SPE1103 2024 Week 9 Lecture Notes - Language Processing Comprehension PDF

Language Processing Comprehension  Learning Objectives × Understand key concepts and experimental methods in psycholinguistics × Understand basic processes involved in the perception and comprehension of the speech signal. × Understand the terminology and processes of the recepti...

Language Processing Comprehension  Learning Objectives × Understand key concepts and experimental methods in psycholinguistics × Understand basic processes involved in the perception and comprehension of the speech signal. × Understand the terminology and processes of the receptive side of the Levelt model. 2  Language – but how?! × Language comprehension is very fast and automatic. × We understand an utterance as fast as we hear or read it. × This is a great deal of work to be done in a very short time! 3  Consciousness × We are not conscious of the complicated processes we use to understand speech any more than we are conscious of the complicated processes of digesting food and utilising nutrients. × Psycholinguistics – aims to describe the processes people normally use in speaking and understanding language. 4  How do we ‘understand’ language? × Let’s remember what language is: × Arbitrary – sounds need to be ‘de-coded’ × Rule-based – morphological and syntactic rules can be used to organise and interpret incoming information × Discrete (made up of many ‘pieces’) need all the pieces to make the whole × Creative – meaning cannot be accessed as a whole (e.g. whole sentences) 5  Activity × Take the following steps to comprehension. × Place them in order from start to finish (1-7) × Its ok if you don’t know – have a go × Check your order as we go through the lecture 6  Step 1 Ear detects the incoming speech stream. 7  Speech signal × A sound = disturbance in the position of air molecules. × Vocal fold vibration = small pulses/variations air pressure × Speed of variations= the pitch (fundamental frequency) × The magnitude of the variations = the loudness (amplitude) × Acoustic phonetics = speech sounds, all of which can be heard by the normal human ear. 8  Auditory discrimination × Detection of the incoming speech stream depends upon: × Adequate hearing mechanisms × Intact auditory pathways of the brain × Attention and habituation – focus on the speech stream and ‘screening out’ of other incoming stimuli 9  Step 2 Word segmentation 10  The Segmentation Problem × Speech is a continuous signal. × Sounds overlap and influence each other. BUT WE THINK WE HEAR… × discrete units, such as words, syllables and phonemes. Why? = ‘segmentation problem’ 11  Normalisation × When we hear speech we ‘normalise’ it. × We account for differences in pitch, loudness and speed. 12  Sound Perception × People don’t perceive all sounds the same! × Speakers of English can perceive the difference between /l/ and /r/ = they affect meaning. × Speakers of Japanese usually cannot. They are allophones. × These perceptual biases develop during the first year × Prior to 9 months we can detect sounds of all languages 13 ‘Finding’ words in the speech stream × No clear pauses between words – so how do we find them? × You (consciously or unconsciously) conduct a search of your mental lexicon for the phonological strings you decide are possible words. × This process is known as ‘lexical access’ or ‘ word recognition’. 14 ‘Finding’ words in the speech stream × If you heard someone say × “ The cat chased the rat” – × You would do a lexical “ look-up” of ‘cat’, ‘rat’ ‘chased’ × The segmentation and search of these ‘words’ relies on knowing the: × phonology, grammatical morphemes and syntax. 15 Lexical access – top down or bottom-up? × Remember how speech comprehension is fast? × Its really fast. We can understand spoken language at a rate of 20 phonemes per second. × Is there a trick? Is this magic? × Kind of… we have 2 different systems to help: × Top-down processing × Bottom-up processing 16  Bottom-Up Processing × Comprehension is done step-by-step × Hear sound = identify word= check lexicon=interpret phrase. × The speaker waits until he or she hears ‘the’ and ‘boy’ and then constructs an NP, and then waits for the next word and so on. × Allows you to interpret something unexpected, or to decide that you don’t know something 17  Top-Down Processing × We ‘predict’ what’s coming next × Use morphological, semantic and syntactic information to make an ‘educated guess’ × Makes processing a lot quicker – filters out the mess and noise × Evidence is shown by: × Noise distraction × Shadow tests × Speech signal obstruction 18 Evidence for Top-Down Processing × People are asked to identify spoken words in a noisy environment. × More errors = × isolated words vs sentences × words given in nonsense sentences × words given in ungrammatical sentences. = Sentences ‘help’ us identify words 19 Evidence for Top-Down Processing × ‘Shadow’ sentences = people are asked to repeat each word of a sentence immediately as they hear it, as fast as they can. × People often produce words BEFORE hearing in anticipation of the input. × Based on their knowledge of syntax and semantics for the sentence so far… × “ You walked up the beautiful X” 20 Evidence for Top-Down Processing × People were given sentences in which some part of the signal is removed with a cough or buzz × People ‘think’ they hear the sentence as complete × They have difficulty saying exactly where in the word the noise occurred. × They ‘correct’ it when they are listening! 