Language Production and Speech Errors Lecture Notes PDF

Language 2: Language Production and Speech Errors Issues: (a) How is language produced? (b) How do speech errors shed light on language production? Speaking seems easy Fast average speech rate of 150 words/minute, up to 5.6 syllables/second Accurate errors in word selection about 1/1...

Language 2: Language Production and Speech Errors Issues: (a) How is language produced? (b) How do speech errors shed light on language production? Speaking seems easy Fast average speech rate of 150 words/minute, up to 5.6 syllables/second Accurate errors in word selection about 1/1000 words errors in sound ordering about 1/2000 sounds Normally effortless and automatic But it should be difficult because: Many types of information involved Need to produce sentences never used before Have up to 100,000 words in vocabulary to choose among Little time or memory to prepare far in advance Rapid sequence of precise movements for articulation Speech errors Language production research started with study of speech errors e.g., Merringer & Mayer (1985) Get information about how system works normally from how it works when it falls apart Sources of speech errors Corpora Listen for errors and document those that occurred But Can't control factors that influence errors Biased to miss some kinds of errors: people often don't notice when a speaker makes an error. Experiments Researcher tries to make errors happen more frequently than usual But may make task too artificial Speech errors Tuesday we talked about the structure of language, how meaning is related to language. Then I talked about some experiments that examined how people understand language: how is language comprehended? Today, I'm going to shift the focus to the production of language. How is it that people speak or write? In other words, how do they translate the meaning they intend to express either into a vocal utterance or into the movement of a pen (or a keyboard)? Why study speech errors It's difficult to control what a person says or writes, and much easier to control what a person hears or reads. The reason speech errors have been helpful is that language is made of rules, and it turns out that speech errors happen when rules are applied incorrectly. So that by seeing what speech errors occur, you can see what rules there are. Types of speech errors Mistakes Errors in choosing the goal or the plan to achieve it. Slips Errors in carrying out the intended plan. We will focus on slips of the tongue. Slips of the tongue are unintended, non habitual deviation from a speech plan. The first step is to characterize what types of errors can and do occur. The reason is that just knowing which slips are possible and which are not has a lot of implications for theories of speech production. There are several different possible levels of speech error, which correspond to the size of the linguistic unit (level) involved in the error. Dell describes all of these in his coursepack reading. Three are most common, however, and we'll focus on those today. Errors and The Size of the linguistic unit Sound errors (phoneme) Sound errors are errors in which a single sound is involved. One type, for example, is when a sound in nearby words is exchanged. York library --> lork yibrary Snow flurries --> flow snurries Can be a single phoneme, a consonant cluster (as above), or a vowel-consonant combination (rime - the vowel plus remaining consonants in a syllable). 70-90% involve a single phoneme Morpheme errors Morpheme errors occur when the unit involved is the morpheme. An example is when morphemes are exchanged in two words. self-destruct instruction --> self-instruct destruction thinly sliced --> slicely thinned Word errors writing a letter to my mother --> writing a mother to my letter. Nature of the disruption Errors also differ in what exactly happens to the particular linguistic unit involved in the error. Exchange (Reversal, Transposition) An exchange is simply when two words appear, and the unit for one word is exchanged with a unit from another word. Probably the most common type of speech error. (All of the examples mentioned earlier are exchanges.) "Heft lemisphere" --> "Left Hemisphere" Spoonerism One particular kind of reversal is called a Spoonerism. This is the type of error where you exchange one sound for another. Spooner was the Dean of New College of Oxford in England, and he was well known for making exchanges. The noteworthy thing about his exchanges were that that they created new words: "You have hissed all my mystery lectures. I saw you fight a liar in the back quad; in fact, you have tasted the whole worm." "Easier for a camel to go through the knee of an idol" "Queer old dean" for "Dear old Queen." "The Lord is a shoving leopard to his flock." Some of these are however considered fictitious. Anticipation An anticipation occurs when a linguistic unit from one word appears earlier in the sentence than it should. Phoneme: reading list --> leading list Word: sun is in the sky --> sky is in the sky Perseveration A perseveration occurs when the linguistic unit from one word appears again later in the sentence. phoneme: beef noodle --> beef needle word: class will be about discussing the text --> class will be about discussing the class. Addition (Insertion) to strain it --> "to strained it" some weeks --> "somes weeks" Deletion "I just wanted to ask that" --> "I just wanted to that" Substitution "after it is too late" instead of "before it is too late" "Liszt's second Hungarian rhapsody" --> "Liszt's second Hungarian restaurant." Blend tennis athlete - tennis player: tennis athler --- athlete + player --> athler call a taxi - call a cab: call a tab --- taxi + cab --> tab Evidence from collections Errors follow rules within a level Words: The Syntactic Category Rule One of the most salient pieces of evidence about speech errors is that when an error occurs, the error usually obeys the syntactic rules pertaining to the category of the linguistic unit. In particular, word errors obey the rules of words, so that nouns replace nouns, verbs replace verbs. In other words, when word errors occur they preserve the syntactic structure of the sentence. (That is, the phrase structure of the error is the same as the phrase structure of the intended utterance.) Never see "letter a writing to my mother" Sounds: Phonological rules When one sound is responsible for the error, it forms a word that is pronounceable. That is, even if the word that is formed is not really a word, it follows the rules of English phonemes.  