LING2214 Psycholinguistics PDF

LING2214 Psycholinguistics Week 1 Jul 23, 2024 What is Psycholinguistics? What is language?? Semantics ○ Word and sentence meaning Syntax ○ Sentence structure, grammatical relationships between words Morphology ○ Inflectional, derivational (word-internal structure) Phonetics and phonology ○ Sounds and production of sounds Pragmatics ○ Contextual use Are there any essential differences between human language and other forms of communication? ○ Eg. birdsong, bee dances, whale song, gestures, sound, etc) In looking for evidence of ‘species specificity’, Aitchison concluded that no animal form of communication can be proven to have: ○ Structure dependence (syntax: use of full sentences with grammar) ○ Creativity or productivity Creativity or productivity We can produce and understand an indefinitely large number of utterances, with meanings that have never been encountered previously We are able to use a finite set of rules and words to create an infinite set of meanings We can restructure meaning as well We can talk about anything that occurred at any time (past, present, future) We can talk about our ability to talk about anything that occurred at any time Distinguishing characteristics of human language The richness of human language ○ Our ability to talk about anything using a finite number of words ○ Language appears to be a lot richer than other type of human behaviour Eg. music (communicates emotions primarily) Language in other species? Since human language has characteristics that are not present in any known form of animal communication, can we conclude that ‘language’ (or some aspect of it) is uniquely human? ○ Can we teach language to animals? Language in non-human primates Kanzi is the most famous bonobo Theory 1: ○ Language is the product of being human ○ Humans evolved as a product which generate language Theory 2: ○ Humans have evolved and as a consequence, we produce language as a product of human behaviour Kanzi learned the lexigrams through this inadvertent exposure. He also learned to comprehend the spoken English words to which they correspond (the first ape to do so) Kanzi has demonstrated more impressive linguistic abilities (though perhaps not “human-like”) than primates studied previously. Why? ○ Theory: this advantage could stem largely from the environment in which he was raised (especially for the first year of life) ○ Comparison of language comprehension abilities between the two bonobo siblings, Panbanisha and Tamuli, who were reared under different conditions and achieved different levels of language competence. Can non-human primates acquire ‘human-like’ language? ○ Theory 1: Language acquisition appears largely dependent on the type of environment experienced during the first year of life. The fact that bonobos can comprehend spoken English provides support for theories positing that cognitive functions in humans and animals are continuous rather then of entirely different origins and natures ○ Theory 2: Chimps can learn some symbols and some easy of combining them, but they cannot acquire a human-life syntax - The complexity of human languages seem to go beyond the ability of non-human primates - Young children and chimps are similar in cognitive abilities but very different in linguistic abilities - Human children can learn language because only they have the special innate equipment necessary to do so - Indicated by a language acquisition device (LAD), in particular, with recursive syntactic rules Psycholinguistics Psycholinguistics concerns with studying the psychological processes involved in using language ○ Most of the processes studied are unconscious processes Approaches to psycholinguistics Verbal learning techniques: used in the 1950s and 60s when behaviourism dominated psychology ○ Eg. paired-associate learning ○ Theory that shows how memories can be encoded and retrieved when paired with other information (usually words) in paired associations. Language use was considered a behaviour like any other The emphasis was on associations between observable inputs (stimuli) and outputs (responses) and how they were learned There was no interest in “the mind” Data driven Skinner : Verbal Behaviour ○ Emphasises significance of environment and the history of the organism Noam Chompsky : Verbal Behaviour ○ Does not believe in behaviourism ○ He believes that language behaviour comes from the innateness in the brain ○ “More is involved in sentence structure than insertion of lexical items in grammatical frames: no approach to language that fails to take these deeper processes into account can possibly achieve much success in accounting for actual linguistic behaviour” Eg. Replace “can be” with “is” or “are” ○ He proposed transformational grammar to account for these and other facts of sentence structure ○ He emphasised the nature of the internalised grammatical rule system, which included “transformations” to account for the similarity in meaning between active and passive sentences Fred kicked Bill Bill was kicked by Fred ○ Almost immediately, psycholinguists began to test its ‘psychological implications’ Derivational Theory of Complexity (DTC) ○ Emerged directly from Chompsky’s theory ○ Sentences involving more transformations were considered derivationally more complex Comparisons including passive vs active, negative vs affirmative Findings can be task-dependent (eg. recall vs grammaticality classification) Information processing ○ The ‘computational metaphor’ is still prevalent today ○ Our brain is like a computer and is designed in a way to efficiently and effectively process information Cognitive science ○ Knowledge from artificial intelligence (AI) and relying largely on computer modelling ○ Using the input to generate output to compare with human data, if findings are similar, it can be concluded that human behaviour is very similar to running a computer Connectionism ○ Variant of the cognitive science approach Processing models ‘involve many very simple, richly-interconnected neuron-like units working together without a governing plan Our brain is structured in a way that we can learn and process information efficiently and effectively → thanks to the neural basis ○ Activation Central concept A continuously varying quantity that can spread from one unit or word or point in a network to another Eg. if we hear “ghost”, its representation received activation which spreads out to other related (“vampire” or “goal”) but not unrelated (“pamphlet”) words Neurons have to collectively fire to recognise a face, or a speech sound etc, therefore activation is vital in connectionism When we recognise a face, we are matching what is stored in our mind with external stimuli, external input and the part of the brain that has LEARNED this information is activated. This activation can be translated into the cluster of neurons firing together Cognitive Neuropsychology ○ Uses data from individuals with brain damage (lesions) to develop and modify models of ‘normal’ processing ○ A complex cognitive task (eg. sentence comprehension) is broken down into component processes (eg. understanding individual words, deciding who’s doing what to whom) ○ It is assumed that: Each component can be impaired independently of the others: ‘Fractionism’ Performance reflects the operation of the normal system minus the damaged component (transparency) Neuroimaging ○ EEGs/ERPs measure the electrical activity of the brain by putting electrodes on the scalp ○ MEG measures the magnetic activity of the brain ○ fMRI (functional magnetic resonance imaging) measures the brain locations where activation occurs LING2214 Psycholinguistics Week 2 Jul 30, 2024 Lexical Processing Pt1 Recap: Language Processing and Lexical Processing Uniqueness of Language Learning: ○ Language learning is a uniquely human trait, involving complex cognitive and social processes. Case of Kanzi: ○ Kanzi, a bonobo, demonstrated some language abilities but does not approach the complexity of human language. Cognitive Processes in Language Processing: ○ Many cognitive processes involved in language processing are unconscious and automatic. Stroop Effect: ○ Demonstrates the conflict between automatic and controlled processes in language processing. ○ Incongruent effect Approaches to Studying Language Processing: ○ Focus on architecture (structure) and mechanisms (how it works). Concept of Modularity: ○ The idea that language processing occurs in specialised modules in the brain. Aims Storage and Representation of Words: ○ Investigate how words are stored and represented in the mental lexicon. Methods in Studying Lexical Processing: ○ Priming: Measures how a preceding word affects the response to a target word. ○ Eye-Movement Measures: Analyses reading and word recognition by tracking eye movements. Types of Cognitive Processes Involved: ○ Visual word recognition, auditory word recognition, and related cognitive processes. Models of Visual Word Recognition: ○ Various theoretical models explain how written words are recognized and processed. What is a Language? Lexical Processing Recognition of Words: ○ Differences between recognition of visually presented words (reading) vs. spoken words (listening). Modality Differences: ○ Print: Easier to study due to standardised features and clear onset definition. ○ Speech: More variable and harder to define onset. Measuring Lexical Processing Psycholinguistic Methods: ○ Naming: Measures reaction time from seeing a word to saying it aloud. - Word recognition, reading words aloud, moving articulators (mouth, tongue, vocal folds) to produce words ○ Lexical Decision: Measures reaction time to judge if a letter string is a real word. Response Times (RTs): ○ Naming: ~400 ms ○ Lexical Decision: ~500-600 ms Empirical Findings: ○ Frequency Effect: Common words recognized more quickly. ○ Word-Nonword Effects: Faster response to words than nonwords. ○ Word Length Effects: Long words take longer to pronounce. Priming Priming Technique: ○ Response to a target word is affected by a preceding prime word. ○ Can be facilitatory (faster response) or inhibitory (slower response). Repetition Priming: ○ Faster recognition of a word if previously seen. ○ Frequency of the word affects the extent of priming. Semantic Priming: ○ Responses to a target word are faster if preceded by a semantically related word (e.g., doctor-nurse). Automaticity of Priming: ○ Semantic Priming: Priming is typically automatic