Language: How Sounds, Signs, and Context Become Meaning

“Can you open the window?” looks like a question about ability, but usually functions as a request. Understanding it requires more than hearing individual sounds: you must identify words, organize their relationships, interpret meaning, and infer what the speaker wants.

Language is a structured system for communication and thought that can be spoken, signed, or written. Speech is one expression of language, not its definition. These notes explain the interacting processes that make language possible.

Learning goals and Feynman practice

Distinguish sound structure, word structure, grammar, meaning, and context. Explain speech segmentation, interpret language-related brain measurements cautiously, and describe how biological constraints and experience jointly shape acquisition.

Read a section, close it, and explain it to a beginner. Give a new example, identify the proposed mechanism, and state one limitation. Repair the step you cannot explain before revealing the self-check answers.

1. Levels of language: different questions about the same sentence

Level	Main question	Example
Phonology	Which sound contrasts and patterns matter in this language?	Changing /p/ to /b/ distinguishes “pat” from “bat” in English.
Morphology	How are meaningful grammatical units combined?	“Walked” combines walk with a past-tense marker.
Syntax	How are words and phrases organized?	“The dog chased the cat” differs from “The cat chased the dog.”
Semantics	What do expressions mean?	Understanding the meanings of dog, chase, and the sentence’s relations.
Pragmatics	What does a speaker communicate in this context?	“It is cold here” may indirectly request closing a window.

The outline’s four-level account is useful, but adding morphology makes the distinction between sound and meaningful structure clearer.

A phoneme is a language-specific sound category that distinguishes words. It is not a letter or a fixed physical sound independent of context. English has roughly forty-plus phonemes in many descriptions, with counts depending on dialect and analysis. Human languages collectively use far more contrasts than a universal inventory of exactly 100 phonemes would suggest.

A morpheme is a minimal meaningful or grammatical unit. A whole word may be one morpheme, while “cats” has cat plus plural -s. The same written ending can serve different grammatical functions: -ing in different constructions cannot be explained solely by calling it a tense marker.

Pragmatics includes indirect requests, sarcasm, and context-sensitive interpretation of metaphor. It overlaps with semantic and social knowledge rather than forming a completely separate final stage.

Teach it back: Explain why “unhelpful” contains more than one meaningful unit, while a letter is not automatically a morpheme.

2. Speech is a stream, not words separated by silence

Written text supplies spaces. Ordinary speech supplies a continuous, variable acoustic signal, with some pauses but no reliable silence at every word boundary. Adjacent sounds influence one another through coarticulation, and speakers differ in rate, accent, pitch, and articulation.

A closure for a stop consonant can create a short low-energy interval within a word. Conversely, adjacent words can run together without a pause. The listener therefore cannot solve word segmentation by detecting silence alone.

Speech rates such as 180–250 words per minute describe some settings, not a universal rate or a fixed conversion into phonemes per second. Speaking style, language, word length, and counting conventions matter.

How does segmentation work?

Listeners combine acoustic details, stress and rhythm, familiar vocabulary, grammatical expectations, and the probability of sound sequences. Several candidate interpretations can compete as speech unfolds. Context helps constrain a noisy or ambiguous signal, but it does not replace sensory evidence entirely.

An isolated clip removed from a sentence may be harder to recognize because boundaries were cut artificially and useful neighbouring cues disappeared. This is especially likely with reduced or coarticulated speech; it does not mean isolated words are generally unrecognizable.

A useful analogy is identifying overlapping brushstrokes in a painting. The surrounding picture helps determine what a local mark belongs to, but the mark still constrains the interpretation.

Teach-back check: why do captions help you interpret unfamiliar speech without proving that hearing is just reading?

Captions provide likely word candidates that guide interpretation of the acoustic signal. Speech perception still relies on auditory processing; context changes the interpretation of evidence rather than replacing its sensory source.

3. Categorical perception: a continuous signal can yield a category judgment

Imagine synthesized syllables gradually changing between two speech categories. Listeners may label many acoustic steps alike, then switch labels over a relatively narrow region. Discrimination can be better for equally spaced sounds crossing that category boundary than for sounds within one category.

