Chapter 9: Language

Comprehending and producing language

Comprehending and producing language

Something we do all day long when we listen to dialogue or conversations, writing paper

AI and Language

Researchers studying AI have found it extremely difficult to build computer systems that can understand language (spoken or written) as easily as a 4-year old can

Learning Language

Language acquisition is rapid, however a person can still take several years to become proficient

Language

Language seems to be a crucial cognitive ability so easily used that we typically overlook its complexity. Some language processes are bottom-up or driven by incoming data, whereas others are top-down or driven by the lister or speaker's expectations. Some language processing appears automatic, carried out without awareness or intention. Other language processing, is performed intentionally and with effort.

Processing Language

Constrained by other cognitive processes such as perception, attention, and memory.

Distinguishing between language and communication

Language is often used as a communication system, there are other communication systems that do not form true languages (i.e. bee dancing, birds)

Necessary Characteristics of natural language

• regular: governed by a system of rule (grammar) - productive: has infinite amount of things that can be expressed in it

Other characteristics of natural language

• arbitrariness - lack of necessary resemblance between a work or sentence and what it refers to - discreteness - system can be subdivided into recognizable parts

Structure of Language

phonology, morphology, syntax, semantics, pragmatics

phoneme

different sounds which can be used to distinguish language. the smallest unit of sound that makes a meaningful difference in a language

Study of phonology

study of ways in which phonemes can be combined in any given language

Morphology

Putting the sounds together in some coherent way, identifying the meaningful units of language. Word endings, prefixes, tense markers, etc are critical parts of each sentence

Morphemes

the smallest meaningful units of language, and can be words

Syntax

refers to arrangement of words within sentences or more broadly to the structure of sentences - their parts and the way the parts are put together

pragmatics

necessary for conversation flow, listeners must pay attention and make certain assumptions, and speakers must craft their contribution in ways that will make the listener's job feasible. rules are the social rules of language - include certain etiquette conventions

Linguistic rules

• make up grammar of the language and define the way a language works

grammar

the set of rules for a language, does not refer to the "rules" of good English

knowledge of language

knowledge of the rules is not explicit (cannot articulate what all rules are/ not consciously aware of all of them), but implicit. We can often articulate prescriptive rules, which tells us how we should talk or write even though we may violate them. In contrast we find it hard to articulate the descriptive rules of English, which characterize which sentences are legal and which are not.

linguistic competence

the underlying linguistic knowledge that lets people produce and comprehend their language, not always evident in actual use or performance of language. Lapses of attention or memory, nervousness, etc can interfere causing us to produce ungrammatical sentence/ comprehend a sentence incorrectly

Linguistic performance (Chomsky, 1965)

linguistic competency only under ideal conditions - never achieved

phonetics

study of speech sounds and how they are produced

Phonology

study of systematic ways in which speech sounds are combined and altered in language

English

• about 40 phonetic segments (phones), but only certains ones are meaningful

Vowels

without obstructing the airflow, simply depending on the shape and position of the tongue and lips

Consonants

made by closing or almost closing part of the mouth, and differ in terms of "place of articulation" (where airflow obstruction occurs), and "manner of articulation", mechanics of how air flow is obstructed.

phonological rules

govern the way in which phonemes can be combined, which explain how to pronounce new words and how to pronounce prefixes and suffixes to word such as plural or past-tense endings.

Syntatic rules

govern the way in which different words or larger phrases can be combined to form "legal" sentences on the language. 2 requirements: be able to describe every "legal" sentence and should never be able to describe an "illegal" sentence

syntax

refers to arrangement of words within sentences or more broadly to the structure of sentences - their parts and the way the parts are put together

Kinds of Syntatic Rules

Chomsky proposed one set of rules, called phrase structure rules, (breaking down phrases to noun phrase , verb phrase -> verb -< particle -> noun phrase -> det and N), transformational rule - turn structures into other structures

Semantics

The branch of linguistics that studies the meaning of words.

Theories of meanings have to explain several things, at a minimum

Anomaly, Self-contradiction, Ambiguity, Synonymy, Entailment, and should also explain how we use word meanings to process whole sentences and discourses

Study of semantics involve

study of truth conditions of sentences and of the relations between sentences

truth conditions

simply the circumstances that make something true

Conversation (Searle)

In listening to another person we must understand not only the sounds, words, and structure of the utterances but also the kind of utterances -> assertive, directives, commissive, expressive, declarations

Searle's Speech Act Theory

Part of our job as listeners is to figure out which of the five types a particular utterance is and to respond appropriately

pragmatic understanding

Often exploited by advertisement to deliver hidden messages

Language comprehension and production

language must be transformed from raw input into meaningful representations. one of the first stages of this transformation is perceptual, then comprehension and the processing of discourse s