21  Video! × https://www.youtube.com/watch?v=Cz vgf-Xc-A4 22  Step 3 Lexical Activation 23  What is ‘lexical activation’? × The process by which we obtain information about a word from our mental lexicon. × Theories often relate to how we think we ‘store’ word knowledge 24 Review: What do we think is in our lexicons? × We think each word is linked to: × Its phonological form × Its potential morphemes There may be more × Its potential syntactic class than 1 of these! × Its meaning or concept 25 Review: What is stored for each word? /r/ 26  Review: How is our lexicon organised? × Lexicon thought to be organised as a semantic network, like a web of interconnected elements. × The elements are concepts or nodes that are connected through having relations with one another. 27 How do we ‘access’ the right word? × Thought to be through ‘spreading activation’ × We activate many words which might ‘fit’ × Activation begins with a single node and spreads throughout the network. × Eventually based on context we select only 1 × The ‘word’ which fits the: × Sounds we hear × Meaning of the sentence overall × Meaning of the discourse overall 28  How do we ‘know’? × We don’t…for sure. × But we have an idea based on: × Slips of the ear × Lexical decision tasks × Semantic priming 29  Slips of the Ear × When someone makes an error of comprehension × Tell us a lot about how language is processed × “Mine is a long and sad tale!” said the Mouse, turning to Alice, and sighing. × “It is a long tail, certainly,”’ said Alice, looking down with wonder at the Mouse’s tail; “but why do you call it sad?” × What does this example tell us? 30  Slips of the Ear × When someone mishears a word, it is typically a whole word. 85% of mishearings involve single words. × What does this tell us? × Which one of these is right? “Let the sky fall, when it “Make a trifle, make a crumbles crumble We will stand tall Build my cake tall And face it all together” And we’ll eat it all together” 31  Lexical Decision Tasks × Do you know this word? OR Is this a word? (spoken orally) × This can be timed (e.g. reaction times). × Longer = more processing, less good ‘access’ to the word × We are quicker for commonly used words – spoken and written (e.g. car vs fig). 32  Semantic Priming × When we ask someone to make a lexical decision. E.g. ‘Palm’ is this a word? × Having just previously asked them to make a lexical decision about ‘hand’ = quicker reaction time to ‘palm’ Called ‘Semantic Priming’ × Tells us about how we store words – what does it tell us? 33  Semantic Priming × After you hear the sentence × ‘“ The gypsy read the young man’s palm for only a dollar”. × ‘Palm’ – primes the word ‘hand’ but also ‘tree’ – why? × What does this tell us about how words are retrieved? × Evidence for bottom-up processing × This priming for unrelated meanings only lasts around 250 milliseconds 34  Complexity of Meaning × Meanings can be complex & individual - linked to our world knowledge. × Imagery and emotion are activated by auditory information × Words can also have different connotations to their denotation (dictionary meaning) Bossy -- Dominant Nag -- Remind Cheap -- Inexpensive Nosy -- Inquisitive Difficult -- Challenging Out of date -- Time-tested 35  Morphological Priming × Measures of brain activity show priming to pairs of verb forms such as teach/taught during the early stages of lexical access. × What does this tell us about the way we store words? 36  Phonological Priming × Significant priming effects were obtained when the prime and target words shared phonological information (word initial and word final) × What does this tell us about the way we store words? 37 Break!! Step 4 Parsing 39  Parsing × Parsing = figuring out the syntactic and semantic relations among the words and phrases in a sentence. × Listeners ‘build’ a phrase structure representation of the sentence as they hear it. × Each incoming word is assigned a grammatical category and ‘attached’ to the tree being constructed in the mind. 40  Parsing × Influenced by: × The rules of grammar/syntax × The linear presentation of a sentence. E.g. what comes first × Intonation and stress × Word knowledge 41  But HOW do you know this? × Again we don’t…for sure. × But we have an idea based on experiments on: × Shadowing × Garden Path Sentences – eye gaze and time × Intonation hints × Memory limitations 42  Shadowing Tasks × Shadowing tasks = repeat what you hear as rapidly as possible. × Good shadowers: × Very short delay of 0.3 of a second × correct speech errors & missing morphemes × don’t remember the errors they corrected × corrections are more likely if they are predictable What does this tell us about language parsing? 