” (rather than “b_g”, “bog”, or “bug” E.g., “Franken-words” ○ “jo” + “b” ➔ “jog” if “jo” came from (intact) “jog” Categorical Perception We mentally represent sounds as abstract categories ○ Phoneme: category of sounds that share the same function; smallest unit of sound that changes the meaning of a word /t/ vs /d/ -> tan vs Dan but NOT [t] vs [t h ] (in English) Continuous changes in a stimulus are perceived as having a sharp break between discrete categories How does categorization interact with perception? ○ Speech gestures are messy and continuous ○ We impose structure on speech – maybe this influences how we perceive sounds Forced Choice Identification Task: subjects are required to categorize stimuli into one of two categories ○ Voice Onset Time (VOT): continuous variable indexing when the vocal folds begin vibrating relative to the release of the articulators IF Incremental perception: gradual shift from /ba/ to /pa/ as VOT increases But actually: sudden changes from /ba/ to /pa/ as VOT increases ○ Evidence for categorical perception! A: indicate whether you hear /ba/ or /pa/  B: indicate whether you hear “beach” or “peach” Seems like “real” perception is much more continuous! Other Aids To Speech Perception Context cues: using information about a sound’s acoustic context to infer what sound you’re hearing ○ especially important for word identification! ○ Ganong Effect: listeners perceive the same ambiguous sound differently depending on what word its embedded in /g-k/ perceived as /g/ in ”_ift” but as /k/ in “_iss” ○ Phoneme Restoration Effect: when a predictable phoneme within a word is replaced with a non-speech sound, people will “hear” both Motor Theory Articulatory phonology theory: ○ Articulation consists of contrastive gestures ○ Each contrastive gesture creates a noticeable change in the speech signal (pattern of sound waves) Speech perception = perceiving a speaker’s intended gestures (Liberman) or speaker’s actual gestures (Fowler) ○ Register acoustic signal ○ Determine gestures that produced the signal ○ Deduce syllables/words from gestures McGurk Effect: non-acoustic information affects speech perception. ○ Speech perception systems uses both visual and auditory info when perceiving speech ○ Auditory AND visual information can help figure out what set of gestures created the auditory signal McGurk Effect only tells us that gestural info informs speech perception ○ Need stronger evidence to show that gestural info is necessary for speech perception Which comes first: perception or production? ○ 5-7 months: babbling stage – development of “mental categories” of sounds?  ○ Babbling is physical practice: immobilizing babies’ tongues will have no effect on their ability to distinguish phoneme contrasts ○ Babbling is representation-building: immobilizing babies’ tongues will impede their ability to distinguish phoneme contrasts At 6 months, English-speaking infants can usually distinguish between /d̪/ and /ɖ/ ○ Sounds have different places of articulation – if articulatory gestures are important for building sound representations, immobilizing tongue will impede sound discrimination ○ Two teethers – one blocks tongue movement, one doesn’t ○ Babies with blocked tongues couldn’t distinguish sounds! Three core cognitive processes: ○ Conceptualization – thinking of something to say ○ Formulation – figuring out a good way to express your idea ○ Articulation – moving muscles to make a sound wave Lexical concepts: concepts for which your language has a specific word. Lemma: an abstract mental representation that incorporates semantic (meaning) and syntactic (combinatory) information. ○ Does NOT include sound information! Syntax: how words are arranged to create well-formed phrases and sentences ○ Determining relationships: who is doing what to whom ○ The same messages can be conveyed using different syntax Dr. Milburn taught the speech production lecture. The speech production lecture was taught by Dr. Milburn. It was Dr. Milburn who taught the speech production lecture. It was the speech production lecture that Dr. Milburn taught. Speech production is hard! ○ Maintaining balance between planning, producing, and comprehending – multitasking! ○ Pressure to minimize pauses in conversation Accessibility: speakers produce whatever word or structure is most accessible at the time ○ happens with both words and syntactic structures Syntactic Priming: speakers are more likely to use a particular syntactic structure if they’ve recently been exposed to that syntactic structure Bock (1986)  ○ Participants were exposed to prime sentences that were either Prepositional or Double Object ○ Participants then saw a picture and had to describe it ○ Participants’ syntax choices were biased by the prime! Morphemes: language “atoms;” smallest unit of language that carries meaning. ○ E.g., eat vs. eats vs. eating vs. ate ○ E.g., [elephant] + [s] ○ Morphological specification (word form) changes depending on what role the lemma plays in the sentence Phonological gestural score: a detailed map of ○ phonemes needed for an utterance ○ metrical information (stress) Used to plan specific motor movements ○ E.g. “banana” ➔ σ σ’ σ ○ vs. “Panama” ➔ σ’ σ σ How Do We Know? Word exchange errors: Word gets replaced with one occurring later in the sentence ○ “He left it and forgot it behind.” ○ Can occur across many words, but usually within same clause Sound exchange errors/”Spoonerisms”: speech errors in which sounds get exchanged between words ○ The Lord is a shoving leopard (loving shepherd) ○ “It is 94 degrees right now and it is noon so we’re approaching a peak of 105 in a few hours.” ○ Usually occurs over adjacent or nearby words Suggests that we plan concepts/words farther ahead than sounds Lexical Bias Effect: speech-based sound errors tend to result in real words instead of non-words Investigating Sound Exchanges in the Lab “I’m not a pheasant-plucker, → ph-pl I’m a pheasant-plucker’s son, → ph-pl and I’m only plucking pheasants → pl-ph  til the pheasant-plucker comes.” → ph-pl SLIP paradigm: Spoonerisms of Laboratory-Induced Predisposition Subjects read pairs of words with a certain sound pattern: ○ fat beer ○ fun bed ○ far base Then are prompted with a target pair: ○ big feet Creates a scenario that increases the likelihood that a speech error will occur ○ Slips more likely when errors produce pairs of real words How Do We Know? Tip-of-the-Tongue (TOT): you know you know the word, but you can’t retrieve and pronounce it ○ What part of the production system is breaking down? Tip-of-the-Tongue A word or sentence that reads the same backward and forward, such as “Madam, I’m Adam” Islands off the coast of Ecuador that Darwin visited to study unique species of animal life Order of lower mammals including kangaroos and opossums which carry their young in an abdominal pouch Word meaning favoritism in hiring based on family relationships People who make maps People who explore caves, either for hobby or for sport Formal term for the collection and study of postage stamps During TOT experiences, people ○ can report many characteristics of the target word ○ accurately predict whether they will come up with the target word soon ○ are more accurate about the beginning and end phonemes than the middle ○ resolve about 40% within a few seconds to a few minutes If you’re stuck… ○ Ask yourself questions about the target word!  