but can be influenced by strategies and context. ○ Category Priming: Stronger priming for high typicality members, not entirely automatic. Semantic Priming Fischler (1975): ○ First Related Pair Effect: Small effect for the very first related pair seen by a subject; strengthens with subsequent pairs. Tweedy, Lapinski, & Schvaneveldt (1977): ○ Proportion of Related Items: Priming is stronger with a higher proportion of related words (e.g., 50% related vs. 20% related). Suggests that priming involves some level of strategy, indicating it is not entirely automatic. Conscious Anticipation of the Target Neely (1976): ○ Category Priming: Example: BIRD-ROBIN (lexical decision on target). Expectation: Slower responses if subjects anticipate the wrong word. Experiment: Two SOAs: Short (300 ms) and Long (2000 ms). Related pairs (e.g., BIRD-ROBIN) showed facilitation at both SOAs. Unrelated pairs showed inhibition, particularly at long SOAs (e.g., XXXX-ROBIN was faster than FRUIT-ROBIN). Conclusion: Category priming is stronger for high typicality members, but not due to simple anticipation. Theories of Semantic Priming Prospective Theory: ○ Concept: The prime pre-activates the unit for the target word, facilitating faster processing. Retrospective Theory: ○ Concept: Faster decisions occur if a semantic relationship between the target word and the prime is perceived. ○ Example: If the target is PILLOW and the prime is SLEEP, the confirmation of the hypothesis speeds up decision-making. Experimental Studies: ○ Techniques to eliminate awareness of the prime to determine the nature of priming effects. Priming Without Awareness Forster & Davis (1984): ○ Masking Technique: Used forward and backward masks to investigate priming. Example: house HOUSE (where "HOUSE" is a backward mask). ○ Stimulus Onset Asynchrony (SOA): Same SOA (50-60 ms) for all subjects. ○ Tests for Awareness: Same-different task Lexical decision Two-alternative forced choice ○ Findings: Strong repetition priming (e.g., house-HOUSE). Form-priming (e.g., horse-HOUSE). Effects last about 1-2 seconds. Semantic effects (e.g., doctor-NURSE) require longer SOAs (60+ ms). Cross-language translation priming is strong with different scripts, suggesting semantic involvement. Masked Priming Paradigm Three-Field Masking Paradigm (Forster & Davis, 1984): ○ Demo: What is Shown: #### 你好 DOCTOR What You See: #### DOCTOR ○ Timing: Mask → Prime → Target Example: (DURATION) Mask: #### 500 ms (mask) Prime: 你好 50 ms (prime) Target: doctor 500 ms (target) ○ “Aware” priming is achieved either by lengthening the prime duration, or removing the first mask. Demo Variations: ○ Demo 1: Mask: #### Prime: window Target: doctor Timing: 500 ms (mask), 50 ms (prime), 500 ms (target). ○ Demo 2: Mask: #### Prime: window Target: doctor Timing: 500 ms (mask), 200 ms (prime), 500 ms (target). Models of Lexical Access Direct Parallel vs. Sequential Search: ○ Parallel Process Models: Examples: Logogen Model, Interactive Activation (Connectionist) Models. ○ Sequential Search Process Models: Example: Forster’s Autonomous Serial Search Model. Forster’s Serial Search Model: ○ Concept: Recognition involves a ' mental dictionary search'. Words correspond to lexical entries specifying pronunciation, spelling, meaning, and form class. ○ Process: Encode input to extract an access code. Serially compare the code with entries in access files (phonological, orthographic, or syntactic-semantic). Upon finding a match, a pointer locates the entry in the lexicon proper. Conduct a post-access check with the complete representation from the lexicon proper. Forster’s Serial Search Model: ○ Components: Orthographic Access File Phonological Access File Syntactic-Semantic Access File Lexicon Proper: Bins arranged by frequency (high frequency to low frequency). ○ Empirical Findings: Frequency Effect: High-frequency words are encountered higher in the search path, reducing recognition time. Word-Nonword Effect: Nonwords take longer to identify as such due to the exhaustive search process. Forster’s Serial Search Model Repetition Priming: ○ Forster & Davis (1984): Concept: Repetition priming is interpreted as a “readout effect.” Explanation: Immediately after an entry in the lexicon is accessed, it remains in a moderately excited state, allowing for faster information extraction. Semantic Priming Effects: ○ Explanation: Lexical-semantic (associative) priming is explained by cross-referencing within the lexicon. Example: "cow" is cross-referenced to "milk," "pig," and "horse"; "doctor" is cross-referenced to "nurse" and "patient." Interactive Activation (Connectionist) Models Developed by: David Rumelhart and Jay McClelland Concept: ○ Direct and Parallel Processing: Recognition is based on activation, with no serial search involved. ○ Network Structure: Consists of layers of 'neuronlike' detector units. Typical levels include: Visual Feature Units Letters Words Activation Process: ○ Activation Flow: Presentation of a word activates associated visual feature units. Activation moves from visual features to letter units. Activated letters pass activation to word units (e.g., T excites "TAKE" and "TASK," inhibits "CAKE" and "CASK"). ○ Inhibition: Inhibitory connections decrease activation in less excited units within a level (e.g., "TAKE" inhibits "CAKE"). Activation passes backward through the system, contributing to the 'word superiority effect' (e.g., "TAKE" excites "T," inhibits "C"). ○ Recognition: A word is recognized when the corresponding unit is the only one to remain active. Empirical Findings: ○ Frequency Effect: Activation flows more readily from constituent letters and feature detectors to high-frequency words due to stronger connections built through experience. ○ Word-Nonword Effect: Nonwords are identified as such if a deadline passes before the network reaches a state of equilibrium, where one word unit is more activated than others. Interactive Activation (Connectionist) Models Repetition Priming: ○ Activation Accumulation: When a word stimulus is presented, activation builds up over time in feature-level, letter-level, and word detectors. Activation can accumulate across stimuli that immediately follow each other (Grainger et al., 2012). ○ Masked Prime Effect: When a masked prime word is presented, activation starts to build up in the network. Even though the prime disappears before recognition, it leads to a build-up of activation. When the same word appears as the target, less additional input is needed to reach a criterion level of activation, allowing for faster identification. Lexical-Semantic Priming Effects: ○ Activation Flow: Word detectors send activation to the semantic level, where meaning is represented. Word detectors activate related semantic units beyond the target word itself (e.g., "nurse" activates "patient," "doctor," "hospital"). ○ Feedback Mechanism: Activated semantic units feed activation back to the word level. This feedback activation facilitates the word detector for the related word, making it easier and faster to reach its criterion level for identification. Hence, related words are identified more quickly due to the priming effect. Comparison of Models Evaluation Criteria: ○ Non-Associative Priming: Key data for evaluating models, as it cannot be explained by intra-lexical processes alone. Interactive Activation Models: ○ Context Effects: Account for sentence context effects through feedback from the semantic to the word level, similar to how they account for lexical-semantic priming effects. Search Theory: ○ Limitations: Cannot account for sentence context effects as it is a modular or autonomous theory, where all priming occurs intra-lexically within the lexicon proper. Lexical Ambiguity (Harley, 2014, pps. 198-206) Homophones: ○ Words that sound the same but have different meanings (lexically ambiguous). ○ Examples: Duck, bank, bear, ball, cabinet, bulb, deck (also homographs) Types of Homophones: ○ Balanced Homophones: Have two equally common meanings. Examples: fly, bat, park, spot, nail ○ Biased Homophones: Have one dominant (more frequent) meaning and one subordinate (less frequent) meaning. Examples: ball, coat, game, jar, pen Reference for Frequency Norms: ○ Twilley, L.C., Dixon, P., Taylor, D., & Clark, K. (1994). University of Alberta norms of relative meaning frequency for 566 homographs. Memory & Cognition, 22, 111-126. Models of Lexical Ambiguity Resolution Integration Model: ○ Meanings are accessed in order of dominance (frequency). ○ Context operates post-lexically, aiding the integration of the contextually relevant meaning. Reordered Access Model: ○ Access is exhaustive; all meanings are initially accessed. ○ Meanings are accessed in an order influenced by meaning dominance and preceding context. Evidence for Post-Lexical Selection Swinney (1979) Cross-Modal Priming Experiment: ○ Procedure: Subject hears an ambiguous word in context, followed by a visual probe related to one of the meanings or unrelated. Example Sentence: “The man was not surprised when he found several spiders, roaches, and other bugs in the corner of his room.” Probes: ant (contextually appropriate), spy (contextually inappropriate), sew (unrelated) ○ Findings: Both meanings (ant and spy) are primed immediately after the ambiguous word. Four syllables later, only the contextually appropriate meaning (ant) remains primed. Implication: Both meanings are activated initially, but contextually relevant meaning prevails over time. Summary of Swinney Findings Example Sentence: ○ “...he found several spiders, roaches, and other bugs in the corner of his room.” ○ Probes: ant (contextually appropriate), spy (contextually inappropriate) ○ Facilitated Probe Words: Ant and spy are primed initially, but only ant is primed when the probe is delayed. Subordinate Bias Effect Definition: ○ When an unbalanced homograph is preceded by context favouring its subordinate meaning, it takes longer to read compared to a matched control word. ○ Example Sentence: “The alligator saw the food. He came from the bank (edge) with a quick lunge.” ○ Interpretation: Preceding context raises activation of the homograph’s subordinate meaning. Increased competition between subordinate and dominant meanings results in longer reading times. Eye-Tracking Techniques Tobii TX300: ○ Used for web-based research and auditory language processing. EyeLink 1000/2000: ○ Used for reading research and auditory language processing. Fixations and