Speech Perception

understanding the speech of someone talking to you is usually quite easy unless it is in a foreign language or the speaker has a very marked speech impediment

George Miller 2 problems in speech recognitions

1. speech is continuous, boundaries do not correspond to word or syllable boundaries, this when it sounds as if someone is taking pauses between syllables and words it is illusory 2. that a single phoneme sounds different, depending on context - men and women different pitches, different accents, shouting, coxing, whispering

speech perception

although the actual acoustic stimulus can vary infinitely in its phonetic properties our perception of speech sounds is categorical, In processing speech sounds, we automatically without awareness or intention force the sound into discrete categories

Lisker and Abranson (1970): Categorical Perception of speech sounds

Used a computer to generate artificial speech sounds consisting of a bilabial stop constant (bah or pah sound - same consonant features and different voice onset time). They varied the voice onset time by computer from 20.15 seconds to 10.15 seconds - generating 31 syllables

Lisker and Abranson (1970): Categorical Perception of speech sounds Listeners

Participants "heard" only two sounds 'ba' and 'pa'. Any syllable with a voice onset time of 10.03 seconds or less was a 'ba', otherwise was a 'pa.

Lisker and Abranson (1970): Categorical Perception of speech sounds Findings

We pay attention to acoustic properties that make a meaningful difference in our language but ignore others

Categorical perception

Has been demonstrated for some nonspeech sounds, such as tones, buzzes, and musical notes played on different instruments.

infants and categorical perception (Eimas 1985)

Infants have shown that although very young infants can discriminate many, if not all of the sound distinction used in all the world's languages, the ability begins to narrow to just the phonemes in the infant's primary language when the infant is about 6 months.

Massaro and Cohen (1983) Visual information in the perception of speech

Examined the categorical perception of the stop consonants 'b' and 'd' (differ only in place of articulation). Participants heard 9 computer synthesized syllables which ranged in acoustic properties from a clear 'ba' to a clear 'da'. In the 'neutral' condition pparticipants heard syllables with no visual info. In two other conditions, participants heard syllables while watching a silent but synchronized videotape of a speaker pronouncing either 'ba' or 'da'

Massaro and Cohen (1983) Visual information in the perception of speech Findings

What the speaker appeared to be saying influenced what was heard, as shown with syllables in the middle of the 'ba'-'da' continuum were perceived slightly different as a function of what the speaker appeared to be saying relative to the perception reported in the neutral condition. This can be considered a context effect.

Perception of Speech Context Effect

Warren and co have demonstrated that in some cases people "hear" phonemes" that are not there. As he presented participants with a recording with coughing sound in the middle. Only 1/20 listers reported detecting a missing sound covered by a cough, and the one who did misreported the location. The other 19 showed phoneme restoration effect, because they apparently restore missing phonemes predicted by other linguistic information during the course of perception

Marslen-Wilson & Welsh (1978) Perception of Speech Context Effect

Required participants to "shadow" speech (repeat aloud), then introduced some distortion to the speech being presented.

Marslen-Wilson & Welsh (1978) Perception of Speech Context Effect Findings

Participants were often likely to restore the distortion to the proper pronunciation, specially if the word was highly predictable from the preceding context. Suggests that readers and listeners typically use the context of previous words in a sentence to predict the next word, and can even "mishear"/"misread" that word if it is presented in a distorted fashion.

Speech Errors

defined as instances in which what the speaker intended to say is quite clear, but the speaker makes some substitution or reorders the elements. Much of the data concerning speech error studies is observational as opposed to experimental, thus assertations about causations are problematic.

Garrett 1998 Classes of Speech Errors

Errors that showed meaning relations and errors which showed form relations. And argued that these two types of errors were distinct, as those that show similarities of meaning rarely involved similarities in form and vice versa.

Garrett 1990 Speech Errors Relation

The relative infreequency of word substitution errors showing both meaning and form similarities indication language production system processes information about meaning and information about form at different points in sentence construction. As if meaning and form processes operated simultaneously, then sentences in which both kinds of similarity ought to produce the most errors because there is a greater opportunity for errors, but this doesn't happen.

lexical decision task

can be used for understanding or recovering the meaning from task

semantic priming effect

people respond faster to an item if it was preceded by an item with similar meaning

Neely tyoes of processes for semantic priming

a fast-acting automatic spread of activation, and a slower expectance-driven process

Neely Semantic Priming Experiment

Included a prime target pairs (BIRD-SPARROW) that were related in meaning, which produced priming as expected. But also instructed participants that the prime word for Building would have a target named after a body part, and Body would be a prime word for a part of the building (switch instructions)

