Language Production: From Speech Errors to Syntactic Planning
Language Production: Speech Errors, Models, and Syntactic Planning
Correction on Spoonerism and Speech Errors
Spoonerism: Specifically refers to an exchange of initial consonants of words, such as "shoving leopard" for "loving shepherd." It is the most widely observed type of speech error in corpus data.
Correction from previous lecture: Spoonerism does not include all types of speech errors.
Other Types of Speech Errors: Speech errors are categorized into different types beyond spoonerism.
Foolium Slips (Freudian Slips): Thinking one thing but saying another, indicating multiple competing thoughts in the mind during speech.
General Categories (practiced in tutorials):
Shift: A segment moves from its correct location to another (e.g., "It certainly run out fast" instead of "it certainly runs out fast").
Anticipation: A segment appears earlier than intended (e.g., a "cup of cofee" becoming "a cuff of cofee").
Perseveration: A segment appears later than intended (e.g., "a phonological fool" becoming "a phonological feem").
Addition: An extra segment is inserted.
Deletion: A segment is omitted.
Substitution: One word or segment is replaced by another.
Blending: Two words or phrases merge into one.
Spoonerism and foolium slips are among the most frequently observed errors, while other types may be less frequent.
Generalizations from Speech Error Data
Evidence for Sentence/Syntactic Planning: Characteristics of speech errors shed light on how sentence structure is planned.
Content Words vs. Grammatical Elements:
Content words (nouns, verbs, adjectives, adverbs) and grammatical elements (function words like prepositions, articles, or morphemes like suffixes) rarely exchange with one another in speech errors.
Word exchange typically happens within a particular type of word (e.g., nouns exchange with nouns, verbs with verbs).
Example: "a weekend for manics" vs. "manics for weekends." You exchange "weekend" and "manics," not "weekend" with "for."
Implication: Content words and grammatical elements might be processed or stored separately during language production. If stored similarly, exchanges between these two categories would be more common.
Suffix Alterations and Sentence Stress:
Suffixes: Tend to be altered appropriately for their new linguistic environment, even when a word shifts position.
Sentence Stress: Remains largely unchanged despite word exchange errors.
Implication: Phonology is specified late in the production process, potentially after an error has been made, suggesting distinct processing stages for meaning/syntax and phonology.
Word Exchange vs. Sound Exchange - A Summary of Observations:
Word Exchange:
Can occur across phrase boundaries.
Involves words that belong to corresponding grammatical categories (e.g., noun with noun, verb with verb), usually content words.
Can span across a set of words (longer distance).
Example: "We sit around the song and sing files" (intended: "We sit around the files and sing songs").
Cognitive Implication: When a word exchange occurs, it suggests that the words of choice were already active in the mind when the sentence started being spoken. For instance, in the example, both "song" and "files" were activated, and their positions were swapped during articulation, implying early word activation and selection competing for output.
Sound Exchange:
Usually occurs within phrases (does not cross phrase boundaries).
Does not necessarily involve words of the same grammatical category.
Tends to span only one or two words (short distance).
Example 1: "You got a lot of pawns and pads" (intended: "pots and pans"). Here, and sounds were exchanged within the words.
Example 2: "A disordar of speech is strictly stricken is" (intended: "disorder of speech is strictly speaking is").
Cognitive Implication: Sound exchanges often happen due to similar linguistic environments surrounding the swapped segments (e.g., both preceded by vowels or similar sounds). These are often articulation mistakes rather than conscious planning errors.
Conclusion from Exchange Differences: Word exchange and sound exchange happen at different time courses and different stages of syntactic formation.
Word exchanges are made at a multi-phrasal level (related to sentence structure and planning) because they cross phrasal boundaries.
Sound exchanges are made at a single-phrase level (later than word exchange stage) because they occur within phrases.
Generalization: There is likely more than one level of syntactic representation during the planning process of language production.
Merrill Garrett's Model of Language Production
Overview: This model explains speech data and encompasses characteristics observed in speech errors.
Core Distinctions:
Major distinction between content words and grammatical elements.
Specifies that phonological processing occurs after syntactic processing.
Proposes two levels of syntactic representation.
Placement within General Language Production Framework: Garrett's model specifically details what happens in the formulation stage (after the message level and before articulation).
Stages of Formulation in Garrett's Model:
Message Level: Formulation of an event, message, or idea.