What does this mean? ○ TOTs reflect the strength of the relationship between conceptual, lemma, and phonological levels of representation ○ Strong relationship (more common word) = less likelihood of TOT Feed-Forward Vs. Interaction (A) Model with unidirectional links ○ Message level → Lexical activation → Sound activation (B) Model with bidirectional links ○ Message level → Lexical activation → Sound activation Evidence for Interactive Models Mixed errors → speech errors that involve similarities of both sound and meaning ○ “Pushing out from the inside, watching the outside cylinder…I mean watching the outside silhouette” – my ceramics teacher Emma ○ “They were given a temporary immigration status called Humanitarian Parole-Parole- allowing them to live and work in the US.” – NPR Error Monitoring We often catch our speech errors in the middle of making them or before they’re made – evidence for a self-monitoring system? ○ Real-word errors are hard to catch Test by using SLIP technique to elicit potentially embarrassing errors ○ kook tin ○ coop tilt ○ coon tips ○ tool kits (vs tool karts) More likely to make errors in non-embarrassing situations! Learning Words (EXAM 2) Learning Word Meanings Children start to speak - 12 months They leave high school knowing - 50 to 60,000 words  It takes about 18 months to learn 50 words (productive vocabulary); then infants have a word spurt. Duality of Patterning: meaningful messages are made up of distinct smaller meaningful units. Point and say (Clark, 2002; Skinner 1957) ○ Many early words refer to concrete objects (Brown, 1957) ○ But other words like greetings (Hi!) imperatives (NO!), substances (milk), and processes (think), do not. Poverty of the Stimulus: words do not contain enough information to uniquely specify meaning Other problems for point-n-say: ○ People do not refer to on-going actions ○ And meaning depends on perspective of speaker, not child: E.g. chase vs. flee ○ And: blind children learn word meanings at about the same rate as sighted children Lexical representation: information stored in LTM about sound and meaning properties of words,including some info about syntactic capabilities Switch Task: word-mapping task for infants ○ Habituation: object is paired with label ○ Test: label is paired with either familiar or “switched” objects 14-month-olds don’t seem to notice the switch if the sounds are very similar! Why? Babies only store as much info as they need to differentiate words in LTM. ○ If your vocabulary is only a dozen words, you don’t need much detailed sound information ○ Mommy vs daddy ○ As we grow, we refine our lexical representations to distinguish between similar-sounding words big/dig, bean/dean Graf Estes et al., 2007 ○ 2.5 minutes of artificial speech ○ Switch paradigm: half “words,” half “nonwords”  ○ 17-month-old babies learned word-pic associations - but only for “words”! Heuristics and Biases Whole-object bias: assumption that a new word refers to a whole thing and not parts, color, surface, etc. ○ Babies can parse physical objects before linguistic utterances ○ But also are biased towards assigning linguistic labels to whole objects Mutual Exclusivity Bias: no two words in a language have exactly the same meaning ○ New labels = new objects ○ Test: familiar and unfamiliar objects. If speaker uses unfamiliar word, child should assign to unfamiliar object Principle of Contrast: if two labels do apply to an object, they must have different meanings ○ Salient parts: 1st term = whole object, 2nd term = salient part or property “See the bunny? Those are his ears.” ○ Speaker choices: highlight differences in meaning “That’s a dog” vs. “That’s a mammal” Iconicity Arbitrariness → No necessary relationship between objects in the world and symbols used to represent them Ideophones → words in which sounds convey sensory information about size, texture, motion…. ○ English: zig-zag, glimmer, twinkle, higgledy-piggledy, wishy washy ○ Siwu: vɛlɛvɛlɛ ‘a dizzy, giddy feeling in the body’ Iconicity → form a symbol resembles the meaning of the symbol ○ Common in signed languages Learning Categories Children must learn to generalize ○ Language is useful because it generalizes to new situations The generalization problem is a categorization problem Categorization challenges: ○ One object can belong to many categories  ○ Some categories don’t get their own words Children appear to assume that new labels refer to basic level categories (rather than subordinate or superordinate ) ○ Chesterfield - chair - furniture ○ Ford Mustang - car - vehicle ○ Thoroughbred - horse - hooved ungulate Gavagai! = whole thing + basic level category Basic-level categories are great for capturing generalities ○ Developing basic-level categories first is more useful - but can be difficult Overextension: mapping new words into categories that are too general Underextension: mapping new words into categories that are too specific. Social Interaction Hockett’s Design Features ○ Semanticity: there are fixed associations between units of language and aspects of the world ○ Arbitrariness: no necessary relationship between objects in the world and symbols used to represent them ○ We collectively agree that particular units of language refer to particular objects or concepts in the world! ○ Linguistic labels are social constructs! Theory of Mind: Knowledge that people think, perceive, and have private experiences ○ Social nature of consciousness: we understand that other people have private thoughts because we observe them in yourself and in others ○ Development of sharing, teasing, persuasion, empathy, lying… Different ToM skills develop along different time scales ○ …but children are learning words and categories BEFORE ToM is fully developed! Babies learn words before they fully develop ToM ○ 15-20 month-old infants know that language is social Children track speakers’ mental states, abilities, attention → infer which objects are being labeled ○ Children understand that speakers can refer to objects that speakers can’t see Children give greater weight to ”reliable” speakers  ○ Adults > children ○ ”Knowledgeable” adults > “un-knowledgeable” adults Nativism VS Empiricism Genetically Guided Learning Hypothesis: children have innate word categories (noun vs. verb) which they populate with words from their native language ○ Paired with “try noun meaning first” strategy “Gavagai” is not an action → pair with salient perceptual experience like an object ○ Doesn’t solve problem, because objects belong to multiple categories simultaneously ○ Also, doesn’t rule out parts that are nouns ○ Possible solution: keep track of multiple episodes General-purpose learning and memory skills ○ Children have a general ability to remember linguistically-conveyed info Adults and children have very good memories for arbitrary facts conveyed by language ○ See an object, either hear “It’s a koba” or “My uncle gave it to me” ○ 1 month later: children remember linguistically-conveyed info ○ But if stickers are used: they don’t remember which object had a sticker Learning new words uses same mental processes as learning new facts conveyed by language → we don’t need a special word-learning mechanism Infant-Directed Speech Exaggerated articulation and prosody; higher and more variable pitch ○ Helps mark boundaries between important words and phrases Infants prefer IDS Children of IDS-speaking mothers have better phoneme discrimination abilities @ 6 months Topic words appear in prominent positions and are marked prosodically Not all cultures have IDS (e.g., Quiche Mayan); but it helps where it is present Experience Matters 30 million word gap between children of white-collar & impoverished parents Within SES groups, maternal education affects the amount of infant-directed speech.  Some infants are better at recognizing familiar words than others; as indicated by gaze Performance at 6 months predicts performance at 24 months Individuals differences largely reflect exposure/practice effects Recognizing Words Words Words are packages of ○ Form (phonology/orthography), AND ○ Meaning (semantics) How Are Words Organized In the Mind? Mental lexicon: information about the meanings, pronunciations, syntactic capabilities, etc of words store in long-term memory ○ How words are activated, stored, processed, & retrieved by individual comprehenders ○ Lexical access: the process of identifying words and recovering word-related information from long-term memory/the mental lexicon How much overlap is there in the words you thought of? What kinds of semantic (meaningful) relationships did you notice? ○ part-whole relationships (bicycle-wheel) ○ conceptual relationships (apple-banana) ○ situational relationships (ambulance-hospital) ○ synonyms/antonyms (black-dark; black-white) ○ co occurrence relationships (apple-pie) What can you conclude about how words are organized in the lexicon? Lexical Decision Task (LDT) A way to gain insight into the mental lexicon ○ yes/no: is (stimulus) a real word? ○ Reaction Time (RT) is related to the amount of a time it takes to “look up” the word in the lexicon (lexical access)  Let’s try it: raise your right hand if the string IS a real word, and raise your left hand if the string is NOT a real word Semantic Networks Nodes ○ Concepts with meanings captured by the