Saccades (eye tracking technology) Saccade ○ Double checking when reading a sentence (going backwards and forwards with eye movement) ○ To some extent, this tells us that we are unsure about the word or information we are processing so we have to go back to integrate the prior information. ○ In this regard, eye movement data will give you much more information or accurate information than the reaction time studies Word Recognition: Summary Distinction: ○ Processes involved in word recognition (lexical access) vs. word understanding (semantic processing). Techniques for Investigating Lexical-Semantic Processes: ○ Naming ○ Lexical Decision (RT measures) ○ Eye movements ○ Brain imaging/scanning ○ Tachistoscopic Identification ○ Categorization Influences on Word Recognition: ○ Frequency effects ○ Repetition priming ○ Masked priming ○ Word-nonword effects ○ Length effects ○ Lexical-semantic priming Models of Lexical Access: ○ Serial Search ○ Interactive Activation (Connectionist) Models Sentence Context Effects: ○ Used to choose between models of lexical access. ○ Integration Model: Consistent with search theory. ○ Reordered Access: Consistent with logogen and interactive activation models. ○ Most current evidence favours the reordered access model. LING2214 Psycholinguistics Week 2 Tutorial prep - Language in non-humans 1. What is “protolanguage?” How did previous work by Greenfield and Savage-Rumbaugh support use of a protogrammatical (word-ordering) strategy in Kanzi? Protolanguage = explores how language has evolved Investigating the development and evolution of symbolic capacities is to investigate the existence of these capacities in our closest living relatives – chimps and bonobos, who along with humans comprise an evolutionary clade Greenfield and Savage-Rumbaugh wanted to investigate a cross-species comparison in the language domain that established commonalities across the clade in the development of symbolic play. It supported the idea of protogrammatical strategy in Kanzi by showing that Kanzi, a bonobo chimpanzee, used a simple form of word-ordering in his communication, which resembled the early stages of human language development ○ Kanzi showed the ability to use lexigrams in a meaningful sequence 2. What research questions are addressed in this study? Could another bonobo replicate Kanzi’s achievements in the domain of protogrammar? Could a chimpanzee replicate Kanzi’s protogrammatical achievements? What are the similarities and differences between the combinatorial symbolic capacities of Pan (bonobos and chimpanzees) and the third member of the clade, Homo sapiens? To the extent that similarities are found, where do they fall on the ontogenetic scale of human language acquisition? 3. How did Kanzi’s rearing history and early lexigram use differ from that of Panbanisha and Panpanzee? Emphasis on keyboard use and spoken english was begun earlier with Panbanisha and Panpanzee than with Kanzi Kanzi was initially exposed to a keyboard of six symbols during reward-based training sessions with his mother, however Panbanisha and Panpanzee were initially exposed to a keyboard of more than 256 symbols and were expected to learn the use of these symbols without specific training. These different rearing histories were reflected in a difference in the timing of the first meaningful lexigram use Whereas Kanzi produced his first meaningful lexigram at 30 months of age, Panbanisha and Panpanzee produce theirs around age one, a starting point very similar to that of a typically developing child. 4. How was data collected for this study? All uses of the keyboard and other gestures by Panbanisha and Panpanzee were recorded by hand by the caregivers and input into a computer database at the end of the day. This provided more than 8 hours of data collection a day. Kanzi had at least one human caregiver with him recording his communicative productions 9 hours a day, but contextual notes were provided when there was a second caregiver present to function as an observer. For all three apes, the data collection procedures were similar to observational protocols used in classic studies of child language; these provided the comparative foundation for this study 5. Describe the following aspects of the data a. What percentage of Kanzi’s, Panbanisha’s and Panpanzee’s utterances contained two or more elements? How did this compare with results for human children? Combinations of two or more elements accounted for 16.1% of Panbanisha’s utterances (1043/6492) and 15.6% of Panpanzee’s utterances (978/6250) Both apes produced a significantly higher proportion of combinations (vs. single symbols) than Kanzi who had 10.4% (or 1422/13,673 were combinations) While Panbanisha and Panpanzee’s pattern had moved slightly in the direction of human children, in comparison with Kanzi, their proportion of combinations was still much lower than that of human children and closer to that of Kanzi. At age 2, one child had a corpus in which 73% of the utterances combined more than one morpheme; by 22 months of age, the other child in the study had a corpus in which 55% of the utterances combined more than one morpheme. ○ Whereas Panbanisha began using single lexigrams meaningfully at roughly the same age (11 months) as children raised in a North American cultural environment and Panpanzee was not too far behind (20 months) The three children in Brown’s sample had, by age three-and-a-half, all reached a mean length of utterance of four morphemes, with maximum length somewhere around 13 morphemes. ○ At the same age, Panbanisha and Panpanzee’s mean length of utterance was under two semiotic elements (lexigrams, gestures); anything over three elements was a rarity in the corpus, with corresponding limitations on complexity. b. What percentage of Kanzi’s, Panbanisha’s and Panpanzee’s spontaneously produced two-element combinations were “requests?” How did this compare with results for human children? 80% for Panbanisha 93% for Panpanzee 96% for Kanzi Human children = 19.5% of the cases, 43.3% c. Which semantic relation that is frequently produced by human children was never produced by the apes? Possessor-Possessed Semantic relation that is frequent for children in many cultures but was never constructed by the apes. d. What three new meaning relations did Panbanisha and Panpanzee co-construct, and what systematic ordering strategy did they use? Panbanisha and Panpanzee co-constructed 3 new meaning relations involving affirmatives Affirmative-Goal: YES OUTDOORS = used once by Kanzi to ask to go outside Affirmative-Entity: YES BANANA = used once by Kanzi to ask for a banana Affirmative-Action: YES HUG = never used by Kanzi e. How does the overall pattern of data presented in Table 2 reveal the use of systematic ordering strategies by the apes? f. On what grounds do the investigators argue that “gesture-last was an ape, not a human creation” (p. 17)? The apes preferred lexigram-gesture combinations, Entity before demonstrative gesture and action before agent gesture. In contrast, human caregivers preferred agent (gesture) before action (lexigram) This pattern of humans using the opposite order shows that gesture-last was an ape, not a human creation g. When two-element combinations were classified according to whether they referred to a social relation or a tool relation, what pattern did the results show? Q6: page 18-21 LING2214 Psycholinguistics Week 3 Aug 6, 2024 Lexical Processing Pt2 Recognising Words: Revision of Empirical Findings Frequency Effect: ○ The phenomenon where more frequently encountered words are recognized more quickly and easily. Priming: ○ A process where exposure to one stimulus influences the response to a subsequent stimulus, often improving recognition or reaction times. Masked Repetition Priming: ○ A form of priming where the initial presentation of a word is briefly flashed (masked) before being presented again, leading to faster recognition on the second presentation. Lexical-Semantic Priming: ○ Priming occurs when exposure to a word (prime) facilitates the recognition of a semantically related word (target), e.g., "bread" priming "butter." Recognising Words: Revision of Models Fundamental Difference Between Models: ○ Forster’s Serial Search Model: Lexical access is sequential; words are searched one by one. ○ Interactive Activation (Connectionist) Model: Lexical access involves parallel processing; multiple word representations are activated simultaneously and interactively. Representing and Processing Word Meaning Previous Focus: ○ Recognition or lexical processing, knowing a string of letters or sounds is familiar and stored in memory. Current Focus: ○ Representation, organization, and retrieval of word meanings (semantics) from memory. Investigating Word Meanings: ○ Examining how people use semantic knowledge for categorization: Examples of categorization questions: Wild or domestic? Large or small? Fruit or vegetable? Representation and Processing of Word Meanings Categorization Example (Reeves et al., 1998): ○ Task: Categorize words or pictures as ‘fruit’ or ‘not fruit.’ ○ Example: Easier to categorize apples, bananas, oranges, and pears as fruit compared to olives and avocados. ○ Goal: Identify characteristics that define the concept of ‘fruit.’ Denotation vs. Connotation: ○ Denotation: Core, essential meaning (e.g., what makes a dog a dog). ○ Connotation: Emotional or evaluative associations (e.g., ‘dirty,’ ‘friendly,’). Semantic Network Model: ○ Words gain meaning through their associations with other words. Example: The meaning of "DOG" is derived from associations like "barks," "four legs," "furry." Spreading Activation: ○ Mechanism where activation spreads through the network, strengthening connections between closely related words (e.g., "DOG" strongly associated with "barks," less with "meows"). Representation and Processing of Word Meanings: Spreading Activation Semantic Networks: ○ Main problem: Ability to explain nearly any pattern of results. ○ Lack of ability to make falsifiable experimental predictions. Alternative Approach: ○ Representation of meaning in terms of semantic features. Feature Theories: ○ Decompositional approach to semantic representation. ○ Meaning of a word or concept is composed of features. ○ Example (from Harley, 2014, p. 326): Types of Features: ○ Perceptual Features: Describe physical characteristics (e.g., a cat with a tawny yellow coat and black stripes). ○ Functional Features: Describe uses or functions (e.g., an animal farmed for wool). ○ Defining vs. Characteristic Features (Classical View): Defining Features: Essential for category membership (e.g., 'birds lay eggs'). Characteristic Features: Usual but not necessary for category membership (e.g., 'birds fly,' but ostriches and penguins do not). Difficulties with Classical View: ○ Defining features is not easy to find for some concepts. ○ Examples of issues: Wittgenstein’s example of 'game' with categories having 'fuzzy' boundaries. Examples of categories with unclear boundaries: Is ‘stroke’ a disease? When does a cup become a bowl? Examples of Fuzzy Boundaries: ○ Cup vs. Bowl: When width exceeds height. When sides are convex it’s a bowl; if sides are straight it’s a cup. When the edge of the sides meets the base at a sharp angle it’s a cup; if the sides gradually curve into the base, it’s a bowl. Prototype or Family Resemblance View (Rosch and colleagues): ○ No necessary and sufficient conditions for category membership. ○ Emphasis on characteristic features only. ○ Instances of a concept will overlap in some aspects but not others. ○ A fruit will have characteristics in common with many other fruits (family resemblance) but not a single set of features possessed by every member of the fruit category. The Prototype or Family Resemblance View: Prototype Theory: ○ Reflects that some category exemplars are more representative than others. ○ The ‘best example’ of a concept is known as the ‘prototype.’ Evidence for Graded Structure: ○ People rate category exemplars based on how representative they are. ○ Examples of rating experiments: Which is a more representative fruit? BLUEBERRY or COCONUT ORANGE or AVOCADO OLIVE or APPLE TOMATO or KIWI FRUIT STRAWBERRY or CUCUMBER BANANA or APRICOT Prototype Theory Details: ○ People rate ‘apple’ as a more representative fruit than ‘olive’ because the biological concept of ‘olive’ is less commonly used. ○ Clearly defined categories can still have a graded structure (e.g., ‘13’ rated as a better odd number than ‘57’). Fuzzy Boundaries and Prototype Theory: ○ Categories with fuzzy boundaries fit well with prototype theory. ○ Example: Tomatoes, olives, avocados, and pumpkins could be categorized as both fruits and vegetables. Problems for Prototype Theory: ○ Difficult to identify a ‘best example’ for abstract categories (e.g., ‘truth,’ ‘freedom,’ ‘hypocrisy’). ○ In determining if a number is odd, the defining feature is divisibility by 2, not representativeness. Feature Theories: Interim Summary: ○ Classical View vs. Prototype View: Classical View: Categories should be clearly definable. Prototype View: Boundaries are ‘fuzzy,’ and concepts are graded by their resemblance to the prototype. Criticisms of Feature Theories: ○ What constitutes a feature? ○ What is the most basic level of meaning, or the ‘atoms’ of meaning? ○ How to address the problem of ‘infinite regress’? ○ How to select relevant features for concept representation? Example: Would you choose the feature ‘heavy’ to represent ‘piano’? Context matters: ‘The man lifted the piano’ may affect feature relevance. ○ Revisiting kinship terms: Should ‘+ older’ be assigned to ‘mother’ or ‘grandmother’? Context Dependence of Features: Context Dependence: ○ Context can determine category membership. ○ Two Types of Features: Independent of Context: E.g., Pianos have keys that produce music; basketballs are round and bounce. Context-Dependent: E.g., Pianos are heavy; basketballs float. ○ Context influences which features and meanings are retrieved from semantic memory. Barsalou (1982) Study: Property Verification Task: ○ Participants verified properties of nouns after reading sentences. ○ Independent Properties: E.g., “Banks typically contain money” (context had no effect on verification time). ○ Context-Dependent Properties: E.g., “Can be walked upon” is not typically associated with a roof (verification time dependent on context). ○ Contextual Verification Times: “The bank had been built ten years ago” vs. “The bank was robbed by three bandits.” “The roof had been renovated prior to the rainy season” vs. “The roof creaked under the weight of the repairman.” Theory Theories: Interim Summary Approach from Work with Children: ○ Knowledge is the basis for representing semantic categories. ○ Features play a less defining role. Theory Theories: ○ People categorize objects based on inherent properties rather than characteristic features. ○ Example: Gold is identified as gold due to its microstructure, not because of characteristic features. People might categorize a zebra as a zebra even if its stripes were dyed and not visible. The Organisation of Semantic Memory (Rogers et al., 2004) Proposed a connectionist model of semantic memory. Aims to explain both representation of meaning and the mapping of language onto the external (visual perceptual) world. Reference: Rogers et al. (2004). Structure and deterioration of semantic memory: A neuropsychological and computational investigation. Psychological Review, 111, 205-235. Note: Visual feature units are separate from units representing verbal propositions about visual features. A Connectionist Model of Semantic Memory Each unit in the verbal layer corresponds to: ○ A name (e.g., animal, bird, goose) ○ A visual property (e.g., has eyes, has wheels) ○ A functional property (e.g., can fly, can roll) ○ An encyclopedic property (e.g., lives in Africa, found in kitchen) "The arrangement of verbal units into separate pools … has no functional consequence" (Rogers et al., 2004, p. 207). Each unit in the visual layer corresponds to: ○ A visual feature (e.g., is round, has limbs). Network Operation (Rogers et al., 2004) ○ Reference: Rogers et al. (2004). Structure and deterioration of semantic memory: A neuropsychological and computational investigation. Psychological Review, 111, 205-235. ○ Note: Visual feature units are separate from units representing verbal propositions about visual features. Connections and Activation ○ All units in the "verbal" and "visual" layers are connected to the set of semantic units. ○ Connections have different "weights" based on previous experience and associations. Presenting a Visual Stimulus ○ Units corresponding to the visual features of the stimulus are activated (e.g., features of a “canary”). ○ Other visual feature units remain inactive. ○ To function correctly, activation must flow to related verbal feature units through the semantic units. Alternative Direction ○ The network can also be presented with a name or verbal proposition and work in the opposite direction, activating visual features. Connectionist vs. Feature-Based Models Advantages of Connectionist Models: ○ No need to identify actual semantic features; meaning emerges as a pattern of activation across a network of detector units. ○ Units do not need to be explicitly labelled. ○ Simulations require explicit detailing of how knowledge is represented and used in processing tasks. Understanding Words: Summary ○ Investigation of semantic memory often involves categorization tasks. ○ Feature-Based Views: Classical View: Defining features and fuzzy boundaries are problematic. Prototype or Family Resemblance View: More flexible in handling defining features and boundaries. ○ Problems with Decompositional Views: Identifying relevant features. Choosing which features to use in representing a concept. ○ Knowledge-Based Theories (Theory Theories): Provide an alternative to feature-based and connectionist models. LING2214 Psycholinguistics Week 4 Aug 13, 2024 Sentence Processing Sentence Comprehension: Experimental Control: ○ Most psycholinguistic experiments investigate how people interpret sequences of words outside of conversational context. ○ Purpose: Eliminate contextual factors that might influence sentence meaning. Experimental Procedure: ○ Sequences of words are presented outside of context to minimise extraneous factors affecting sentence interpretation. ○ Internal operations, such as brain activity, that are outside the experimenter’s control, play a role in constructing meaning. Contextual Processing: ○ Normally, sentence processing occurs within a context with a goal. ○ Speakers and hearers align their mental models of the conversational context. Mental Model Alignment: ○ Hearers use the sequences of words to update their mental models, often by eliminating possible states of affairs. ○ Example: In a context with John, Mary, and Bill, the sentence “John or Mary stole my iPhone” indicates Bill did not steal the iPhone. This context-based process is challenging to investigate outside of context. Research Focus: ○ Chapter 10 explores whether sentence grammar is psychologically real. ○ Examines if features of sentence structure (e.g., tree diagrams) can explain how people resolve structural ambiguities. Types of Ambiguities: ○ Structural ambiguities ○ Lexical (form class) ambiguities Form Classes (Harley, 2014, pps. 37-38): ○ Nouns: Refer to objects or ideas (e.g., pig, truth) ○ Verbs: Refer to actions, states, or assertions (e.g., kiss, modify, is) ○ Adjectives: Describe attributes (e.g., pink, lovely) ○ Adverbs: Qualify verbs (e.g., quickly, slowly) ○ Determiners: Articles that agree in number with nouns (e.g., the, a, some) ○ Prepositions: Indicate relationships (e.g., in, to, at) ○ Conjunctions: Connect words or clauses (e.g., and, because, so) ○ Pronouns: Substitute for nouns (e.g., he, she, him, her) Function vs. Content Words: ○ Content Words: Convey most of the meaning (e.g., nouns, verbs, adjectives, adverbs) ○ Function Words: Perform grammatical work (e.g., determiners, prepositions, conjunctions, pronouns) ○ Open Class Words: Large number and expandable (e.g., texting, bromance) ○ Closed Class Words: Small, fixed number Lexical (Form Class) Ambiguity Reference: Bucaria, C. (2004). Lexical and syntactic ambiguity as a source of humor: The case of newspaper headlines. Humor, 17, 279-309. Examples: ○ Doctor testifies in horse suit ○ Red tape holds up bridge ○ Stolen painting found by tree ○ NJ judge to rule on nude beach ○ Squad helps dog bite victim ○ Eye drops off shelf ○ Dealers will hear car talk at noon ○ Research fans hope for spinal injuries Structural Ambiguity Reference: Bucaria, C. (2004). Lexical and syntactic ambiguity as a source of humor: The case of newspaper headlines. Humor, 17, 279-309. Examples: ○ L.A. voters approve urban renewal by landslide ○ Dr. Ruth to talk about sex with newspaper editors ○ Killer sentenced to die for second time in 10 years ○ Youth steals funds for charity ○ Man shoots neighbour with machete ○ Enraged cow injures farmer with axe Understanding Sentences Acceptable Continuations of “The horse raced past the barn …” ○ … in the dark late last night ○ … that was full of hay ○ … and jumped over the barrier ○ … jumped over the barrier Reduced Relative Clauses: ○ The horse raced past the barn … and jumped the barrier ○ The horse (that was) raced past the barn … jumped the barrier ○ The horse is the agent (did the racing and jumping) ○ The rider is the agent (made the horse race; “raced” could be replaced by “ridden”) Syntactic Parsing Word Order as a Cue: Fred chased Bill. Bill chased Fred. ○ Distinguishes meaning based on sentence structure. ○ Examples: 3. The sparrow built the nest. 6. The nest built the sparrow (not sensible). 