hiseht

Linguistics & Cognitive Science: An Overview

1. Cognitive Science and the Study of Human Intelligence

• Cognitive science investigates human intelligence through an interdisciplinary approach, integrating linguistics, psychology, neuroscience, and, computer science.
  

• A central focus is on understanding language as a cognitive ability:

  • Language is viewed as a set of linguistic capacities enabling real-time comprehension and production.This includes exploring how individuals acquire language, process it, and utilize it in communication, thereby shedding light on the intricate connections between thought and language.

  • These capacities are supported by mental representations and computational processes.
    

2. Real-Time Language Processing

Language comprehension and production involve:

• Adjectival Modifiers: Adjectives modifying nouns (e.g., "red apple").

• Resultative Constructions:

  • Adjective denoting the state of the noun resulting from a verb (e.g., "She painted the wall red").

• Object Depictive Constructions:

  • Adjective denoting the state of the object (noun) when the verb's action occurs (e.g., "He ate the meat raw").

• Subject Depictive Constructions:

  • Adjective denoting the state of the subject (noun) during the verb's action (e.g., "She danced happy").

3. Linguistic Mental Representations

• Nature of Representations:

  • Mental representations are a construct that explains purposeful behaviors such as reasoning, holding beliefs, executing plans, body movements, and language composition and production.

  • MRs are mental stats that bear a lot of information. That mental processes use to encode, transform and store

  • Ex) Column Arithmetic(Old School Math)

• Mental Representations are not specific to language

4. Neuronal Bases of Mental Representations

• Language processing relies on specific brain regions:

  • Broca's area: Linked to syntactic and morphological processing.

  • Wernicke's area: Associated with semantic processing.

  • Temporal and frontal lobes: Coordinate phonological and lexical access.

• Advanced imaging techniques (e.g., fMRI) reveal neural networks underlying linguistic computations.
  

5. Differentiating Cognitive Psychology/Science from Behaviorism

• Behaviorism:

  • Restricted explanations to observable stimulus-response relations.

  • Dismissed the role of internal mental states.
    Believed all behavior is stimulus equals response

• Cognitive Science:

  • Embraces mental representations and processes as central explanatory constructs.

  • Investigates how the mind encodes, stores, and manipulates information.
    

6. Theories and Models in Cognitive Mechanisms

• Cognitive science employs theoretical models to explain linguistic abilities:

  • Symbolic Models: Represent language as rules and structures.

  • Connectionist Models: Use neural networks to simulate learning and processing.

  • Probabilistic Models: Incorporate statistical patterns in language acquisition and use.
    Marr'S three levels: computational, representational, implementational

  • modeling stroop Task:

  • WEAVER: Word Encoding by Activation and Verification: a computational model designed to explain how speakers control and plan the production of spoken word

  • Reading a word that is a color but the font is a different color.

  • Information is received from the associative network by spread activation the perceived color activates the corresponding concept mode. Activation then spreads through the network following a linear activation rule with a decoy factor. Each node sends a proportion of its activation to the nodes it is connected to and also to its lemma node.

• Reverse Engineering:

  • AKA backwards engineering or back engineering

  • Process through which on atttemps to understand through deductive reasoning how a previously made device, process, system, or software accomplishes a task with every little insight/experience on how to do so

  • 3 basic steps: information extraction, modeling, and review

  • Information extraction: practice of gathering all relevant information for performing the operation

  • Modeling: practice of combing gathered information into an abstract model

  • Review: testing the model to ensure the validity of the chosen abstract
    

Addressing Speech Perception Topics

1. Consonants vs. Vowels

• Speech perception involves a balance of 85% consonants and 35% vowels.

• Vowels are differentiated based in their height and blackness of the tongue and the rounding of the lips

• Consonants are differentiated by the place and manner of articulation and voicing

Classifying Consonants

• Voicing: Whether the vocal cords vibrate.

  • Voiced: Vocal cords vibrate (e.g., /b/, /d/).

  • Voiceless: No vocal cord vibration (e.g., /p/, /t/).

• Manner of Articulation: How airflow is obstructed.

  • Stops: Complete closure and release (e.g., /t/, /k/).

  • Fricatives: Narrow constriction causing turbulence (e.g., /f/, /s/).

  • Affricates: A combination of stop and fricative (e.g., /ʧ/ as in "church").

  • Nasals: Airflow through the nose (e.g., /m/, /n/).

  • Glides: Smooth transition (e.g., /w/, /j/).

  • Liquids: Minimal obstruction (e.g., /l/, /r/).

• Place of Articulation: Where in the vocal tract obstruction occurs.

  • Labial: Lips (e.g., /p/, /b/).

  • Interdental: Between teeth (e.g., /θ/ as in "think").

  • Alveolar: Tongue at the alveolar ridge (e.g., /t/, /s/).

  • Alveopalatal: Tongue near the palate (e.g., /ʃ/ as in "shy").

  • Velar: Back of the tongue against the velum (e.g., /k/, /g/).