Functional Stage:
Selection of words from a mental lexicon based on meaning.
Formulation of the semantic and syntactic roles of these words (e.g., selecting a noun for a subject). Here, the abstract concepts for words are chosen.
Positional Stage:
Planning the order of the selected lexical items according to the syntax of the language (e.g., English subject-verb-object order).
This stage involves creating a surface phrasal frame, including functional words.
Phonological Stage:
Retrieval of phonological information for each word (e.g., the sounds that make up "dog").
Articulation: The physical production of the retrieved phonology (e.g., forming the sounds of "dog" with the vocal apparatus).
Evidence from Acquired Expressive Aphasia
Support for Garrett's Model: Investigations into acquired expressive aphasia provide empirical support.
Broca's Area and Expressive Aphasia: Patients with damage to Broca's area (primarily responsible for language production) often experience expressive aphasia. Symptoms are heterogeneous due to varied brain injury.
Double Dissociation: A key observation supporting Garrett's model, indicating separate language processing components.
Definition: Double dissociation occurs when two related mental processes are shown to function independently. If a brain injury impairs process A but not B, and a different injury impairs B but not A, then A and B are doubly dissociated.
Application to Garrett's Model: This supports the distinction between content word retrieval (functional level) and syntactic/positional planning (positional level).
Patient Type 1: Agrammatism:
Symptoms: Impaired production of syntax (difficulty forming grammatically correct sentences, often lacking verbs and function words) but relatively maintained ability to produce content words and convey meaning.
In terms of Garrett's Model: Difficulty progressing from the functional to the positional levels (struggle with syntactic ordering after word selection).
Example Patient Speech: Description of a kitchen scene: "In the stool, is it boy, is it landing down. The girl is laughing and cookie jar window, curtains, and out the garden, and trees, little grass…" (Mostly nouns, struggling with verbs and sentence structure).
Patient Type 2: Fluent Aphasia (Wernicke's Aphasia, though not explicitly named in transcript):
Symptoms: Fluent spontaneous speech (maintaining syntax and rhythm) but with little informational content and frequent use of incorrect or meaningless content words (neologisms or semantic paraphasias).
In terms of Garrett's Model: Difficulty retrieving meaning from the message to the functional levels (struggle with selecting appropriate content words).
Example Patient Speech: A patient speaking fluently but producing nonsensical responses like "End of the football," "It causes it comes in," or making many repetitive, vague statements and struggling to name objects or answer simple questions accurately, despite maintaining sentence structure.
Challenges to Garrett's Stage-Based Model
Blending Errors: Suggests that stages might not be strictly serial but can interact or occur in parallel.
Example 1: "It's difficult to verify" (blend of "validate" and "verify").
Example 2: "I'm making the kettle on" (blend of "making some tea" and "putting the kettle on").
Implication: These errors suggest that syntax and lexical items are processed in a more parallel fashion, leading to a blend when two potential structures/words are active simultaneously. This challenges a purely sequential, stage-based model where one stage completes before the next begins.
Mixed Word Substitution: Errors like "column" for "comma" suggest interaction between different levels of processing (e.g., semantic and phonological similarity influencing selection).
Lexical Bias Effect: The tendency for speech errors to result more often in actual words than non-words.
Problem for Garrett's Model: If many errors (like sound exchanges) are presumed to occur at phonological encoding, after lexical access (i.e., after the word has been selected), then why would there be a bias towards forming valid words at this later stage? A purely feed-forward, stage-based model struggles to account for this. It suggests a feedback mechanism or interaction where the lexicon can influence later phonological stages.
Overall: Blending errors and the lexical bias effect indicate that there might be interactions between the functional and positional levels (and potentially others) in Garrett's production model, rather than a strictly modular, feed-forward process.
General Inferences from Language Production Research
Focus of Psycholinguistic Investigations: Largely on the formulation stage, specifically how word choices are made and syntactic structures are assigned.
Data Sources: TOT state, speech error data, and aphasic speech patterns.
TOT State Implication: Supports a two-stage lexical retrieval process: a meaning selection stage followed by a phonological retrieval stage.
Word exchanges are characteristic of the meaning stage.
Sound exchanges are characteristic of the phonological stage.
Dual-Task Paradigm (Mediate Effect): Suggests that syntactic structure is planned rather than automatic.
If automatic, one wouldn't expect differences in error patterns or processing times based on sentence structure.