7. Colourless green ideas sleep furiously (grammatical but meaningless). 8. Cf. Sleep colourless green furiously ideas (ungrammatical). Sentence Parsing Additional Cues Beyond Word Order: 7. Bill was chased by Fred. 8. It was Fred that chased Bill. 9. It was Bill that was chased by Fred. 10. Fred was the one that chased Bill. 11. Bill was the one that was chased by Fred. 12. The one that Fred chased was Bill. 13. The one that was chased by Fred was Bill. Assigning Sentence Structure Models of Syntactic Processing: ○ 1. Autonomous Two-Stage Model: Constructs a purely syntactic representation incrementally, word-by-word. Semantic information is added later. ○ 2. Interactive Model: Uses both referential (contextual) and syntactic information to construct sentence representations. Assigning Sentence Structure ○ The major difference between models of sentence comprehension concerns the timing of processing: When referential information is used. When meaning is computed. ○ Models of parsing aim to address three main questions: Are sentences assigned grammatical structure? Why do we resolve structural ambiguities in particular ways? Why is it harder to understand some sentences than others? ○ Most psycholinguists believe that sentences are understood on-line (incrementally, as each new word is encountered). ○ Note: Parsing = assigning syntactic structure to a sentence. The ‘Garden Path’ Model: The Sausage Machine ○ Parsing occurs in two stages: Stage I: Relies solely on syntactic information. If the input is ambiguous, a single structure is created based on syntactic preferences. Minimal attachment Late closure Stage II: Semantic or thematic information is used to revise an incorrect initial parse if necessary. Minimal Attachment ○ Definition: When the parser faces a choice about how to integrate new words into the current phrase/clause, it will prefer the ‘simpler’ option. ‘Simpler’ is defined as the structure requiring fewer nonterminal nodes. This preference assumes that an interpretation involving fewer nodes can be processed more quickly. Minimal Attachment The manager opened the door with … ○ … a master key (attachment to verb phrase) ○ … a broken knob (attachment to noun phrase) Terminal vs. Non-Terminal Nodes Terminal Nodes: Individual words in a sentence. ○ Example: the … horse … ran … past … the … barn Non-Terminal Nodes: Syntactic and phrasal categories assigned to groups of words. ○ Examples: VP (Verb Phrase) DET (Determiner) N (Noun) NP (Noun Phrase) Examples of Node Structures: 1. Minimal Attachment: ○ Attachment to Verb Phrase: Structure: VP → V NP PP Example: opened the door with a master key Simplifies to a single addition to the phrase structure. ○ Attachment to Noun Phrase: Structure: VP → V NP NP PP Example: opened the door with a broken knob Requires an additional NP node. 2. Minimal Attachment: ○ The secretary found … … the wallet in her backpack (continues the verb phrase) … that the wallet was damaged (begins a new sentence) Explanation: Simpler Option (Minimal Attachment): ○ Attach the noun phrase directly to the verb phrase, e.g., found the wallet in her backpack. ○ This option avoids the need to construct a new sentence and requires fewer structural changes. Complex Option (Non-Minimal Attachment): ○ Would involve attaching a new sentence or complex structure, which is more involved than the simpler option. By following minimal attachment, syntactic structures are simplified, minimising the need for additional nodes and maintaining clarity in sentence parsing. Non-Minimal Attachment The more complex option: ○ Involves constructing a new sentence (e.g., “…the wallet was damaged”). ○ Requires an additional sentence node (S) in the syntactic structure. Structure Example: ○ VP V: found S NP: the wallet VP was damaged The ‘Garden Path’ Model Parsing Stages: ○ Stage I: Relies solely on syntactic information. Creates a single structure based on syntactic preferences. Minimal Attachment: Prefer simpler structures by attaching new material to the existing phrase. Late Closure: Incorporates new material into the current clause or phrase. ○ Stage II: Uses semantic or thematic information to revise an initial parse if necessary. Being Led Up the Garden Path: ○ Example: The manager opened the door … with Initially Preferred: Attach preposition to the verb phrase (“opened with”). Correct Interpretation: Sometimes leads to reanalysis when additional information (e.g., “broken knob”) disambiguates the structure. Incorrect: opened the door with a broken knob. Correct: opened [the door] [with a broken knob]. Late Closure: ○ Definition: Incorporate new material into the clause or phrase currently being processed. ○ Example: Since Jay always jogs two kilometres … Late Closure: …this seems like a short distance to him (attaches “two kilometres” to the verb phrase). Early Closure: …seems like a short distance to him (begins a new clause). ○ Findings: Reading times are longer in disambiguating regions when late closure is preferred. Early Evidence for the Garden Path Model: ○ Frazier and Rayner (1982): Compared reading times for late vs. early closure sentences. Late Closure: Since Jay always jogs a mile and a half, this seems a short distance to him. Early Closure: Since Jay always jogs, a mile and a half seems a very short distance to him. ○ Rayner and Frazier (1987): Compared ambiguous reduced complements with unambiguous complements. Ambiguous: The criminal confessed his sins harmed many people. Unambiguous: The criminal confessed that his sins harmed many people. ○ Findings: Reading times longer for ambiguous sentences due to the need for reanalysis. Predictions and Findings: For Complements: ○ If Minimal Attachment is Preferred: Reading times should be faster in the disambiguating region for sentences aligned with minimal attachment. ○ If Non-Minimal Attachment is Preferred: Reading times should be longer in the disambiguating region. ○ Result: Sentences with ambiguous structures lead to longer reading times in the disambiguating region compared to unambiguous sentences. The Garden Path Model - Evaluation: Evidence: Supported by studies of typical language users and individuals with brain damage. Addresses Key Questions: ○ How sentences are assigned phrase markers. ○ Why structural ambiguities are parsed in particular ways. ○ Why some sentences are harder to parse than others. Parsing Summary: Sentence Meaning: Involves assigning thematic roles and depends on syntactic processing. Syntactic Processing: Involves the assignment of phrase structure and is incremental, occurring as each new word is encountered. Week 4 Tutorial Prep 1. What three cognitive skills did Anema and Obler identify as potentially influencing reading comprehension in older adults? The three skills linked to the study are processing speed, inhibition of distracting material and working memory. 2. Explain the distinction between on-line and off-line tasks. On-line tasks: participant assigns the meaning at the time of reading (via tests of word-by-word reading) Off-line tasks: participants are required to understand the material and perform other tasks (such as sentence and picture matching) 3. What two views have been proposed in the literature in regard to the role of working memory (WM) in sentence comprehension? The on- and off-line contrast Kemtes and Kemper (1997) Focus: Investigated on-line reading times and off-line comprehension of syntactically ambiguous sentences. Participants: Younger and older adults. Findings: ○ Older adults had slower on-line reading times compared to younger adults. ○ Within the older group, those with higher working memory (HS readers) read faster and comprehended better than those with lower working memory (LS readers). ○ Higher working memory correlated with better comprehension performance. Conclusion: Off-line measures (like reading comprehension) can reveal age-related declines in language processing, particularly in relation to working memory. Waters and Caplan (2005) Focus: Examined whether off-line performance reflects syntactic processing or "sentence review processing." Participants: Younger and older adults. Findings: ○ Both age groups performed similarly on plausibility judgments of semantically plausible and implausible sentences, regardless of sentence complexity. ○ No significant age-related differences in on-line syntactic processing were observed. Conclusion: On-line syntactic processing is not affected by age. The working memory resources used for syntax processing remain stable across the lifespan. Off-line performance might reflect review processing rather than direct syntactic processing. Comparison On-line Processing: ○ Kemtes and Kemper: Found that older adults show slower on-line processing compared to younger adults. ○ Waters and Caplan: Found no significant age-related differences in on-line syntactic processing. Off-line Processing: ○ Kemtes and Kemper: Suggested that off-line measures can highlight age-related declines in language processing, linked to working memory. ○ Waters and Caplan: Argued that off-line measures might reflect sentence review processing rather than pure syntactic processing, with no age-related decline in on-line syntactic processing. Working Memory: ○ Kemtes and Kemper: Found that working memory impacts both on-line and off-line language processing, with higher working memory correlating with better performance in older adults. ○ Waters and Caplan: Suggested that working memory resources for syntactic processing are stable across the lifespan, implying that age does not affect the availability of these resources. 