  • Glottal: At the glottis (e.g., /h/).
    

Classifying Vowels

• Height: How high the tongue is (e.g., high /i/ vs. low /a/).

• Backness: How far back the tongue is (e.g., front /i/ vs. back /u/).

• Lip-Rounding: Whether the lips are rounded (e.g., rounded /u/ vs. unrounded /i/).

• Tenseness: Degree of muscle tension (e.g., tense /i/ vs. lax /ɪ/).
  

2. Nature of Speech Input

Physical nature of sound: Speech is carried by sound waves, with properties like frequency, amplitude, and duration.

Acoustics of speech:

Fundamental Frequency:

• basic pitch of voice

• Rate of which the whole vocal cords vibrate

Resonance: sympathies vibration

• rest of the vocal tract enhances some frequencies and inhibits others

• Frequencies that are enhanced or inhibited depends on vocal tract shape which depends on positions of articulatiors

• Produces formants: enhanced frequency bands, usually 3-4 formants in speech

Fourier’s Theorem: The cochlea in the ear separates complex sounds into their component frequencies, enabling neural encoding of frequency-specific information.of a series of sine or cosine terms, called the Fourier series, each of which has specific amplitude and phase coefficients, known as Fourier coefficients.

Cochlea as a Fourier Analyzer: The cochlea in the ear separates complex sounds into their component frequencies, enabling neural encoding of frequency-specific information using Fourier's theorem. This mechanism allows the auditory system to decode sounds into fundamental frequency components, assisting in the perception of speech and other auditory stimuli. The cochlea functions effectively as a Fourier analyzer by breaking down sound signals into individual sine or cosine components, known as Fourier series, each possessing specific amplitude and phase characteristics, termed Fourier coefficients.

Parallel Transmission of Phonemes: Phonemes overlap in time, requiring listeners to decode multiple sounds simultaneously.

3. Source-Filter Theory of Speech Production:

• Source: The vocal cords generate sound with a fundamental frequency (pitch) and harmonics.

• Filter: The vocal tract shapes this sound by amplifying certain frequencies (formants) based on its shape.

• Speech: speech production can be divided into two independent parts: sources of sound/signals such as the larynx. And filters that modify the source/systems such as the vocal tract
  

4. Formant Frequencies

• What are they? Formants are resonant frequencies of the vocal tract, visible as dark bands on a spectrogram.

• Phonetic properties of formants:

  • F1: Related to vowel height (low F1 = high vowels, high F1 = low vowels).

  • F2: Related to vowel backness (high F2 = front vowels, low F2 = back vowels).

• Reading a spectrogram:

  • Dark bands represent formants.

  • Transitions in formants indicate consonant articulation.

  • Steady-state formants: Vowels.

  • Formant transitions: Consonants.
    

5. Challenges to Speech Perception

• Segmentation Problem: Speech is continuous, with no clear boundaries between words or phonemes.

• Lack of Invariance Problem: Phonemes vary due to speaker differences, context, and coarticulation.

• Linearity Problem: Speech sounds are not strictly sequential but overlap due to coarticulation.

• segmenting the continuous acoustic signal into discrete regions relevant to identifying the phonemes the speaker intended

• Identifying phonemes when their acoustic signal varies across the linguistic context, speaking context, and speakers dialects, gender, ge, etc.

6. Design Solutions to Challenges

• Categorical Perception: Listeners group continuous variations in sound into discrete categories.

  • Voice Onset Time (VOT): The time between consonant release and vocal cord vibration; a key feature in distinguishing voiced and voiceless stops (e.g., /p/ vs. /b/).

• Top-down Strategies:

  • Phoneme Restoration: Listeners fill in missing sounds based on context (e.g., hearing a complete word despite a noise obscuring a phoneme).

  • Dealing with Mispronunciations: Context and expectations help recognize words even when mispronounced.

  • McGurk Effect: Visual cues can alter the perception of spoken sounds, suggesting that speech perception integrates auditory and visual information (analysis-by-synthesis model).
    
    

Lexical Access and Word Recognition

1. Lexical Representations

• Lexical representations in the brain are content-addressable, meaning they can be directly accessed using specific cues, such as the phonological, orthographic, or semantic properties of a word.

• Access occurs through a cue-driven, direct-access process, enabling rapid retrieval of word meanings during comprehension.
  

2. Time-Course of Lexical Access

1. Early Activation:

   • Multiple candidates are activated simultaneously based on the input's initial cues (e.g., phonemes or letters).

   • For example, hearing "ca-" may activate "cat," "car," and "cab."

2. Competitive Phase:

   • Inhibitory processes reduce the activation of less viable candidates as more input becomes available.

   • For example, hearing "cat-" narrows the candidates to "cat."

3. Selection Phase:

   • The best-fitting candidate remains active, while others are fully suppressed.

   • For example, hearing "cat" resolves to the single word "cat."

3. The Cohort Model

• Developed as an early explanation of the lexical access time-course.

• Mechanism:

  • Words are activated as soon as their initial phonemes match the auditory input.

