LING 142: Language in the Age of AI - Comprehensive Course Review

Comprehensive vocabulary flashcards covering articulatory and acoustic phonetics, automatic speech recognition, phonology, speech synthesis, syntax, parsing algorithms, lexical semantics, and word embeddings from LING 142.

Discreteness

The linguistic property whereby human language is constructed from a small inventory of distinct, meaningless sound units that combine to form meaningful expressions.

Duality of Patterning

The structural organization of human language at two or more combinatorial levels: meaningless discrete phonemes combine into words, words combine into phrases, and phrases combine into sentences.

Displacement

The ability in human language to communicate about entities, abstract ideas, or events that are separated in time and space from the immediate here and now.

Descriptive Linguistics

An objective, scientific approach to linguistics that observes, analyzes, and models how language is naturally used by its speakers, rather than prescribing how it should be used.

Autoregressive Generation

A text generation method utilized by language models where output tokens (characters or words) are generated sequentially one by one, with each newly generated token conditioned on the previous context.

Glottis

The anatomical opening or space between the vocal folds located within the larynx that regulates airflow from the lungs.

Phonation

The rapid vibration of the vocal folds within the larynx, produced when pulmonary air pressure forces the vocal folds apart and Bernoulli forces draw them closed again.

Active Articulators

The mobile structures of the vocal tract that actively move to shape or restrict the vocal airflow during speech production, including the lips, tongue tip/blade, tongue body, tongue root, and vocal folds.

Passive Articulators

The stationary anatomical structures of the vocal tract toward which active articulators move to create an obstruction, including the upper teeth, alveolar ridge, hard palate, velum, uvula, and pharyngeal wall.

International Phonetic Alphabet (IPA)

A universal phonetic notation system designed to represent every distinct speech sound across all human languages with a unambiguous one-to-one symbol-to-sound correspondence.

Schwa

The mid-central neutral vowel $[\partial]$ situated at the center of the vowel space, toward which unstressed vowels typically reduce in English.

Aspiration

A strong, turbulent burst of air produced during the release phase of a voiceless stop consonant prior to the onset of vocal fold vibration for the following vowel.

Sonority Hierarchy

A ranking of speech sound categories according to their inherent acoustic audibility and openness, where low/open vowels possess the greatest sonority and voiceless stops possess the lowest.

Electromagnetic Articulography (EMA)

A phonetic instrumentational method that tracks the position and movement of speech articulators inside and outside the oral cavity over time by placing sensor coils in an alternating electromagnetic field.

Fundamental Frequency ($F_0$)

The lowest frequency component of a complex speech wave corresponding to the rate of vocal fold vibration during phonation, typically ranging between $100\,Hz$ and $500\,Hz$.

Spectrogram

A two-dimensional acoustic visual representation displaying how spectral frequencies change over time, where the $x$-axis represents time, the $y$-axis represents frequency, and shading or color indicates sound intensity.

Formants

The characteristic resonant frequencies of the vocal tract ($F1$, $F2$, $F3$) that amplify acoustic energy into dark bands on a spectrogram, serving as acoustic fingerprints for vowel identity.

Source-Filter Model

A foundational acoustic theory modeling speech production as a combination of a sound source (the larynx producing periodic pulses or turbulent noise) and an acoustic filter (the vocal tract shaping resonant frequencies).

Radio Rex

A 1920s mechanical toy dog operated by an electromagnet that released an internal spring when triggered by $500\,Hz$ acoustic energy (the first formant of the vowel $[\varepsilon]$ in 'Rex'), serving as an early pioneer of speech recognition.

Word Error Rate (WER)

The primary evaluation metric for automatic speech recognition, computed via minimum edit distance as $100 \times \frac{\text{Insertions} + \text{Substitutions} + \text{Deletions}}{\text{Total Words in Correct Transcript}}$.

Mel Spectrogram

A speech spectrogram whose frequency axis is warped to the Mel scale, reflecting the human auditory system's heightened sensitivity to low-frequency variations compared to high-frequency variations.

Phoneme

An abstract, contrastive speech sound in a language's phonological system that distinguishes one word meaning from another, typically denoted within slashes / /.

Minimal Pair

A pair of words with distinct meanings that differ by exactly one single sound segment in the identical position (e.g., 'pack' and 'back'), establishing that the contrasting sounds represent distinct phonemes.

Allophone

A non-contrastive phonetic variant or surface realization of a phoneme whose occurrence is often conditioned by surrounding phonetic environments, denoted within square brackets [ ].

Complementary Distribution

The mutually exclusive occurrence of allophones of the same phoneme, where one variant occurs exclusively in specific phonetic environments where the other never appears.

Free Variation

The condition where two or more allophones of the same phoneme can substitute for one another in the exact same phonetic environment without altering the word's meaning.

Natural Class

A group of phonemes within a language that share one or more distinctive phonological features to the exclusion of all other sounds, patterning together in phonological rules.