4. How was hyphenisation used in Anema and Obler’s study to reduce syntactic complexity? Furthermore, within the set of (syntactically) complex sentences, older adults’ reading comprehension is further reduced when WM demands are high. To best facilitate reading comprehension in older adults, then, we must attempt to reduce WM load and/or syntactic complexity. Break down complex phrases if it contributed to comprehension of our sentences with hyphenated and non-hyphenated ambiguous noun phrases (NPs), predicting that the hyphens would reduce WM load. 5. On what variables were sentences matched in Anema and Obler’s study? Can you think of any potentially important unmatched variables? Sentences were matched on length (in syllables) across the 4 conditions (ambiguous, non-ambiguous, hyphenated vs non-hyphenated) Should have matched on word frequency (eg. ambiguous compound “little-used” were similar to unambiguous compounds like “nursing-home”) 6. Why did the authors present comprehension data for only the whole-sentence condition? It only reflects “the natural reading process” 7. What did the analysis of comprehension data show? Ambiguous sentences: younger participants did better than older Unambiguous sentences: non-hyphenated compounds were easier than hyphenated There were no comprehension differences between high span and low span readers 8. How were sentences divided into regions for analysis of on-line reading time data? What did the analysis of these data show? Region 1: text preceding the ambiguous compound Region 2: the ambiguous compound Region 3: text following the ambiguous compound Younger readers were faster than older participants in region 1 The authors compared reading times across three regions in both types of sentences (Region 1 included the sentence initial cleft and was the same for both sentence types, Region 2 included a NP and verb and the word order varied between the two types of sentences, and Region 3 included the sentence final prepositional phrase and was the same for both sentences). Of the three regions only the second region distinguished subject-cleft sentences from object-cleft sentences. Whereas younger adults responded to both the overall imposed memory load and to a more specific interference caused by the form of NP to be remembered, older participants responded only to the more general form of memory load that resulted in slower on-line reading times 9. According to the authors, why might WM have played a relatively minor role in understanding the sentences used in their study? The authors point out that a WM theory specifying separate resources for syntactic processing (e.g., that of Waters and Caplan) cannot explain the different performance by younger and older adults on the tasks described above. They concluded that memory load and syntactic processing most likely draw on the same WM capacity 10. What did you learn from this paper about the effectiveness of hyphenation for improving sentence reading in young versus older adults? Hyphenation improves reading particularly in the older participants LING2214 Psycholinguistics Week 5 Aug 20, 2024 Sentence Processing Assigning Syntactic Structure Two Main Models: 1. Autonomous Two-Stage Model: ○ Construct a purely syntactic representation using only syntactic information. ○ Add semantic information to the syntactic framework afterward. 2. Interactive Model: ○ Use both semantic (contextual) and syntactic information from the outset to construct the syntactic representation. Questions Addressed by Both Approaches: Why are sentences assigned the grammatical structures they are? Why are humans biased to parse structural ambiguities in particular ways? Why is it harder to parse some sentences than others? Psycholinguistic Belief: Syntactic structure is assigned incrementally as each new word becomes available. Parsing: Parsing = assigning a grammatical structure to a sentence. The ‘Garden Path’ Model: 1. Stage I: ○ Processor uses only syntactic information. ○ Creates a single structure aligned with syntactic preferences if the input is ambiguous. ○ Preferences include: Minimal Attachment Late Closure 2. Stage II: ○ Use semantic or thematic information to revise an incorrect initial parse if necessary. Minimal Attachment: Parser chooses the ‘simpler’ option, where simplicity means fewer nonterminal nodes. Simpler structures are processed more quickly. Example: Attach noun phrase directly to the verb phrase if no indication of needing a new sentence. Non-Minimal Attachment: More complex structures involve constructing an additional sentence node. Example: Construct a new sentence when necessary, adding an extra node. Late Closure: New material is incorporated into the current clause or phrase being processed. Late Closure: What It Means in Practice Complete the Following Sentence Fragment: ○ Since Jay always jogs two kilometres… Late Closure: “… this seems like a short distance to him.” (Here, “two kilometres” attaches to the verb phrase, which is closed late.) Early Closure: “… seems like a short distance to him.” (“Two kilometres” begins a new clause, so the verb phrase is closed early.) The ‘Garden Path’ Model Parsing Takes Place in Two Stages: 1. Stage I: The processor uses only syntactic information. If ambiguous, only a single structure is created based on syntactic preferences. Preferences include: Minimal Attachment Late Closure 2. Stage II: Semantic or thematic information can be used to revise an incorrect initial parse if necessary. Being Led Up the Garden Path Example Sentence: ○ “The manager opened the door… with” Prefer to attach the preposition to the verb (“opened with”), but this can lead to incorrect parsing: Correct Interpretation: “The manager opened the door with a master key.” Incorrect Interpretation: “The manager opened the door with a broken knob.” In this case, we’ve been “led up the garden path” by our preference for attachment to the verb. To correct: revise the initial parse upon reaching disambiguating regions, e.g., “broken knob.” Parsing Diagram: ○ Minimal Attachment (Incorrect): VP → V NP PP opened the door with a broken knob ○ Non-Minimal Attachment (Correct): VP → V NP opened the door PP → with a broken knob Early Evidence for the Garden Path Model: Late Closure Study by Frazier and Rayner (1982): ○ Compared readers’ eye movements for late closure versus early closure sentences. Late Closure: “Since Jay always jogs a mile and a half / this seems a short distance to him.” (Ambiguous region extends the verb phrase.) Early Closure: “Since Jay always jogs / a mile and a half seems a very short distance to him.” (Ambiguous region begins a new clause.) Predictions and Findings: ○ Late Closure Preference: Reading times in the disambiguating region should be faster in late closure sentences than in early closure sentences. ○ Early Closure Preference: Reading times in the disambiguating region should be faster in early closure sentences than in late closure sentences. ○ No Preference: Reading times in the disambiguating region should be similar if neither closure type is preferred. ○ Findings: Reading times were longer in the disambiguating region of early closure sentences than late closure sentences. Early Evidence for the Garden Path Model: Reduced Complements Study by Rayner and Frazier (1987): ○ Compared eye movements while reading ambiguous reduced complements (without "that") versus unambiguous "that" complements. Example Sentences: ○ Ambiguous Reduced Complement: "The criminal confessed his sins harmed many people." (Non-minimally attached; ambiguous, more complex parsing required) ○ Unambiguous “That” Complement: "The criminal confessed that his sins harmed many people." (Non-minimally attached; explicit complementizer “that” clarifies structure) Non-Minimal Attachment: ○ More complex parsing involves constructing an additional sentence node. ○ In the ambiguous case, no indicator of needing a new sentence is present until the verb phrase "harmed many people." No Need for Minimal Attachment When "That" is Present: ○ The only option is to construct a new sentence following the verb "confessed." ○ The presence of the complementizer "that" indicates the need for this construction. Experimental Logic: ○ Unambiguous Sentence (2): No need for minimal attachment as structure is clear. Readers will not be garden-pathed. ○ Ambiguous Sentence (1): Minimal attachment applies but is not consistent with the actual structure. Readers will be garden-pathed, experiencing difficulty in the disambiguating region. Predictions and Findings: Complements If Minimal Attachment is Preferred: 1. Reading times in the disambiguating region should be faster in the unambiguous sentence than in the ambiguous sentence. If Non-Minimal Attachment is Preferred: 1. Reading times in the disambiguating region should be slower in the ambiguous sentence than in the unambiguous sentence. Findings: 1. Reading times were longer in the disambiguating region of sentences with ambiguous complements than in those with unambiguous complements. Example Sentences: 1. The criminal confessed his sins harmed many people. 