semantic network Links ○ Different relationships between concepts Work meaning = pattern of activity in a network Spreading Activation → activity at one node causes activation at other nodes via connecting links ○ Fast, outside conscious control ○ Decreases the further it travels ○ Decay: activation fades over time Evidence: semantic priming ○ Faster response to a target stimulus because a related stimulus appears first ○ Mediated priming Lion → tiger → stripes Effects To Explain Facilitation: processes that make it easier for word recognition to be completed ○ Semantic priming! Inhibition: processes that result in word recognition becoming more difficult ○ Similar-sounding words compete with each other for activation! ○ Neighborhood density: the number of different words that can be made by adding, deleting, or changing one phoneme of a word High: feet Low: fruit Modeling Word Recognition Excitatory connections: increase the activity of linked nodes Inhibitory connections: decrease the activity of linked nodes Identifying phonemes causes activation to flow to linked words  ○ BUT ALSO: incompatible words are inhibited – explains neighborhood effects What is a Word Representation? What is in a word representation? ○ Localist view : discrete unit of meaning, sound info, and syntactic information ○ Distributed view : bundles of features, with no central node Lexical Ambiguity Words have multiple meanings! ○ Bank, sheet, bug, gem… Is this a bug (!) or a feature? Exclusive vs. exhaustive access: do we activate all the meanings of an ambiguous word, or just the contextually-appropriate one? Swinney (1979) Cross-modal Lexical Decision Task ○ Subjects heard a biased sentence: “Rumor had it that for years, the government building had been plagued with problems. The Man was not surprised when he found several spiders, roaches, and other bugs in the corner of his room.” Subjects saw: Contextually-related word: ant Contextually-inappropriate word: spy Unrelated word: sew Nonword: luf Measured reaction time (RT) to each type of word ***Words could appear on the screen either IMMEDIATELY when the target word was heard, or 200 ms later RT for words related to either meaning of the ambiguous word (ant or spy) were shorter than for unrelated words (sew), only when presented at the same time as the ambiguous word When the Lexical Decision Task was presented 200 ms after the onset of the ambiguous word, RT was faster only for the contextually appropriate word (ant) ○ Even in a strongly biased context, both meanings of an ambiguous word are briefly activated Ambiguous words have dominant and subordinate meanings  ○ Simultaneously activated meanings compete – like neighborhood studies ○ When context favors subordinate meanings, reading is slower ○ When meanings are more balanced, reading is unhindered Spoken Words “Imagine reading this page through a two-letter aperture as the text scrolled past, without spaces separating words, at a variable rate you could not control.” (Magnuson et al., 2007) Lexical access: the process of identifying words and recovering word-related information from long-term memory. Lexical access is fast: ○ Speech: ~5 syllables/second ○ fast shadowers lag ≈ 250 ms ≈ 1 syllable. ○ Errors are not random: they fit the semantic and syntactic context. ○ Lexical access is incremental – higher order relationships between words are computed before phrase/sentence boundaries are encountered Cohort model: multiple cohort competitors become active immediately and are gradually winnowed down to a single candidate as additional acoustic information is taken in. ○ Uniqueness point: enough information in the speech stream for a word to be differentiated from all competitors Activation ○ Auditory stimulus activates cohort of matching word representations ○ Autonomous Selection ○ Selecting best match from activated representations for acoustic stimulus ○ Words that fit context have advantage Integration ○ Word is incorporated into the utterance Reading Reading involves ○ Linguistic/comprehension processes ○ Eye-movement control processes ○ Coordination of 1 & 2  Readers fixate most of the words in a text ○ ~90% of content words ○ Fixations last ~200-500 ms Some words are skipped ○ Words that are really predictable from context ○ Highly frequent function words (the, a, and, but) Most saccades are progressive, but some are regressive ○ Regressive saccades help resolve comprehension problems Methods in Reading Studies Eye tracking: position of eye is tracked using sensitive camera ○ Camera “samples” eye position usually every millisecond ○ What’s measured: how fast eyes move, where readers pause, where readers look back, which words readers skip… Eye Movement Control Reading is a behavior ○ Direct eye movements along a piece of text → develop a mental representation of that text Fixation: maintaining the gaze on a single location ○ ~200-250 ms Saccade: short, rapid, movements between fixations ○ ~1/50th of a second Saccadic Suppression: no visual information is extracted during a saccade ○ Brain “fills in the blanks Fovea: the area in the center of vision where visual acuity is highest ○ Extends from center of vision 1º in all directions ○ Saccades bring fixated objects into the fovea Parafovea: area circumscribing the fovea where visual acuity is less good ○ Extends about 6º in all directions out from the fovea ○ Near-parafovea: area close to the fovea; visual acuity is good ○ Far-parafovea: area far from the fovea; visual acuity is poor The Perceptual Span  Perceptual span: the region of text from which readers extract useful visual information during reading about 4 characters to the left of fixation to 12-15 characters to the right ○ Because of English writing system ○ When skilled adults read, they directly fixate the… Optimal viewing position: fixation point just to the left of the middle of the word ○ Lexical processing is fastest when saccades land here ○ If saccades land closer or farther to the edge of a word, processing is slower OVP places the majority of the word in the visual region with the best acuity Low acuity = no differentiation between letters (far parafovea) OK acuity = some differentiation between letters (near parafovea) Best acuity = best differentiation between letters (fovea) Words must be directly focused to be identified But sometimes words are skipped! Why? ○ Predictable words are skipped ○ Short words are skipped Skipped words are not identified via clear images in the parafovea Degraded peripheral image + top-down knowledge = good guess Learning to Read Reading emerges only with special instruction in the vast majority of children Alphabetic script: letters/groups of letters correspond to phonemes Precursors to reading an alphabetic script: ○ phonemic awareness ○ the alphabetic principle Phonemic awareness: spoken words can be broken down into subparts Differences can be measured in children who haven’t learned to read → predicts reading success 2/3 years later Test of phonemic awareness ○ Elision: “say cat without saying /k/” ○ Sound categorization: “Pin, fun, gun, bun – which one doesn’t belong?” ○ Blending: “Put fi and sh together” Children’s test scores are correlated with reading success later – but interventions are very effective!  Alphabetic principle: sets of letters go with certain speech sounds Deep vs. shallow orthography ○ Shallow: one-to-one mapping between letters and speech sounds (Spanish, Russian) ○ Deep: multiple mappings between letters and sounds (English) English orthography is a compromise between conveying phonological information and preserving morphological information ○ Sign, signal, signature Dual-Route Model of Reading How are regular words read? How are exception words read? How do we read words that we’ve never seen before Two separate ways that readers can use visual input to access entries in the mental lexicon 1. Phonological route: “sound out” the word by converting letters into phonemes a. Use grapheme-to-phoneme correspondence rules on a letter-by-letter basis b. Great for regular words c. Evidence: monitor list of words for categories like food or clothes i. “false alarm” response for meet and sute – not category members but are pronounced the same 2. Direct route: bypass phonology and access lexicon directly a. Orthographically irregular/exception words: have, pint, Colonel b. Exception words accessed via visual code → pronunciation is available following lexical access Why not just use the direct route for all words? ○ Sometimes we encounter totally unfamiliar words Mave, slood ○ Sometimes we encounter words that we’ve heard but not seen How are regular words read? ○ Phonological route How are exception words read? ○ Direct route How do we read words that we’ve never seen before?  