That pattern is categorical perception. It helps explain stable recognition despite physical variability, but does not imply that the auditory system loses all within-category detail or makes every category decision instantly and without error.

Different experiments manipulate different cues. Voice onset time, for example, can distinguish voiced and voiceless stop categories in a language. Place-of-articulation contrasts such as /ba/ versus /da/ involve other acoustic cues, including formant transitions. There is no universal 40-millisecond switch that defines every phoneme contrast.

Experience shapes category boundaries and cue use. A distinction important in one language may not function as a phonemic contrast in another, even though the physical variation remains detectable under some conditions.

Teach it back: Why is an abrupt change in labels not, by itself, evidence that all sounds within each category are perceptually identical?

4. Ambiguous words: activation and context interact

Consider “She waited beside the bank.” Riverbank and financial interpretations are possible until more context appears. Experiments can use response speed or other measurements to investigate which related meanings are temporarily accessible.

Exhaustive lexical access is an influential account proposing early activation of multiple meanings before context selects among them. It should not become “every meaning of every ambiguous word is always activated equally.” Frequency, prior context, timing, task, and the particular word can change the pattern.

A better working explanation is competition among candidates informed by both the input and context. Strong context may favour one interpretation early; later information can require revision.

Teach it back: Give a sentence in which “bank” remains ambiguous, then add a clause that resolves it. What does that show about the limits of understanding one word in isolation?

5. N400 and P600: time-sensitive evidence, not mental-state labels

An event-related potential, or ERP, is estimated from EEG aligned to an event, often by averaging many trials. It measures electrical patterns at the scalp; component timing does not identify a unique source or reveal a complete thought.

Component	Common association	Qualification
N400	Meaning-related processing affected by context, association, and expectancy; often prominent around 400 ms.	Not restricted to absurd words, not a binary semantic-error detector, and not necessarily identical across tasks.
P600	A later positive response often found with grammatical difficulty, reanalysis, or repair.	Not exclusive to syntax; meaning-related and task-related manipulations can also produce late positivities.

Kutas and Hillyard’s classic work linked out-of-context sentence words with an N400 response, helping establish electrophysiological measures of comprehension. [1]

“I take my coffee with milk and shoes” creates an unusual semantic continuation. “The child throw the toys,” under a present-tense singular-subject reading, creates a grammatical mismatch. These examples illustrate contrasts; they do not guarantee a specific wave in every individual or prove that grammar and meaning occupy isolated brain boxes.

Teach-back check: why compare plausible and implausible continuations across many trials?

An experimental contrast helps estimate activity associated with the manipulation while reducing unrelated variability. Individual words also differ in frequency, length, and other properties, so those factors need control. One waveform cannot uniquely establish a single cognitive operation.

6. Language networks and sign language

Many people show left-biased language processing, but language depends on distributed frontal, temporal, and other regions and their connections. The familiar Broca = production, Wernicke = comprehension pairing is an introductory historical shortcut. Both production and comprehension recruit interacting processes beyond these labels.

Infant imaging provides evidence of early organization. For example, a widely cited fMRI study tested three-month-old infants, not two-day-old newborns. [2] An early asymmetry does not, by itself, prove a fully genetically specified adult language network: prenatal experience, maturation, and early input must also be considered.

Signed languages such as American Sign Language, ASL, have linguistic structure and are not merely pantomime or spoken English conveyed by hand. Sign-language studies find substantial overlap with spoken-language networks along with modality- and experience-related differences. [3] “Identical regions with identical functions in every signer and speaker” overstates the evidence.

Infants exposed to sign language can show rhythmic, language-related manual activity often described as manual babbling. Distinguish this from arbitrary hand movement and from the claim that every deaf infant automatically babbles in signs regardless of accessible language input.

Teach it back: Why does language processing in signers challenge the claim that language is simply the brain’s control of speech muscles?