2. The criminal confessed that his sins harmed many people. The Garden Path Model - Evaluation Support for the Garden Path Model: ○ Evidence from studies of typical language users and individuals with brain damage provides some support. Helps Answer Three Questions: ○ Why are sentences assigned the phrase markers they are? ○ Why are humans biased to parse structural ambiguities in particular ways? ○ Why is it harder to parse some sentences than others? Two Models of Parsing Garden Path Model: ○ Described as "one of the most influential models of parsing" by Harley (2014, p. 295). Constraint-Based Model: ○ Main competitor to the garden path model in recent years. Constraint-Based Parsing (An Alternative to the Garden Path Model) Role of the Verb: ○ The verb plays a crucial role in the initial assignment of syntactic structure. Study by Holmes, Stowe, and Cupples: ○ Investigated whether readers’ preference for minimal attachment could be altered by changing the verb. ○ Used reduced complement sentences, such as: "The reporter saw her friend was not succeeding." Experimental Logic: Reduced Complements Unambiguous Sentence: ○ Sentence 1 is unambiguous, so minimal attachment does not apply. ○ Readers will not be garden-pathed. Experimental Logic: Reduced Complements Sentence 1 is unambiguous, so minimal attachment does not apply. Readers will not be garden-pathed. Sentence 2 is ambiguous, so minimal attachment does apply. ○ Minimal attachment states that readers will attach the ambiguous noun phrase (her friend) directly to the verb phrase (saw) because that is the simplest structure. ○ When reaching the disambiguating verb phrase (was not succeeding), readers will need to go back and reanalyze the sentence in line with the more complex structural reading, inserting a new sentence node between saw and her friend. Pre-ambiguity “that”: The reporter saw that her friend was not succeeding Ambiguous: The reporter saw her friend was not succeeding Reduced Complements: Example Items Holmes, Stowe, and Cupples showed that the garden-pathing effect in reduced complement sentences could be altered by the verb. Readers had more difficulty with verbs like “saw” than with “said.” Constraint-based Parsing (An Alternative to the Garden Path Model) Why did readers have more difficulty with sentences containing ‘saw’ than ‘said’? ○ Verbs like ‘saw’ tend to be used with a simple noun phrase more often than with a ‘that’ clause. Example: The reporter saw the car accident (more frequent). Example: The reporter saw (that) the car was damaged (less frequent). ○ Verbs like ‘said’ tend to be used more often with a ‘that’ clause. Example: The lecturer said nothing interesting (less frequent). Example: The lecturer said (that) the phrase was unusual (more frequent). Findings and Implications Findings by Holmes et al. suggest that the frequency with which verbs are encountered in specific syntactic contexts can influence comprehension ease in both those contexts and different ones. Results are not entirely consistent with general parsing strategies like minimal attachment, which are assumed to apply universally. Findings align more with constraint-based models of parsing. The Role of the Verb in Parsing Results from studies like those of Holmes et al. and Frenck-Mestre and Pynte support the idea that syntactic parsing preferences can be influenced by the type of verb in a sentence. Not all verbs are equally likely to be interpreted according to general syntactic preferences like minimal attachment and late closure. The ultimate goal of understanding a sentence is to assign thematic roles. Thus, focus should be on both syntactic processing and the assignment of meaning. Investigation is needed to determine if verb type effects are present in less complex sentences where there is no attachment ambiguity. Experiencer Verbs Experiencer verbs describe states or emotions. Two types of experiencer verbs differ in how thematic roles map onto syntactic functions: ○ Stimulus-Experiencer SURPRISE: [NP --- NP]; [Stimulus Experiencer] AMUSE: [NP --- NP]; [Stimulus Experiencer] ○ Experiencer-Stimulus FEAR: [NP --- NP]; [Experiencer Stimulus] CHERISH: [NP --- NP]; [Experiencer Stimulus] Thematic Hierarchy According to the thematic hierarchy, some roles are more prominent than others (e.g., agent is more prominent than theme; experiencer is more prominent than stimulus). Sentences are easier to understand if nouns filling more prominent thematic roles also appear in more prominent syntactic positions (e.g., agent or experiencer as subject). ○ Example 1 should be easier than Example 2, but Example 3 should be more difficult than Example 4. 4. The boy feared the dog. 5. The dog scared the boy. 6. The dog was feared by the boy. 7. The boy was scared by the dog. Experiencer Verb Sentences Reference: Cupples, L. (2002). The structural characteristics and on-line comprehension of experiencer-verb sentences. Language and Cognitive Processes, 17, 125-162. Word-by-word reading times were examined for simple active and passive sentences containing experiencer verbs of different types. Experiencer-Verb Sentences Participants: ○ 40 native English speakers (undergraduate students) Procedure: ○ Word-by-word reading times recorded in a meaning-classification task. Participants pressed the right-shift key to see each word in a sentence. If they noticed an anomaly, they pressed the left-shift key. Focus: ○ Response (reading) times at and just after the main verb were of special interest. Experiencer Verb Results As predicted and consistent with the thematic hierarchy: ○ Readers took significantly less time to judge the acceptability of sentences with an experiencer subject (1b, 2a) than sentences with an experiencer object (1a, 2b) at the verb and the word immediately after it. Influences on Sentence Comprehension ○ Findings extend results from studies on temporary structural ambiguities to show: Differences in meaning assignment can impact sentence comprehension difficulty even when sentences are matched in syntactic structure (both active or both passive) and are relatively simple, unambiguous structures. ○ Individual Differences Might there be individual differences in comprehension ability? Individual Differences ○ How might individuals differ? Working Memory Syntactic Ability What is Working Memory? ○ Working memory is a cognitive processing system responsible for storing and processing information. ○ It has limited capacity, so storage and processing compete for the same resources. If processing is efficient, more resources are left for storage. If processing is inefficient, fewer resources are left for storage. ○ Working memory is assessed using tasks that require both processing and storage (e.g., Daneman & Carpenter’s ‘reading span task’). Reading Span ○ In the traditional reading span task: Participants read aloud sets of individually presented, unrelated sentences (containing from 13 to 16 words each). After each set, they recall as many sentence-final words from that set as they can. LING2214 Psycholinguistics Week 6 Aug 27, 2024 Language Acquisition The Big Question How do children acquire the language spoken in their environment? By ages 3 or 4, children have a grammar enabling communication. Development typically avoids: ○ Overshooting: going beyond the target grammar. ○ Undershooting: not reaching the target grammar (not knowing enough grammar). Two Potential Problems of Language Learnability Overshooting: errors that exceed the target grammar. Undershooting: errors that fall short of the target grammar. The Debate Learning vs. Innate Knowledge: ○ How much language knowledge is learned vs. innate (genetic make-up)? Models of Language Acquisition Approach 1: Experience-based Approaches (Empiricist): ○ Skinner, Tomasello. ○ Children learn the rules from the input, the input either gives them positive or negative feedback and then they build on this feedback, learning about the rules and language. Approach 2: Generative Approaches (Rationalist, Nativist): ○ Chomsky. ○ Language is built upon innate knowledge. Approach 3: Interactionist Approaches: ○ Piaget (Cognitive). ○ Language is learnt through social interactions. Cues and rules are learned from social interactions. A Difference in Emphasis Innate Linguistic Approach Learning Behavioral Approach Social Interaction Approach Cognitive Approach Are the Different Approaches Talking About the Same Thing? Learning vs. Innate Knowledge: What do researchers mean by "language"? A Distinction Competence (I-language): ○ Mental grammar internalised by an idealised speaker/hearer. Performance (E-language): ○ Actual language use in real-life situations. Competence vs. Performance Linguistic competence is part of a performance system. Actual speech may include: ○ Interruptions, non perfect sentences ○ Speech errors, hesitation, anticipation ○ Memory lapses, forgetting what we are saying etc. What Should Be the Object of Study? Competence, Performance, or both? The Object of Study The Generative Approach Focuses on developing internalised grammatical competence (phonology, syntax, semantics, etc.). Nativist researchers study I-language. Nativism Proposes that a significant portion of linguistic competence is innate, part of our genetic endowment. Emphasises Principles and Parameters of a Universal Grammar. Places less emphasis on language use and communication. Studying Competence Children’s competence is influenced by a performance system: ○ Limited memory and attention. ○ Articulatory immaturities (stuttering). ○ Word-finding difficulties. Researchers design experiments to minimise performance factors for studying competence. Experience-based Approaches Interested in how the environment contributes to language learning. Emphasise: ○ Frequency of input. ○ Errors as omissions. Focus on E-language: observable performance. What Do Speech Pathologists Study? Interest in: ○ Competence? ○ Performance? ○ Are both important? The Nativist Approach 1950s Challenge: Chomsky questioned the idea that language is solely a learned behaviour. Biological Basis: Proposed that key aspects of language knowledge are rooted in our biology (part of the human genome). Universal Linguistic Principles: Claims that some linguistic principles are universal, applicable to all languages. Universal Grammar Theory: Chomsky’s theory became known as the theory of Universal Grammar. ○ Chompsky is particularly interested in syntax; he doesn’t mention anything about phonology or second language learning etc Language Acquisition Device (LAD) Initial State: Refers to the innate mental capacity for language. Final State: Represents the adult speaker’s mental grammar, which includes: 1. Innate Knowledge of Language: Linguistic principles or constraints. Parameter setting that explains language variation. 