Phonological route Sentence Comprehension Syntax When we read, we parse it - use cues to discover how words in sentences relate to each other ○ Syntax → how words are arranged to create well-formed phrases and sentences The meaning of a sentence is not just the meaning of words within the sentence ○ “The dog chased the cat.” ○ “The dog was chased by the cat.” Incrementality → building meaning “on the fly” as new information becomes available ○ Fast shadowing: lag = 20 ms = 1 syllable; errors are not random: they fit the semantic and syntactic context. Language comprehension systems must suppress inappropriate interpretations as well as calculate correct interpretations - this is hard! ○ How are we able to parse sentences correctly most of the time? Models of Parsing Garden Path Model Constraint-Based Model Good Enough Model Global VS Local Ambiguity Global ambiguity → at least two distinct syntactic parser ○ Examples Kids make nutritious snacks Dealers will hear car talk at noon British left waffles on Falkland Islands Complaints about NBA referees growing ugly Police begin campaign to run down jaywalkers Grandmother of eight makes hole in one Enraged cow injures farmer with ax  Sisters reunited after eighteen years in checkout counter Local ambiguity → contain a sequence of words that can be interpreted in more than one way, but the sentence as a whole has only one correct parse ○ “Garden path” sentences ○ Disambiguation point: the first point at which only the correct interpretation is consistent with the unfolding sentence ○ Measure reading times on disambiguating regions Longer reading times = processing difficulty Who cares? Why would wwe want to study sentences like these? They almost never occur in the real world But we DO encounter ambiguities in language all the time ○ When part of a sentence is ambiguous, how do we decide which meaning to choose? Modularity in Cognition Modularity → cognition of lots of small, separable, automatic systems, each of which is responsible for performing a certain task ○ Modules process input and spit out output ○ Outputs of modules are incorporated by a central processor 1. Informationally encapsulated a. Systems can only access information given in inputs (plus any info already in system) b. Restriction of flow of info into the system 2. Informationally isolated (opaque/inaccessible) a. Outputs are available to central monitoring, internal processes are not b. Only final products are consciously available c. Restriction of flow of info out of the system 3. Domain-specific a. Processes info about a narrowly circumscribed class of objects and properties 4. Fast a. Pretty obvious! b. Fodor: fast cognitive process ≤ 500 ms  5. Automatic a. Not under conscious control b. Switched on by relevant stimuli; operations run to completion Muller-Lyer Illusion ○ Even when you know the lines are the same length, you still see the illusion - perceptual system is informationally encapsulated ○ Evidence for a modular visual system? Garden Path Model Syntactic parser is modular Syntactic parser uses ONLY syntactic information to decide how to construct and interpret a sentence ○ Locally ambiguous sentences “lead you down the garden path” when you build an incorrect syntactic structure Stage 1: identify syntactic categories and build initial structure Stage 2: assess outcome against context, semantic plausibility, real-world knowledge Revise if necessary Assumptions ○ Incrementality → word-by-word parsing ○ Serial processing → one structure at a time ○ Simplicity → no unnecessary structure Late Closure → no unnecessary structure; build the least complex representation. “The burglar blew up the safe with the rusty lock” Strengths ○ Explain how syntactic structures are built ○ Explains why processor is very fast ○ Assumes on-line interpretation - words are processed as they are input ○ Makes strong predictions about ambiguity resolution ○ Extends to languages other than English Weaknesses ○ Underestimates influence of lexical frequencies, plausibility, context, on initial parses Constraint-Based Approach Constraint-Based Approach → multiple interpretations of an ambiguous structure are simultaneously evaluated against a broad range of information sources (or constraints) that can affect the parser’s early decisions ○ Parsers build or activate all licensed structures simultaneously (parallel processing). ○ Structures compete for activation ○ Interpretations are ranked. Most likely structure gets highest activation The parser uses many sources of information to compute likelihood ○ Ambiguity = syntactic preferences + semantic preferences + word frequency + …. “The horse raced past the barn fell.” “Raced” more likely to be a main verb than a past participle Horses are likely to race People tend to build main verb phrases much more often than subordinate clauses Story Context Effects ○ E.g., The burglar blew up the safe with the rusty lock Why is this difficult? ○ Garden-Path Model: complexity/late closure ○ Constraint-based Model: pre-supposition violations ○ The burglar was planning his next job. He knew that the warehouse had two safes. Although one was brand new from the factory, the other had been sitting out in the rain for ten years. The burglar blew up the safe with a rusty lock. ○ GP: sentence should be hard because parser will need to reanalyze syntax ○ CB: sentence should be easy because parser can use contect, even if it means building a more complicated structure This sentence is easy to process in this story context Semantic effects: information about what words mean can be used to predict how a sentence should be parsed The man examined by the lawyer turned out to be unreliable ○ “Man” could be examining something OR being examined  ○ Deciding between parses → slower processing The evidence examined by the lawyer turned out to be unreliable ○ “Evidence” can ONLY be examined! ○ Only one reasonable parse → faster processing The syntax in these two sentences is identical but our semantic knowledge of animate vs. inanimate (alive vs. unalive) nouns has an early effect on parsing Visual Context Effects: syntactic choices can be influence by info outside the language system - from visual context ○ Put the apple on the towel in the box. ○ 1. Initial parse: the apple should go on the towel ○ 2. Reanalysis ○ 3. Final parse: the apple is on the towel On the towel is ambiguous between a goal (of put) and a modifier (of apple) ○ Visual context can support the modifier interpretation (the apple is on the towel already; we’re not putting it on the towel) Visual Context Effects Altmann & Kamide (1999) Visual World Paradigm ○ “The boy will eat the cake” ○ “The boy will move the cake” Participants looked toward the “cake” in the “eat” condition 87 ms BEFORE hearing the word “cake” But 127 ms AFTER hearing the word “cake” in the “move” condition. ○ Place the apple on the towel in the box (temporarily ambiguous sentence) ○ Place the apple that’s on the towel in the box (unambiguous sentence) Good Enough Parsing Sometimes we don’t need parsing! Especially when it’s redundant with lexical information ○ E.g., mouse, cheese, eat GEP predicts parsing errors when lexical semantics contradict structural information ○ The mouse was eaten by the cheese  ○ Relying on syntax: the cheese ate the mouse = BAD ○ Relying on semantics: the mouse ate the cheese = OK Do we build a full syntactic model of these sentences? Or do we say “Eh, good enough” and rely on lexical semantics? “While the hunter was stalking the deer in the zoo drank from the puddle.” Was the hunter stalking the deer? Most participants say yes! ○ Why comprehenders set thresholds from comprehension Stakes of comprehension situation determine how complex of a syntactic frame is built Ambiguities remain unresolved. Which Model is Right All of them, alas All of parsing models explain some of the ways people behave when parsing sentences ○ No model explains ALL of the ways