Text Normalization

The initial text-processing pipeline in speech synthesis that transforms non-standard lexical tokens like numbers, dates, times, abbreviations, and symbols into fully articulated written words.

Cut-and-Paste Synthesis

A domain-restricted speech synthesis method that concatenates pre-recorded words and phrases spoken by a single voice talent, often used in automated announcements.

Unit Selection Synthesis

A concatenative speech synthesis technique that selects acoustic fragments of varying lengths (phones, diphones, syllables, words) from a multi-hour single-speaker speech database to assemble natural utterances.

Diphone

A sub-word speech unit extending from the acoustic center of one phoneme to the acoustic center of the immediately following phoneme, capturing the coarticulatory transition between adjacent sounds.

Formant Synthesis

A parametric, rule-based speech synthesis approach that computes speech waveforms by explicitly calculating acoustic parameters and formant frequencies rather than splicing real human recordings.

Audio Deepfake

A synthetic or computationally altered voice recording created using deep neural networks to closely mimic the acoustic properties, vocal timbre, and speech patterns of a specific target individual.

Open Word Classes

Part-of-speech categories that contain a large, unbounded number of words and continually admit newly coined lexical items, including nouns, verbs, adjectives, and adverbs.

Closed Word Classes

Part-of-speech categories that consist of a fixed, highly restricted inventory of grammatical words to which new items are rarely added, such as determiners, pronouns, conjunctions, and prepositions.

Penn Treebank

A widely used natural language processing corpus containing approximately one million words of annotated English newswire text tagged for parts of speech and syntactic phrase trees.

Brill Tagger

An error-driven, transformation-based part-of-speech tagger that initializes words with their most frequent tag and iteratively learns contextual transformation rules using templates to correct tagging errors.

Syntactic Head

The primary, structurally essential word within a phrase that determines the phrase category and syntactic behavior (e.g., the noun inside a noun phrase).

Context-Free Grammar (CFG)

A formal grammatical system consisting of rewrite rules where each production maps a single non-terminal symbol on the left-hand side to a sequence of terminal and/or non-terminal symbols on the right-hand side.

Syntactic Ambiguity

A property of a sentence that can be derived in multiple distinct ways by a grammar, resulting in multiple valid hierarchical parse trees representing different interpretations.

Earley Item

A data structure in the Earley parsing algorithm representing a grammar production rule with an internal dot ($\bullet$) indicating how much of the rule has matched, accompanied by start and end fence-post indices.

Earley Predictor

The top-down Earley parsing step that adds new items into the parse table whenever an active item has a non-terminal symbol appearing immediately to the right of its dot.

Earley Scanner

The bottom-up Earley parsing step that reads the current input word, validates that it matches an expected terminal category, and creates a finished lexical item spanning the word.

Earley Completer

The Earley parsing step triggered when an item finishes matching, which searches the parse table for earlier active items waiting for that completed non-terminal and advances their dots.

Denotation (Extension)

The actual set of entities, real-world objects, or referents to which a linguistic expression or concept corresponds in the world.

Intension

The conceptual mental representation, definition, or internal semantic features that define the essential properties required to identify a concept.

Prototype Theory

A cognitive semantic model postulating that conceptual categories are structured around exemplary, typical members (prototypes) with graded membership and fuzzy boundaries rather than strict binary criteria.

Distributional Hypothesis

The foundational linguistic principle formulated by J. R. Firth (1957) stating that words occurring in similar textual contexts share similar semantic meanings ('You shall know a word by the company it keeps').

Meronymy

A semantic hierarchical relation where one concept designates a component or part of a larger whole (e.g., 'midsole' is a meronym of 'sole', which is a meronym of 'shoe').

Hypernym

A generic, superordinate semantic term whose meaning encompasses the more specific, subordinate concepts known as hyponyms.

WordNet

A manually constructed lexical database of English that groups nouns, verbs, adjectives, and adverbs into sets of cognitive synonyms called synsets, interlinked by semantic relations.

Vector

An ordered array of numerical values representing a geometric point or directed arrow within a multidimensional mathematical space.

Euclidean Distance

A distance metric calculating the straight-line length between two vectors $\mathbf{u}$ and $\mathbf{v}$, defined as $d = \sqrt{\sum_{i=1}^n (u_i - v_i)^2}$.

Cosine Similarity

A metric evaluating vector similarity based on the cosine of the angle $\alpha$ between them, defined as $\cos(\alpha) = \frac{\mathbf{u} \cdot \mathbf{v}}{\|\mathbf{u}\| \|\mathbf{v}\|}$, independent of their magnitude.

Jaccard Index

A coefficient measuring document similarity defined as the ratio of shared content words ($M$) to total unique content words appearing across both documents ($N$), expressed as $M/N$.

Dimensionality Reduction

A mathematical compression procedure, such as Singular Value Decomposition (SVD), that transforms sparse, high-dimensional word vectors into dense, low-dimensional vectors while preserving relative geometric distances.