Requires a mental representation of sentence structure to be formed and planned, aligning with Garrett's positional stage.
Syntactic Planning and Preferences
Historical Emphasis in Linguistics: Formal linguistics, particularly under Noam Chomsky, placed significant emphasis on syntax as a core mechanism of human language.
Syntactic Priming: A phenomenon where an interlocutor's consistent use of a particular syntactic structure (e.g., passive voice) increases the likelihood of a naive participant also using that structure.
Example: A speaker consistently using passive voice ("John was kissed by Mary") primes listeners to produce more passive sentences themselves.
Implication: This suggests that syntactic structures are not just passively understood but actively influence subsequent production.
Verb Types and Syntactic Choices (Experiment 1):
Background (Comprehension): Sentences where the "experiencer" is in the subject position (as an agent) are generally easier to process than when the "experiencer" is in the object position (e.g., "John feared the dog" is easier than "The dog feared John").
Production Study Question: Do speakers prefer to produce specific active/passive structures based on verb types?
Methodology: English native speakers were asked to write sentences ( words or more) in response to single verbs.
Findings:
"Experience-Stimulus" Verbs (e.g., "cherish," "misunderstand," "suspect"): Speakers generated significantly more active sentences than passive sentences (e.g., only passive sentences).
"Stimulus-Experience" Verbs (e.g., "amuse," "outrage"): Speakers generated significantly more passive sentences than active sentences (e.g., passive sentences, a statistically significant difference from the first type).
Conclusion: Speakers actively make choices about syntactic structure based on the type of verb being used, supporting the idea of planned syntax.
Prosody and Syntactic Preferences (Coveney & Harley, 2013 - Experiment 2):
Contrast: Sentences with different underlying syntactic structures for the word "have."
Model Perfect Construction: "He could have worked at a shoe factory."
Verb Construction (noun phrase with 'have'): "He could have worked at the shoe factory."
Background (Comprehension): Reading times were faster for modal perfect constructions (one a and two a) than for their verb construction counterparts (one b and two b), suggesting a preference.
Production Study Methodology: university students (English and non-English speakers) were given scrambled words and asked to reorder them to form a grammatical English sentence as quickly as possible.
Dependent Measure: Time taken to reorder the words.
Findings (Native Speakers):
Time to resolve modal perfect sentences: approx. seconds.
Time to resolve noun phrase sentences (with 'have'): approx. seconds.
Conclusion: The time to resolve word order was longer for noun phrase sentences compared to modal perfect sentences. This indicates a syntactic preference in sentence production, consistent with comprehension data, further supporting that syntax is not automatic but planned.
Summary of Factors Impacting Syntactic Choices: Verb types and structural preferences (e.g., modal perfect construction preference).
Syntax, Universal Grammar, and Brain Representation
Universality of Syntactic Planning: The presented studies are largely specific to English. Different languages have different argument structures and ways of expressing concepts (e.g., Mandarin Chinese uses a passive marker "bei" and has a different word order for passive sentences).
The question of whether these specific syntactic planning patterns apply universally across all languages is complex.
Chomsky's Universal Grammar (UG): Proposes that human languages share common underlying principles (e.g., the existence of subjects and objects) but differ in parameters (e.g., word order like SVO, SOV, VSO, VOS).
UG suggests an innate, domain-specific capacity for language.
Counterarguments (Learning/Epigenetics): Propose that the linguistic input children receive from their environment is sufficient for them to extract syntactic patterns, without needing an innate 'organ' for syntax (Skinnerian perspective).
Genie Case Studies: Examples like "Genie" (a child deprived of linguistic input during a critical period) who never fully acquired complex syntax despite later exposure, support the idea of developmental time windows and neurological constraints on language acquisition.
Modern Neuroscience Insights: Brain imaging data currently show strong evidence for specific brain regions associated with phonology and semantics, but less reliable evidence for a dedicated region specifically for syntax.
Challenge to Nativism: This can be a "bummer" for strong nativist views emphasizing a dedicated innate syntax module.
Abstract Nature of Syntax: Syntax is highly abstract, making its direct neural representation difficult to pinpoint.
Evolving Theories: Theories of language, including those on syntax, must be falsifiable and evolve with new scientific evidence. This requires continuous investigation and critical evaluation of existing models and hypotheses, moving beyond static theoretical frameworks.