people behave Why have multiple models ○ Explain the same parsing behavior with different underlying assumptions ○ Allow us to develop different theories about parsing mechanisms Top-Down Processing When our expectations, knowledge, or experience with the world influence what we sense; information processing guided by higher-level mental processes Top-down processing helps the brain assign meaning to sensory messages, including information based on prior knowledge, learning, and goals Language: use world knowledge, language patterns, environmental context, to resolve linguistic ambiguity ○ Choosing among multiple options AND anticipating upcoming sounds, words, and syntactic structures (Altmann & Kamide, 1999) Prediction Kamide, ALtmann, & Haywood (2003) Comprehenders use world knowledge to predict upcoming words ○ The man will ride the motorbike ○ The man will taste the beer  ○ The girl will ride the carousel ○ The girl will taste the candy Listeners combined subjects and verbs to predict direct objects based on WK! Do comprehenders predict general categories of words or specific words? ○ Use brain waves to examine brain activity while people comprehend language Event-Related Potential (ERP) : change in brain response that is the result of a specific event ○ N400: associated with contextual implausibility ○ P600: associated with syntactic violations DeLong et. al, 2005 ○ The day was breezy, so the boy went outside to fly a kite ○ The day was breezy, so the boy went outside to fly an airplane Comprehenders can generate a prediction for a specific word based on context + article (a/an) N400 after low predictability article = prediction for specific words! Discourse Comprehension (End of Exam 2) Discourse Discourse happens when more than one sentence go together Focusing on narrative texts The goal of discourse processing is to build a representation of the situation that the texts describes ○ “Reading between the lines” → inferences Four main processes 1. ID content of text 2. Referential processes: connect words in text with ideas, objects events that words refer to 3. Cohesion/coherence processes: connecting different pieces of text to each other 4. Discourse representation/mental model: a representation of what of what the text is about Construction-Integration Model  The system builds three related representations: ○ Surface form: words in text + syntactic relations; phrase-structure tree ○ Textbase: propositions represented by the surface form Doesn’t include exact working; can include info that wasn’t mentioned ○ Situation model: ideas and events of the text Propositions: Capture action, states, and changes of state ○ A verb and its arguments (adjoining role-players) What actors and objects are involved in action “The customer wrote the company a complaint” Verb (agent, recipient, direct object) Wrote (customer, company, complaint) ○ The smallest unit of language that can be assigned to a truth value True or false: wrote the company? True or false: the customer wrote the company? Propositions are psychologically real. Proposition-membership, rather than verbatim form, drives language behavior. The geese crossed the horizon as the wind shuffled the clouds. ○ Task → remember word “wind” from second proposition ○ Cue → ‘horizon’ = poor recall ○ Cue →’clouds’ = good recall Situation Models: A representation of what the text is about ○ The mental model combines verbatim information from the text and general world knowledge (inferences - info that isn’t directly stated). ○ When general world knowledge is absent, the representation of the text can be incoherent. Why can’t we just stop at propositions? Propositions, Situation Models persist; Verbatim (surface) form does not. Causation Argument overlap: look for arguments in active props that represent the same concepts ○ “John drove his car through the valley and over the bridge. A goose swam under the bridge.” ○ Allows one proposition to connect to another  Cohesion: different parts of a text go together because they talk about the same thing ○ “It was a sunny day on the hill. Hills are higher than valleys. The central valley contains a number of interesting museums. Museums often have antique weapons. Weapons of mass destruction are a threat to security. Linus’ security blanket needs washing.” Modeling causation ○ Casual coherence Is this statement a cause of events that occur later in the story? Is this statement a consequence of events that occurred earlier in the story? ○ Determine whether something is a cause using the ‘Necessity in the circumstance’ heuristic A causes B if, in the circumstances of the story, B would not have occurred if A had not occurred. Stories can be described as networks of causal connections Discovering causal connections allows comprehenders to answer why questions Propositions on the central causal chain are remembered better than other propositions. Construction Integration a. Three turtles rested on a log and a fish swam beneath them b. Three turtles rested on a log and a fish swam beneath it i. Three turtles rested on a log and a fish swam beneath it/them c. Three turtles rested beside a log and a fish swam beneath it d. Three turtles rested beside a log and a fish swam beneath them Task: pick exact sentence you read from pair ○ Very inaccurate when picking between a) and b) - map onto same situation model ○ Much more accurate between c) and d) - map onto different situation models General World Knowledge Knowledge affects ○ How stories are structured  ○ The info we need to make inferences to keep stories coherent ○ The situation models we build to represent what stories are about 1. How stories are structured a. Story Grammars i. Bartlett, 1932 “War of the Ghosts” 1. English participants read a story from Canadian indigenous folklore 2. Repeated reproduction: participants tried to recall the story at longer and longer intervals 3. As time passed, participants transformed the story to make it consistent with their own culture 2. The info we need to make inferences to keep stories coherent a. Schemas: structured, pre-existing knowledge about a particu;at domain in the world i. Supports inference generation by allowing us to “fill in the gaps” b. Susan and Bill went into the restaurant, sat down, and ordered lunch c. Susan dropped her fork… They paid the check and left 3. The situation models we build to represent what stories are about a. If inappropriate information is out of focus, comprehenders may not notice. i. How many animals of each type did Moses take on the Ark? ii. A plane crashed on the border between two countries. Where should we bury the survivors? b. Moses Illusion → situation models “overwhelm” meanings of single anomalous words. c. But: Constructing the sentence to syntactically focus attention on the “oddball” decreases the Moses Illusion. How do we use world knowledge to decide whether a sentence describes something true? ○ Distinction between an expression’s semantics and its truth value in our mental representation. ○ “The current US President is divorced” vs “The White House is divorced” Points to a possible distinction between world knowledge and linguistic knowledge in semantic memory  ○ Do we determine sentential meanings and then check them against world knowledge? ○ Or do linguistic and WK analysis happen simultaneously? Hagoort et. al, 2004 ○ Can we use EEG to distinguish between world knowledge and linguistic knowledge during language comprehension? True sentence: The Dutch trains are yellow and very crowded WK violation: The Dutch trains are white and very crowded Linguistic violation: The Dutch trains are sour and very crowded Observed an N400 response for BOTH types Retrieval of word meanings and WK happens at the same time! What if we’re reading fantasy or sci-fi or other “weird” discourses, where our WK might not apply? ○ “The clock was depressed” = N400 Nieuwland & Van Berkum, 2006 ○ Participants read passages establishing cartoon like interpretive contexts for inanimate objects ○ At what point in the story does the pragmatic context override the local semantic anomaly? ○ Strong discourse models can neutralize local violations ○ WK is also used immediately when comprehending language Comprehenders construct and update situation models as they gain new information ○ Maybe based on genre expectations ○ Partitioning off “real-world” knowledge and analyzing based on what appropriate for genre Inference Generation Inferences: information and conclusions that are not directly stated in the text (world knowledge) ○ Help maintain coherence of a text Minimal Inference: Inferences are drawn when ○ They are necessary to maintain coherence ○ Based on “quickly and easily available” information.  