  • As more phonemes are heard, the cohort of activated words narrows.

• Experimental Evidence:

  • Cross-Modal Priming:

    • Participants are faster to recognize or respond to a target word (e.g., "dog") if it is semantically related to a word in the cohort of candidates (e.g., "cat").

    • This supports the idea that multiple candidates are initially active and compete for selection.

4. Neural Network (Connectionist) Models

• Organic Explanation:

  • Connectionist models simulate lexical access using networks with nodes representing words and connections reflecting relationships.

  • Feed-forward Activation: Input propagates through the network, activating relevant nodes.

  • Feedback Activation: Higher-level contextual information influences earlier processing stages.

  • Lateral Inhibition: Competing candidates within the same level inhibit one another to refine the selection.

5. TRACE Model

• Information processing occurs through excitatory and inhibitory process interactions amount a large number of simple processing units

  • Three levels of units interact with one another in TRACE: feature phoneme and word

  • features activate phonemes which activate words

  • Within each layer, connects are inhibitor

  • Between each layer, connections are excitatory and bidirectional

• TRACE is broadly compatible with lexical effects on phoneme identification: phonemic restoration effect and lexical influence on compensation for coarticulation

• It predicts results of experiments on cue trading relations and categorical perception

• It recognizes words even if the initial phoneme is distorted or ambiguous

• It can predict bias towards shorter words within a cohort, consistent with gating results, but over larger stretch’s of speech’s is biased towards longer words, so that it is not crippled by embedded words

• Competitive recognition permits top down segmentation

• Evidence:

  • Visual World Paradigm:

    • Eye-tracking studies show that participants look at objects whose names are phonetically similar to the spoken input (e.g., looking at a picture of "cat" and "cap" when hearing "ca-").

    • This supports the TRACE model's prediction of simultaneous activation of multiple candidates.

Mental Lexicon

1. Information in Lexical Representation

The mental lexicon contains a vast amount of information about words, including:

• Phonological Information: Sound patterns and pronunciations.

• Orthographic Information: Written form and spelling.

• Morphological Information: Structure and components of words.

• Syntactic Information: major category

  • content word, noun, verb, adjective, or adverb.

  • Function words: pronoun, preposition, etc.

• Semantic Information: Meanings and conceptual associations.

• Pragmatic Information: Contextual usage and connotations.
  

2. Morphological Knowledge of Lexical Representation

• Free Morphemes: Can stand alone as words (e.g., "cat," "run").

• Bound Morphemes: Cannot stand alone and must attach to other morphemes (e.g., "-s" in "cats," "un-" in "undo").

• Roots vs. Affixes:

  • Roots: The base form of a word that carries the core meaning (e.g., "read" in "reader").

  • Affixes: Morphemes attached to roots to modify meaning or grammatical function.

    • Prefixes: Added to the beginning (e.g., "un-" in "undo").

    • Suffixes: Added to the end (e.g., "-ed" in "walked").

• Derivational vs. Inflectional Morphemes:

  • Derivational: Create new words or change word class (e.g., "beauty" → "beautiful").

  • Inflectional: Modify grammatical properties without changing word class (e.g., "walk" → "walks," "walked").
    

3. Differences Between Inflectional and Derivational Forms

• Inflectional Morphemes:

  • Functionally constrained (e.g., tense, number, case).

  • Do not change the core meaning or syntactic category of a word.

  • E.g., "dogs" (plural form of "dog").

• Derivational Morphemes:

  • Allow for greater semantic and syntactic flexibility.

  • Can change the syntactic category and meaning of a word.

  • E.g., "run" (verb) → "runner" (noun).

4. Full Parsing vs. Full Listing, and Hybrid Models

• Full Parsing Models:

  • Words are broken down into their morphological components during lexical access (e.g., "walked" is decomposed into "walk" + "-ed").

  • Efficient for managing a large lexicon since fewer stored entries are needed.

• Full Listing Models:

  • Entire words are stored as separate entries in the mental lexicon.

  • Reduces the cognitive load of recombining morphemes during access.

• Hybrid Models:

  • Combine elements of both approaches.

  • High-frequency words may be stored as whole forms, while less common forms are parsed.
    

5. Evidence for and Against Decomposition

• Evidence Supporting Decomposition:

  • Priming Studies:

    • Morphologically related words (e.g., "teach" → "teacher") show faster recognition, suggesting decomposition into shared morphemes.

  • Neuroimaging:

    • Brain regions involved in morphological processing (e.g., left inferior frontal gyrus) show activity during tasks requiring decomposition.

  • Error Patterns:

    • Speakers produce errors like overregularization (e.g., "goed" instead of "went"), indicating reliance on morphological rules.

• Evidence Against Decomposition:

  • Frequency Effects:

    • High-frequency complex words are processed faster than expected under decomposition models (e.g., "walked" is accessed as a whole unit).

  • Dual-Route Models:

    • Suggest both decomposed and whole-word pathways exist, depending on context, frequency, and task demands.