Never see someone say tca instead of cat. Won't create a word beginning with "TL" Could say "slips of the tongue" as "stips of the lung" but would never hear "tlip of the sung" The consonant-vowel rule Consonants only replace or exchange with other consonants, and vowels only replace other vowels. The experimental study of slips of the tongue Problem with collections Most of the work on speech errors initially relied on people's collections of speech errors, and of course the problem with this is that it is not only very difficult to do, but the collections are subject to biases on the part of the experimenter, i.e. the experimenter may only be looking for certain types of slips. Even greater problem, perhaps, is that listeners don't detect slips of the tongue. Often tend to hear what was intended, rather than the actual mistake. Role of top-down processing. Tongue Twisters have been used to elicit speech errors. Laboratory method Present word pairs 1 per second. Subjects read word pairs. For some pairs, a signal is presented which tells the subject to say the word pair as quickly as possible Biasing pairs: 3-5 pairs preceding some word pairs bias the subject to make a slip. (see slides) Findings: With experimental control over slips, it's possible to test specific hypotheses about speech errors that have to do with the interaction between levels. Errors at one level follow the rules of another level. Lexical (word) bias effect: interaction between word level and phonological level. Phonological errors that create words are more likely than errors that create nonwords. barn door --> darn bore (more likely than) dart board --> bart doard. Syntactic-phonological interaction Phonological errors tend to create phrases that are syntactically (grammatically) appropriate:  ker hane --> her cane (more likely than) ker hame --> her came. Sound-meaning interactions (Semantic/phonological) Phonological errors tend to happen more often for words that not only sound alike but are also similar in meaning. That is, sound errors occur more often when the resulting word has a meaning that is related to the intended word. "Oyster" substituted for "Lobster," "Let's stop" instead of "Let's start" Freudian slips? As you probably know Freud thought that slips of the tongue often expressed unconscious conflicts. Basically the idea was that when we let our defenses down, we involuntarily let unconscious thoughts slip out. Theoretical explanations of speech error patterns Models of speech production have been designed partly to account for these speech error data. Most models agree that there are: Semantic level- determines which words you will use. Word level- represents the words that are used. Sound level- specifies the sounds you will use. Each level governed by rules, e.g. phoneme rules. Speech editor One way of explaining these data is to say that the speech system composes the sentence, and checks the sentence before it is uttered. So you assemble the string of phonemes and then the output editor checks to weed out errors, but it can only check whether the utterance is a word. If the phonemes form a word, the editor lets the word be uttered, but if the phonemes are not a word, then the speech system starts over and produces a new set of phonemes. The non-word errors don't get uttered. Interactive activation approach (Dell) Doesn't require postulating a special editor. Automatically filters out non-word errors because of interactive activation. (Model is similar to McClelland and Rumelhart's model of word recognition). Levels of representation: Word, phoneme. At the time when you speak, the phonemes that are most highly activated determine what you say.  Top-down and bottom-up activation. Top-down is usually what produces a word. Example: Deal. (see slides) Error happens when the wrong phoneme is activated when you speak. Can happen for several reasons: Wrong phoneme can be active because spreading activation from the word you intend to say can activate words and phonemes that aren’t needed. Phonemes from previously spoken words can still be active Extraneous cognition and perception can activate unintended words and their associated phonemes. Word errors are more likely than non word errors (lexical bias) because word errors receive top- down activation from word nodes. When activation of a phoneme is already high (like from previous trial), a word error is more likely if that activation is reinforced by top-down activation. If there is a word, the activation can flow up and then back down again. Creates a positive feedback loop. With Dean -> Bean slip, two sources of activation: the top-down activation from Dean, and the bottom up activation from the B phoneme. The bottom-up activation from Bean combines with the other nodes to activate the word "Bean." The "bean" phonemes -- the ones that make the slip -- therefore can receive additional activation from the word node. Creates a positive feedback cycle - mutual activation flows in both directions. Makes it likely that Bean will be spoken instead of Dean. With Deal -> Beal, the phonemes for Beal don't receive from word nodes. So there's less competition between Deal and Beal. And there's no positive feedback cycle. Advantage over editor model -- it's a simpler explanation: no separate cognitive system is necessary to produce the effect.