Bridging inferences: that connects some of the content in a sentence with previous material in the text, or with information encoded in the mental mode ○ Help maintain causal of a text Faster reading times and best memory when causal connections are strongest! ○ Cathy felt very dizzy and fainted at her work. She was carried away unconscious to the hospital. ○ Cathy worked very hard and became exhausted. She was carried away … ○ Cathy worked overtime to finish her project. She was carried away … ○ Cathy had begun working on her project. She was carried away Presuppositions: implicit assumptions that are signaled by specific linguistic expressions and whose existence or truth is taken for granted as background information ○ Instructions to search the situation model for specific content that has already been encoded “The problem with my car turned out to be the battery” → search your situation model for the information that I have a car, my car has a battery, and that my car is experiencing problems. False memory? “Did you see a stop sign?” vs. “Did you see the stop sign?” Elaborative inferences: inferences that are not required in order to make a discourse coherent but that enrich the meanings of sentences. ○ Most likely in situations of high contextual support Show don’t tell: allows readers to make elaborative inferences! ○ Overtly explaining everything in a text can decrease readers’ interest ○ But only if they have enough time/cognitive resources Patty felt like she had been in graduate school forever. Her stipend was minimal and she was always low on cash. Some weeks, she had nothing to eat but peanut butter and jelly. Patty packed her lunch every single day to save money. She yearned for the day she could afford to eat in a restaurant. Alas, she pulled out her sack lunch and looked at its contents. Patty bit into her apple, then stared at it. ○ Final sentence: It had half a worm in it/It had an unpleasant taste/It had little flavor. ○ Patty spit out the bite of apple (consistent)/Patty swallowed the bite of apple (inconsistent).  Longer reading times for inconsistent sentence ONLY when it mismatched the final sentence -> participants made elaborative inferences. But only high WM participants inferred – suggests that elaborative inferences are costly. Lexical Semantics (Study slide 31 map) Sense vs Reference ○ “The blue shape.” “The rectangle.” “The left-hand object.” “The larger object.” Same referent, mean the same thing Words can mean the same thing, but different things Sense: - dictionary definition ○ Blue =/ rectangle =/ left-hand =/ larger Reference: pointers to specific instance ○ Blue = rectangle = left-hand = larger Talking specifically about SENSE meaning in this unit Anaphoric Reference Referring expression = anaphor ○ ‘Evelyn’; ‘she’; ‘My professor’ ○ Used as a memory cue to activate preceding information in the discourse Target = antecedent (or referent) ○ Evelyn Milburn ○ Ante = ‘before’ = ‘the piece that comes before’ Co-reference: deciding that two different expressions refer to the same thing Connecting an anaphor to an antecedent Anaphoric reference is a form of inference ○ Steve wanted to buy his son a gift for his birthday. He went to the pet store and bought a puppy. He was delighted with the gift. What factors make resolving anaphoric reference easier? Antecedents: Focus Focused antecedent: the current topic; most important element; most salient element  Easier to establish co-reference with words in some syntactic positions than others ○ Syntax: subject > object > prepositional object ○ Stever bought his son a puppy. He… Listeners will have the least difficulty resolving an anaphor that refers back to a syntactic subject (performing action of a sentence) Position in the sentence: advantage of first mention Can override advantage from syntactic roles ○ Ann beat Pam in the state tennis match. She…. ○ It was Ann who beat Pam in the state tennis match. She… ○ According to Ann, Pam was a terrible loser. She … Comprehenders will co-refer ‘she’ with ‘Ann’ even though ‘Ann’ is not syntactic subject Antecedents: Semantic Factors Semantic factors associated with the referent word ○ Typicality The ostrich lived in the zoo and the bird (slow to read!) was very docile. The pigeon lived in the zoo and the bird (fast to read!) was very docile. ○ Pigeons are typical birds; ostriches are not ○ ‘bird ‘ serves as a better retrieval cue for ‘pigeon’ than ‘ostrich’ Semantic factors associated with the referent word ○ Implicit causality: verb implies that one of its arguments caused the action ○ Effect of increasing focus on one or another argument Susan blamed Rick because… Focuses object/Rick, so we expect ‘he’ Susan apologized to Rick because…. Focuses subject/Susan, so we expect ‘she’ Antecedents: World Knowledge When are we allowed to use definite vs indefinite articles during reference? ○ Definite article → the; usually used when referring back to previously-introduced antecedent ○ Indefinite article → a/an; usually used when introducing new information  But! If previously-introduced info has a supportive world knowledge structure, we can use a definite article with new information. ○ Normally: ‘an engine’ ○ But with world knowledge: ‘John got into his car. The engine started right away.” ○ And it would be weird to say: ‘John got into his car. An engine started right away.” Anaphors: Explicitness Explicitness: the degree to which the anaphor shares features with the antecedent ○ Zero anaphors - least explicit Use only when a single entity is the focus of attention John went to the store and (zero anaphor) bought some milk ○ Pronouns and noun phrases Share few features with antecedent → can be mapped onto multiple antecedents John and Clint went to the store and he bought some milk ○ Proper nouns - more explicit Degree of featural overlap is perfect → only one possible antecedent John and Clint went to the store and Clint bought some milk ○ When proper noun anaphors are licensed, they speed reading times and make it easier to establish co-reference Repeated Name Penalty Participants read a passage containing an antecedent ○ More prominent antecedent → less explicit anaphor ○ Less prominent antecedent → more explicit anaphor Speakers vary explicitness of anaphor based on prominence of antecedent → listeners can use explicitness as a cue to figure out which antecedent is being referred to. Why not just use proper names/explicit anaphors all the time? ○ Steve bought a puppy and Steve gave it to Steve’s son = BAD  1. There’s usually enough information from lexical, discourse, and pragmatic cues to ID the antecedent 2. Comprehenders process sentences with pronouns and noun phrases faster and more easily than sentences with just proper names a. Comprehender wonders why speaker is giving more info than necessary. b. Comprehender assumes that anaphor is actually new discourse antecedent. Pragmatics & Conversation (EXAM 3) Gricean Maxims What’s wrong with this picture? ○ “Dear Potential Employer, I encourage you to hire Susan. Susan has nice hair. Sincerely, Susan’s Former Boss.” Cooperative Principle → speakers cooperate with listener to share information ○ Take into account listeners' presuppositions, beliefs, and knowledge ○ “Bob succeeded in keeping his eyes open” → only makes sense if we know Bob is tired. Quality: don’t lie; make things up ○ If something is obviously false, there’s probably a reason Relation: make statement relevant to topic of conversation ○ Aids inference process Quantity: say neither more nor less than is required ○ Helps ID correct referents Manner: make your expressions clear; avoid obscure language ○ If violated, we’re doing it on purpose What inferences might a listener make, and what Maxim violation drivers each inference. ○ “How long have you been married?” “4 years, and 16 hours.” ○ Miranda began a relationship with James Holt in April. Three months later, she was dead. ○ Don’t get me wrong. Rick has a number of great qualities as