43 Processing Times & Eye Gaze Tracking × Written sentences are given to subjects × “ Read this sentence, tell me when you understand it” × Length of time taken, eye gaze = how we process sentences × ‘garden path’ sentences = show how we process in real time 44  Garden Path Sentences × We don’t wait to understand a sentence until it is done × We are predicting what comes next and building structure as we hear or read it. 45  Garden Path Sentences × The old man the boat * *Not a typo! × Sounds wrong huh? You probably read it a few times and your eyes jumped back and started to read it again. × You are using some assumptions (based on frequency): × the subject comes first × ‘old’ is an adjective, usually used with ‘man’ × ‘man’ is usually a noun, not a verb. 46  Activity – Garden Path Sentences Read following garden path sentences and: 1. identify how the actual meaning is different to what you first assumed. 2. identify what assumptions you made. Garden Path Sentences: × The cotton clothing is made of grows in Mississippi. × The sour drink from the ocean. × The man who hunts ducks out on weekends. × When Fred eats food gets thrown. × Fat people eat accumulates. 47  Intonation and Stress Helpful Hints × Read the garden path sentences again - try to use intonation to help make the meaning clear. × Useful huh?! Stress and intonation provide some clues to syntactic structure. × E.g. ‘ He lives in the White house’ versus ‘He lives in the White house’ – can be signalled by differences in their stress patterns. 48  Parsing Hints × We seem to interpret incoming speech using 2 ideas to help us: 1. Minimal attachment – build the simplest structure (usually not embedded) The horse raced past the barn fell. 2. Late closure – attach incoming material to the immediate phrase. The Dr said the patient will die yesterday. You want to put ‘yesterday’ with ‘will die’ rather than ‘the doctor said’ × These two hints help us process syntax quickly, but result in garden pathing 49  No Real-World Override × Which of these sentences led you more down the garden path? 1. The performer sent the flowers was very pleased 2. The florist sent the flowers was very pleased. × Actually both! Our ‘real world’ knowledge doesn’t help. Grammar overrides 50 it.  Memory Limitation This is the dog that worried the cat that killed the rat that ate the malt that lay in the house that Jack built Versus Jack built the house that the malt that the rat that the cat that the dog worried killed ate lay in Embedding has limits! 51  Step 5 Avoid Ambiguity & Comprehend 52  Lexical Ambiguity × Example ‘pear/pair’. × Both are activated, which one is selected? × Frequency of the words (high vs low) and the context (meaning of the utterance & discourse so far). Example: × Your friend says to you ‘I am hot’.. × Meaning of ‘hot’ activated first will depend on which is most common for the individual & the speaking context. How would it vary across generations? How would context influence the interpretation? 53 Structural Ambiguity Both meanings are activated. Selection against is based on: Syntactic frequency Environmental context Linguistic context 54  Structural Ambiguity × Worksheet 55  Stage 6 Update the discourse record. 56  What is discourse again? × Discourse – communication, usually taking place over more than 1 sentence. × Examples – conversation, debate, speech, argument, lecture, instructions, story-telling × To comprehend discourse, you need to reflect on and remember the whole transaction. × For ongoing relationships you need to reflect on and remember your entire discourse history 57 Reflecting on the discourse record 58  Example – debate rebuttal × Rebuttal requires debaters to listen to what is being said by the other side and respond to their arguments. × Requires precise comprehension and memory 59 Models of Receptive Language Levelt Stackhouse & Wells 60 Review: What are models of language? × Theoretical representations of how language is processed by the brain × Many different models – each with strengths and weaknesses × Two examples: Levelt and Stackhouse and Wells (named for their authors) × Allow us to hypothesize areas of impairment and treatment 61 Models of Language Processing 62  Levelt Model Acoustic-Phonetic Processor × Requires adequate hearing and perception × Identification of speech vs nonspeech × The phonetic information (e.g. individual phonemes) are held in working memory while they are interpreted using the parser × Prosodic information is used to organise the incoming speech 63  Levelt Model Parser × Phonological decoding – i.e. of words from the speech stream. Activation of words in the lexicon based on sounds × Words are activated in the lexicon- linking to phoneme, morpheme, syntactic and conceptual info × Grammatical decoding – hearing each word activates a syntactic structure – we jump ahead! × Everything is bi-directional 64  Levelt Model Conceptualiser × Looks after the discourse record and monitors your own speech × The parsed message is then interpreted in the context of the discourse record × This is bi-directional – the discourse record helps you work out the meaning × Also allows you to work out the inferred intention if its different to the literal (e.g. sarcasm) 65  Stackhouse & Wells Activity × Match the task for each stage to the model 66  What To Revise Identify and describe the steps of auditory comprehension, starting from auditory input to discourse comprehension Identify the areas of the Levelt model, and what happens at each area for receptive language – including the acoustic/phonetic processor, the parser and the conceptualiser. What evidence exists for how we store and process words using our brains? 67