2. Learned Aspects of Language: Lexicon (vocabulary). Language-specific rules. Pragmatic knowledge (contextual language use). Studying the Final State Theoretical Linguists: Analyse grammars of various languages to identify core principles of Universal Grammar. Experimental Linguists: Investigate whether these principles are present early in child language acquisition. Importance of Stages Stage Analysis: Examining the stages children go through helps reveal properties of Universal Grammar. Continuity Hypothesis: Children’s non-adult stages should reflect characteristics of other human languages. Stages of Language Development Language Faculty Components: ○ Phonology: Sound systems of languages. ○ Syntax: Structure and rules governing sentence formation. ○ Lexicon: Vocabulary of a language. ○ Semantics: Meaning of words and sentences. ○ Working Memory Buffer: Temporary storage for language processing. Modularity of the Language Faculty Separate Module: Nativists view the language faculty as a distinct module, isolated from other cognitive systems. Computational Properties: The language faculty has its own processors that manage information flow. Cognitive Independence: Language understanding is not influenced by general knowledge (e.g., "Mice chase cats" understood despite contradicting common sense). Language and Cognition Anecdotal Observations: Children may show advanced language skills while lagging in other cognitive abilities (e.g., complex sentences by age 3, but struggles with physical tasks). Components of the Language Faculty Hardware/Mechanics: ○ Syntactic Categories: Inventory (e.g., S, NP, PP). ○ Rule Formation: Ability to create grammatical rules. ○ Transformations: Capability to manipulate sentence structures. Principles/Constraints: ○ Universal Principles: Apply across all languages. Parameters: ○ Cross-Linguistic Variation: Parameters explain differences among languages. Core Principles of Universal Grammar Universal List: Every child is born with the same core principles. Constraints on Hypothesis Space: Core principles limit the range of hypotheses children can make about language, guiding their learning process. Speeding Up Acquisition: By constraining options, these principles help children avoid incorrect language paths, facilitating quicker language acquisition. Principles Limit Errors Error Prevention: Core principles restrict the types of hypotheses children can generate, reducing logically possible errors. Unlearning Mistakes: If children were to make numerous errors, they would need to "unlearn" them to align their grammar with that of adults. Corrective Feedback Limited Correction: Research (e.g., Brown and Hanlon, 1970) indicates parents rarely correct grammatical errors but focus on the truth of utterances and phonological mistakes. Implication of Inherent Knowledge: Chomsky argued that if children aren’t corrected, it suggests they possess innate knowledge of certain language principles from the start, helping them avoid specific errors. The Ensuing Debate Behaviourist and Interactionist Views: Some researchers argue that parents play a significant role in language learning by providing experiences (e.g., recasts and expansions) that aid grammar construction. Parental Input: According to these approaches, parental feedback helps children learn from their mistakes, influencing their language development. Knowledge of Language Pre-Wired Knowledge: Evidence suggests that certain grammatical principles are innate and not learned through experience. Types of Knowledge: 1. Sentence Structure: Understanding how sentences are divided into parts. 2. Ambiguity: Recognizing which sentences can have multiple meanings. 3. Syntactic Constraints: Knowing which sentences cannot be formed in certain ways, despite understanding their intended meanings. 4. Semantic Restrictions: Awareness of what specific sentences cannot mean, even if their structure appears correct. A Constraint on Meaning: Principle C 1. Examples: ○ Sentence 1: "The Ninja Turtle said he likes pizza." Meaning 1: The Ninja Turtle said the Ninja Turtle likes pizza. Meaning 2: The Ninja Turtle said someone else likes pizza. ○ Sentence 2: "He said the Ninja Turtle likes pizza." Meaning 1: The Ninja Turtle said the Ninja Turtle likes pizza. Meaning 2: Someone else said the Ninja Turtle likes pizza. 2. Principle C: ○ An innate constraint that prevents language learners from hypothesising impossible meanings. ○ Should be evident across languages that include pronouns and names. ○ Operative in child language from an early stage. Linguistic Principles are Abstract Pronoun and Name Placement: ○ Configuration A: Does not allow co-reference (e.g., the pronoun cannot refer back to the name). ○ Configuration B: Allows co-reference. Principle Overview: ○ These principles are not derived from experience or observable sentence forms; they go deeper than input suggests. The Diversity of Language Universal Grammar vs. Surface Properties: ○ Universal Grammar applies universally but doesn’t explain the vast differences in language surface features. ○ Variation among languages is attributed to parameters, or "menu choices." Setting Parameters Input and Language Acquisition: ○ Children encounter parameters one by one as they develop their mental grammar. Initial State: ○ Children are born with knowledge of important linguistic options present in world languages. Parameters: ○ Hypothesised as binary decisions (A or B): P1: Do verbs precede or follow objects? P2: Omit subjects or use full subjects? P3: Position question words at the front or leave them in their original location? The Switch Metaphor Binary Choices: Parameters can be visualised as switches that yield different language structures. Different Settings: ○ Combinations of parameters result in distinct languages. An Analogy with Baking Cooking Analogy (Mark Baker, 2001): ○ Just as baked goods use basic ingredients (flour, eggs, etc.), children start with Universal Grammar as their foundational "ingredients." ○ Choices made in the recipe (e.g., mixing order, ingredient type) reflect the parameter setting process in language acquisition. Adjusting the Recipe Parameter Setting: ○ Children make specific choices in language (e.g., verb placement, subject usage) similar to choosing ingredients in cooking. The Nativist View Facilitating Language Acquisition: ○ Universal Grammar guides children through the language learning process, limiting problematic hypotheses. Need for Input: ○ Children require meaningful interaction with caretakers for effective language acquisition, not just any input (e.g., TV isn’t sufficient). Special Input: Child Directed Speech (CDS) Nativist Perspective: ○ Acknowledges that parents use special speech styles to engage children, but denies that this alone constructs grammar. Contrast with Behaviorism Behaviorist View (Skinner): ○ Language is learned like other behaviours, using the same mechanisms. Piaget’s View: ○ Cognition and language interact, differing from the strict behaviourist perspective. Behaviorism Prediction Skill Development: Just as children progress in maths and music, they may develop varying language skills. Final State: All children ultimately reach an equivalent final state in language acquisition. Caveat: Learning and education can lead to differences in vocabulary, writing skills, and rhetorical abilities. Advantages of Behaviorism 1. Parsimony: ○ The theory is simple; the same learning mechanisms apply across different domains (language, maths, music). ○ Avoids unnecessary complexity by not introducing multiple mechanisms for each domain. 2. Common Sense Appeal: ○ Aligns with everyday observations—parents naturally teach their children. 3. Parental Teaching: ○ Parents and caretakers play a significant role in language development. ○ Children pay attention to input and strive to mimic what they hear, suggesting a close relationship between exposure and language output. Key Questions Do Children Only Produce What They Hear? ○ Investigates whether children's utterances are solely based on the sentences they have been exposed to. Operant Conditioning (B.F. Skinner) Voluntary Behaviour: Language, termed "verbal behaviour" by Skinner, is voluntary and does not need to occur. Reinforcement: ○ Verbal behaviour can be strengthened or weakened based on consequences. ○ Approval (e.g., smiles, affirmations like “That’s right”) reinforces correct language use. ○ Disapproval serves as a generalised punishment, discouraging incorrect language use. Summary of Skinner's View Language acquisition is shaped by the selective reinforcement of syntactically correct utterances through parental feedback. Children learn to produce syntactically correct forms by responding to reinforcement and punishment. Quote from Skinner: “Syntactically correct utterances come to prevail over syntactically incorrect utterances through the selective administration of signs of approval and disapproval.” (Science and Human Behavior, 1953, p.78) Experiments with Pigeons Positive Reinforcement: Food serves as a reward for behaviour. Free Operant: The initial pecking of the key. Conditioned Operant: Continuous pecking of the key after learning the connection. 1. Setup: A hungry pigeon is placed in a Skinner box. 2. Initial Reaction: The pigeon pecks the key by chance and receives food. Initially, it does not associate the pecking with the reward. 3. Learning: Eventually, the pigeon pecks continuously, demonstrating that it has learned the connection between pecking and receiving food. Behaviorism and Language Learning From Specific to General Learning: The question arises of how children move from learning language in specific contexts to broader applications. Stimulus Generalisation Definition: Stimulus generalisation allows a response to be evoked by stimuli similar to the original. ○ Example: A child learns to say "ball" when seeing a green tennis ball and later applies "ball" to other similar objects, like various balls. Potential Problem: ○ While generalisation is helpful, it can lead to errors. ○ For instance, a child might mistakenly call an orange a "ball." Corrective Feedback: ○ If a parent corrects the child by saying, “That’s an orange, not a ball,” the child can learn from this feedback. Generalisations and Sentence Patterns Examples of Sentence Patterns: ○ a. √ John painted the red barn. ○ b. √ John painted the barn red. ○ c. √ John saw the red barn. ○ d. * John saw the barn red. Correction Challenge: ○ Unlike naming errors (e.g., calling an orange a ball), correcting incorrect sentence structures requires specific guidance (e.g., “You can’t say it that way”), which may be harder for children to grasp. Reinforcement and Punishment Reinforcement: ○ Occurs when a stimulus increases the likelihood of a behaviour in similar future contexts. ○ Example: Giving a child a cookie for polite behaviour reinforces that behaviour. Punishment: ○ Occurs when a stimulus decreases the likelihood of a behaviour happening again. ○ Example: Taking away a cookie when a child throws a tantrum reduces the likelihood of future tantrums. Creativity of Language Property of Recursion Definition: Recursion allows for the repeated use of phrases, enabling the creation of complex and novel sentences. Functions of Recursion 1. Novel Sentence Production: Enables us to understand and produce se

LING2214 Psycholinguistics PDF

Document Details

Tags

Related

Summary

Full Transcript

Upgrade to continue