a boyfriend. He keeps his fingernails trimmed. He enunciates his vowels. He has never  had three different blondes show up at his door on the same morning, each claiming to be his new wife. ○ “What does Terrence do for a living?” “He works in an office.” ○ “I’m having a little bit of trouble this morning. Someone appears to be in my parking spot.” Gricean Maxims accomplish two goals ○ Describe factors that speakers should take into account to craft effective statements/ ○ Describe clues that listeners can use to draw inferences But! They describe ideal conversations rather than actual conversations ○ Real conversations involve time pressure ○ Grice never considered feedback and interaction Read Dialog is Interactive Old view of dialog - “chunks of monologue stuck together” (Pickering & Garrod, 2004) Real dialog ○ Overlaps ○ Interactive/collaborative/cooperative ○ Built from multi-turn exchanges A1. Okay. Um let's see. So we need a yellow two by two. Okay and that's going to fit on the right side of the blue block. B1. M-hm. A2. So that half of it oh yeah on one row of the right side of the blue block. B2. Okay *so half of it’s pointing to the right. A3. *So half of it is pointing off to the right. Yeah. B3. Got it Real Dialog Main Channel vs. Back Channel ○ Main Channel → messages being sent by the speaker ○ Back channel → listener indicates understanding (without claiming speakership) Real dialog violates Gricean Maxims ○ Speakers don’t always use the best ways to express their ideas  ○ Speakers produce tentative, provisional descriptions Issue of shared knowledge ○ What both people know about the issue determines what information has to be revealed. Given-new contract ○ Statements should contain both given and new information for communication to be successful. ○ Conversations are collaborative! Common Ground Theory Common ground: mental knowledge and beliefs shared among conversational parties (Brown-Schmidt & Hanna, 2011) ○ Common Ground ≠ Common Knowledge Not only about what each person in the conversation knows but ALSO about what each person knows about what the OTHER person knows Goal of conversation is to increase common ground = successful communication How do conversational partners choose referring expressions? Referential Communication Task ○ Partner up and designate one person as Card Holder and one person as Card Orderer Your job is for the Card Orderer to put their cards in the same order as the Card Holder, without looking at each others’ cards How? Communicate! When you’ve finished, switch roles at least three times (shuffle cards in between) Observe how your conversation changes Asked pairs to match abstract tangram figures Repeated task several times Observed how referring expressions changes over several conversations Early Late A. Okay the next one is...resembles A: Stair climber. someone that looks like they're B: OK. trying to climb stairs. There's two  feet, one is way above the other, and… B. And there's a, there's a, a diamond on the right side, on a slant? A. Yeah B. Got it. A. Like, kind of looks like it’s off the back. B. Right, I got it, Referential pact → pairs collaborate to create mutually understood referring expressions. ○ Makes communications more efficient Violations of “referential pacts” are costly ○ Specific between speaker and listener Different pairs of speakers and listeners have different common grounds Experiments asking pairs of friends to describe university locations while keeping the identity of locations hidden from third persons. ○ Friend could use common grounds related to private experiences (“This is where I put your teddy bear”) ○ Third party failed to understand conversations Disfluencies → non-fluent deviations in speech ○ Common ○ Often create ungrammatical utterances ○ Don’t generally interfere with comprehension ○ May sometimes improve comprehension Production errors? But drunk people say “ummm” less than sober people ‘Filled pauses’ - a means of ‘holding the floor’ Conventional ‘words’ that speakers use purposefully ○ Speaker is experiencing an upcoming delay - length of disfluency = length of delay Arnold et al (2004) (Same Jennifer Arnold who authored the singular ‘they’ paper) Do speakers use disfluencies as a cue to whether a speaker is going to say the name of something given or something new?  ○ Object just mentioned → name should still be in WM = no disfluency ○ Object is new → name isn’t available = possible disfluency Arnold et al (2004) Setup sentence followed by four possible target sentences: Setup: Put the grapes above the candle 1. “Now put the camel below the salt” 2. “Now put the candle below the salt” 3. “Now put thee uh camel below the salt” 4. “Now put thee uh candle below the salt” Which of ‘candle’ and ‘camel’ is given, and which is new? Arnold et al (2004) Fluent article: listeners looked more at given nouns (candle) Disfluent article: listeners looked more at new nouns (camel) Listeners assume speakers are less likely to be disfluent when they are naming something they have recently named Syntactic priming → hearing of a certain syntactic construction increases the chances that the same grammatical structure will be used by the other speaker ○ Confederate reads sentences: The girl gave the book to the boy. The girl gave the boy a book. ○ Participants: Had to find the card from the ones in front of them Pick a response card from the display and describe it to the confederate. ○ On 78% of the trials, the syntax is copied by the participant = syntactic priming Audience Design Speakers adapt what they say ti fit needs of listeners ○ “Macro” adjustments → adjustment applied over a long period of time Bilinguals “code-switch”; infant-directed speech; out-of-towners; volume in noise ○ “Micro” adjustments → continuous, small adjustments based on moment-to-moment changes in the listener’s knowledge, attentional state, and needs.  Mumbling: speakers produce higher-quality output when words are less predictable ‘A stitch in time saves nine’ vs. ‘the next word in this sentence is nice’ ‘Egocentric Production’ Speakers don’t adapt their utterances to needs of listeners Driven by availability ○ Speakers will produce whatever is easiest for listeners to understand Negotiation and repair How Can We Study Dialog? Use natural speaker-listener pairs Exuberant-responding problem: participants can say whatever they want, however they want → hard to control variables ○ Solution → confederate We know speakers make macro-adjustments Unclear how much/often they make micro-adjustments Audience Design Vs. Egocentrism Locally ambiguous sentences are hard for listeners → speakers should avoid ambiguous sentences if they engage in (micro) audience design ○ I knew the coach and his sister would arrive late (Garden path sentence!) ○ I knew that the coach and his sister would arrive late Read sentences, then repeat to the listener. Do speakers insert ‘that’? ○ Nope! Depends on how easy ‘coach; is to retrieve Speakers disambiguate sentences to make their life easier for themselves ○ That is a fancy way of saying ummm Sometimes, failing to engage in audience design = costs for the speaker ○ Special efforts Privileged ground: information that only one participant in a conversation ○ Speakers “leak” privileged information because they fail to consider listeners’ perspective Does audience design or salience/availability have a stronger influence on speaker behavior?  ○ Speakers asked to describe a target object so listeners can ID the object Some objects can be seen by both, but one is in speakers’ privileged ground. Keep identity of privileged object a secret Audience design → speaker should refer to the ‘triangle’ Egocentrism → speaker should refer to the ‘small triangle’ ○ Violates the maxim of Quantity! More mistakes when explicitly instructed to keep occluded item secret ○ Accessibility drives sentence construction choices. What about individual and cultural differences? Wu & Keysar (2007) ○ Listeners: Chinese and American participants ○ Speakers: Study confederates “Move the block one square up.” ○ Audience design: should infer that speaker can only be referring to block they can see ○ Measured via eye tracking Wu & Keysar (2007) ○ Number of fixations: more interfere from competitor for American participants ○ Latency to fixate target: slower selection of target for American Participants ○ Interpretation → Chinese participants were more efficient at perspective-taking to interpret ambiguous utterances Influence of collectivist vs individualist cultures? 