Neely Semantic Priming Experiment Methods

Held true to the instructions, but sometimes would violate the switch instructions and pair the primes and targets in a regular related way

Neely Semantic Priming Experiment Finding

when the target word was presented very shortly after the prime word (250 ms), the switch instructions did not matter and normal semantic priming was observed. Served as proof for automatic spreading activation. When there was more time between words (700 ms) participants priming based off switch instructions was observed, and for normal semantic priming it had a slower response time, served as evidence for slow-acting expectancy process

Just and Carpenter Reading Experiments

Often used computer-driven instruments to measure and record eye fixation on parts of the written text

Fixations

brief pauses that everyone makes as their eye scan text. Average fixation lasts about 250 ms

Reading

fixations and jumps, known as saccades between fixation, average saccades is about 10 to 20 ms

Just & Carpenter's model of reading

assumes that as soon as readers encounter a new word they try to interpret it and assign it a role (immediacy assumption)

Just & Carpenter's eye-mind hypothesis

interpretations of each word occurs during the time is fixated, there fore time spend on each fixation provides information about ease of interpretation.

Tannenhaus, Magnuson, Dahan, and Chambers model of eye movements

indicate access of stored words from memory during spoken language comprehension

Just & Carpenter Increase fixation duration factors

Word length, word infrequency, and syntactically or semantically analogous word

Just and Carpenter's Fixation study

presented college students with passages from magazins describing scientific inventions, technical innovations or biological mechanisms. Results suggested that more time is spent on the meaningful/semantically rich part of the test

Kintsch and Keenan Semantic factors Influence on Reading

the difficulty in two sentences of equal length lies in propositional complexity of the sentences (number of ideas conveyed).

Kintsch and Keenan Semantic factors Influence on Reading Experiment

Participants were asked to press a button after reading a sentence silently and then to immediately recall as much as they could.

Kintsch and Keenan Semantic factors Influence on Reading Findings

The more proposition a sentence had the longer it took for participants to read and comprehend. Also participants were more likely to recall "central" proposition, which are critical for the meaning of the sentence than peripheral ones. Which suggests that propositions are mentally represented in some sort of hierarchy with more central propositions at the top of the hierarchy

Haviland and Clark (1974): The given/new strategy

A pragmatic approach approach to processing sentences whereby listeners and readers divide sentences into two parts: the given and the new. Listeners first search memory for information corresponding to the given information, and then update memory by incorporating new information as an elaboration of the given

The given/new strategy Limitation

Only works if given information corresponds to some information in the listener's memory called the antecedent

The given/new strategy basis

the listener will have to make some connections- the obvious ones on their own. these connections are called bridging inferences, which bolster text coherence by linking ideas of concepts across sentences

Haviland & Clark's Given-New Experiment

Presented people with passages consisting of context followed by target sentences, The target sentence may have given information that exactly match antecedent information (from context), other times had to draw a bridging inference.

Haviland & Clark's Given-New Results

Took participants longer to read and comprehend the target sentence where they had to draw bridging inference.

Inference

can be for sentences far apart in text. Researchers have consistently shown that number of inferences readers make and their strength affects how well readers remember and understand what they read

Context in processing language John Bransford and Marcia Johnson Experiment

if you read a text then hide it and try to recall as much a possible, participants found this task very difficult and recall only included a few of the ideas. But if you were first provided context, participants were able to recall 8/14 distinct ideas, and without context only 3.6/14 ideas

Van den Broek and Gustafson conclusion on reading texts

1. mental representation is a construction by the reader that differs from and goes beyond info of the text itself -> people recruit their own background knowledge to draw inference to comprehend text 2. good representation is coherent -> structures such as schemata or story grammars are used to make information in a text fit together 3. reader's attentional resources are limited, thus they don't draw every logically possible inference, instead as needed to create coherence

Gricean Maxims of Conversation def

a pragmatic set of rules for people to use language appropriately or effectively specifically for conversations, as people must do more than say things

conversation

considered the essence of language, with both speaking and listening in their natural, universal states

Maxims of quantity

Make your contribution as informative as required Do not make your contribution more informative than is required

Maxim of quality

Try to make your contribution one that is true (Do not say things that are false or for which you lack adequate evidence)

Maxim of relation

be relevant

maxim of manner

Be clear. Avoid obscurity of expression. Avoid ambiguity. Be brief. Be orderly.

Violations of the Maxims

Produce noticeably odd conversations, although for the maxim of quality exceptions can be made, for ironic purposes, however there should be a signal letting the listener know the speaker is being ironic

Maxim of manner purpose

governs the way we chose to construct conversation contributions, and how you should speak appropriately to your listener and context (think how you talk to an interviewer vs. friend), and requires that you at least attempt to organize what you say before you begin speaking