1. Are there other potential sources of individual or cultural differences that might be interesting to explore? What effects do you predict you’d find? 2. Are there situational variables that might affect the degree to which people show sensitivity to a partner’s perspective or mental state? An Alternative Perspective Constantly tracking and updating a conversational partner’s knowledge, mental states, beliefs, etc, WHILE dealing with comprehending and producing language seems really hard…  Is all that mind-reading necessary? What if we used a simpler, dumber, mechanism? Interactive Alignment Model: most alignment between conversational partners is the result of simple, automatic, priming mechanisms ○ Misalignments can be fixed by a simple repair mechanism ○ Representing another person’s knowledge, beliefs, etc is used only in special cases Figurative Language Non-Literal Expressions Indirect requests ○ Can you open the door? Can you pass the salt? Idioms ○ Dave kicked the bucket. Kathy spilled the beans. Metaphors ○ Sally flew down the street on her bicycle. That lecture was a sleeping pill. Irony and sarcasm ○ Wow. What a great lecture. Non-literal meaning vs. Standard or literal meaning: words in the expression /= meaning! Relationship between what is said and what is meant is not straightforward -> difficulty? ○ (Non-Literal) The investors looked to the Wall Street banker for advice. ○ (Literal) The animals were grazing on the hillside. ○ Target Sentence: The sheep followed their leader over the cliff. No reading time differences between contexts = fast computation of non-literal meanings Parallel interpretive processing ○ Both literal and non-literal meanings are simultaneously computed ○ The best/most compatible meaning for the context is eventually selected Similar to ambiguous word processing! Metaphor Why metaphor?  ○ Face-saving properties of double messages. ○ Efficiency ○ Metaphors aid comprehension of a new domain E.g., “crime” as “disease” Texts using the metaphor are remembered better than more “literal” texts. Metaphor: indicates a relationship between two elements ○ Topic is vehicle “Freya is a haunted doll” ○ Freya = topic ○ haunted doll = vehicle Grounds of comparison: the relationship that connects the topic to the vehicle ○ “Freya is old and scruffy and has a thousand yard stare” Recognition Problem → we know that literal and non-literal interpretations are computed simultaneously. ○ When the literal meaning is different from the non-literal (intended) meaning, how does the listener compute the non-literal meaning? Career of Metaphor Hypothesis: the processes comprehenders use to compute metaphors change as metaphors progress through their “lifespan” When metaphors are first coined: comparison and property matching ○ Metaphors are mentally converted to similes and shared characteristics are evaluated A is B ➔ A is like B (Freya is like a haunted doll.) Evaluate new simile using same literal simile evaluation mechanisms Meaning! When metaphors are familiar: class inclusion ○ “A is B” is interpreted as “The topic (A) is a member of the category represented by the vehicle (B).” Freya is a haunted doll. Haunted doll exemplifies the category: Things that are endearingly scruffy and woebegone. Prediction: comprehenders should prefer simile forms (A is like B) for novel 🙁 comparisons and category-inclusion forms (A is B) for established metaphors ○ (novel C-I) The dancer is a butterfly ○ (novel S) The dancer is like a butterfly ☺ ○ (Est. C-I) The problem is a roadblock ☺  ○ (Est. S) The problem is like a roadblock 🙁 Conceptual Mapping Hypothesis Most of language is based on implicit metaphors! Understanding everyday words is based on discovering links between different domains ○ Argument – war ○ crime – disease ○ Anger – heated fluid in a container E.g. we talk about crime as ‘disease’ because we mentally analogize eliminating crime to curing a disease How do spatial terms used to express temporal relations affect the way speakers think? ○ Space and movement: The meeting went from 3 to 4. English: primarily horizontal spatial terms to talk about time ○ June is before July, the future is ahead of us, and the past is behind us… Mandarin: both horizontal and spatial terms ○ Earlier events are up and later events are down Boroditsky, 2001 Mandarin- and English-speaking participants watched horizontal or vertical events Then judge statement about temporal relationship between months ○ “March is before/after April” Both groups faster after viewing horizontal events ○ “March is earlier/later than April” English-speaking group faster after horizontal events Mandarin-speaking group faster after vertical events ○ Boroditsky argues that conceptual mapping is mostly likely when we can’t rely on sensory experience Language shapes abstract thoughts Literal or Figurative Under the weather Fall off the wagon Touch grass Deliver the goods  It’s just blueberries Idioms Idioms: multiword phrases in which the meanings of individual words are different from the overall meaning of the phrase 750 most common English idioms: one is found for every 2,666 words of text in the Cobuild corpus ○ Some idioms have both literal and figurative meanings, others only have figurative ○ Idioms frequently refer to abstract moods, emotions, or states, or evoke entire events that are hard to encapsulate ○ Idioms might rely on cultural context ○ It might be more or less easy to see connections between literal and figurative meanings Lexical Representation Hypothesis: each idiom has its own entry in the mental lexicon ○ [kick the bucket] = “to die” in your mental dictionary ○ The individual words in an idiom aren’t analyzed and don’t contribute to the meaning at all Evidence shows that we are sensitive to to words within an idiom We can constrain idiom meanings based on subparts ○ kick the bucket = a sharp and distinct act We can create and interpret novel idioms that are related to old ones ○ Often involves lexical and syntactic flexibility ○ ”shatter the ice”, “didn’t spill a single bean” Hybrid Models: compositional analysis of idiom words AND direct retrieval of idiom meanings happens simultaneously ○ Explains fast reading of idioms: meanings are grabbed directly from the lexicon ○ Also explains single-word effects: compositional analysis happens at the same time Whether compositional analysis or direct retrieval “wins” depends on… ○ Familiarity, meaning dominance, meaning relatedness… Irony  Verbal irony: juxtaposition between what is said and what is meant ○ ironic compliments vs ironic criticisms Asymmetry of affect: ironic criticisms are faster/easier to understand than ironic compliments (Sidenote: sarcasm = irony directed at a person, with intent to criticize) Tinge hypothesis: social function of irony involves muting literal meanings Changes the way ironic speakers are perceived ○ Ironic criticisms: more polite ○ Ironic compliments: less polite Purposes of irony: frustration diffusion, embarrassment diffusion, face-saving ○ Differs by region: more irony in northern US than southern Mentalizing → thinking about others’ mental states ○ Theory of mind? Perspective-taking ? Greater mentalizing capacity might boost irony comprehension ○ Children who haven’t fully acquired mentalizing abilities also can’t understand irony Bilingualism and mentalizing ○ Stronger mentalizing abilities across children and adults ○ Why? Greater language use diversity → better mentalizing What are the relationships between ecological language diversity, mentalizing, and irony comprehension? Tiv, O’Regan, & Titone, 2023 Participants: French/English bilingual living in Montreal Introduction sentence: You and Henry play with your newly-adopted cat. ○ Positive scenario: The cat is energetic and playful for most of the day. ○ Negative scenario: The cat is sleepy and sluggish for most of the day Positive statement: “What a lovely cat,” Henry remarks. Negative statement: “What a lazy cat,” Henry remarks. (+) statement (-) statement (+) scenario Literal compliment Ironic compliment (-) scenario Ironic criticism Literal criticism  Participants rated each statement on appropriateness and ironicness Also evaluated