7. Acquisition: specialization without an absolute universal-to-native switch

Early infants can discriminate many speech distinctions that adults without relevant experience find difficult. During the first year, perception becomes increasingly tuned to experienced languages, with some non-native contrasts becoming harder and native contrasts becoming more efficient. Classic cross-language studies documented this perceptual reorganization. [4]

Newborns do not demonstrably discriminate every possible human phoneme under every test. Nor is later reduction necessarily permanent loss of every ability to hear a distinction. Contrast type, task, bilingual exposure, and later training can affect performance.

Child-directed speech, also called infant-directed speech, often includes expressive intonation, repetition, and adjustments to rate or vocabulary. It can support attention and learning, but varies across communities, speakers, and situations. It is neither universally identical nor the only route to acquisition. Children learn from accessible interactions in their language environment, including signed input where relevant.

Teach-back check: why can improved native-language discrimination and reduced non-native discrimination occur together?

Learning tunes attention and category representations to the distinctions most useful in the experienced language. Specialization changes how information is organized; it is not simply a general decline in hearing ability.

8. Nativism, learning, and generativity

Skinner emphasized environmental contingencies in verbal behaviour. Chomsky challenged the sufficiency of such accounts, highlighting the structured and generative nature of language: people can understand and produce sentences they have never encountered before.

A child saying “holded” may be extending a productive past-tense pattern to an irregular verb. This is overregularization. It is evidence of generalization beyond copying the exact adult form.

The error does not, by itself, prove a particular innate grammar or disprove every learning-based account. Statistical, connectionist, usage-based, and rule-based approaches offer different explanations of how patterns are acquired. Reinforcement and imitation can contribute without constituting a complete account.

The useful question is what capacities the learner brings, what information the environment provides, and which model explains the resulting development. “Biology or learning” is too simple: biological organization makes learning possible, and experience shapes language knowledge.

Teach it back: Why is “swimmed” evidence against pure verbatim imitation but insufficient to decide among all theories of language acquisition?

9. Sensitive periods: timing matters, but components differ

Age of acquisition is associated with differences in pronunciation, grammar, and some processing patterns. A sensitive period is a developmental interval during which experience has especially strong effects. A strict critical period implies a stronger boundary; evidence must specify the ability and conditions concerned.

Aspect	Broad pattern	Avoid this conclusion
Pronunciation and phonological categories	Early exposure often supports native-like patterns.	Adults cannot improve pronunciation or acquire new contrasts.
Grammar	Age, exposure, instruction, and learning history can affect eventual proficiency.	One universal birthday permanently ends grammatical learning.
Vocabulary and meanings	Learning remains possible throughout life.	Every adult finds vocabulary effortless or unaffected by ageing and context.

Distinguish late first-language acquisition, after limited early access to any full language, from learning an additional language after a strong first language is established. These are different developmental conditions.

Extreme deprivation cases such as Genie are important but heavily confounded by trauma, health, isolation, and uncertain histories. They do not cleanly isolate age or establish one precise deadline. Research on access to signed first languages offers additional evidence about timing, while also requiring careful attention to participants’ experiences.

10. Worked example: understanding an indirect request

Someone says, “Could you pass the salt?” You must recognize a variable acoustic sequence, find word candidates, organize the grammatical relationships, interpret the sentence’s literal meaning, and infer a socially appropriate request. If a word is obscured by noise, context may help recover it. If the tone is sarcastic, speaker intent may differ.

These operations overlap and interact rather than waiting in a rigid queue. A spoken response adds word selection and articulation; a signed response uses a different output modality while still requiring linguistic organization.

Teach it back: Explain the example using phonology, syntax, semantics, and pragmatics. Add a morphological contrast by changing “pass” to “passed.”

Retrieval workshop and spaced review

From memory, define phoneme and morpheme using examples that are not in these notes. Explain segmentation, categorical perception, and semantic ambiguity. Reconstruct the ERP table, then state what a brain measurement cannot establish alone.

Repair: “N400 means semantic error, and P600 means grammatical error”

These are common experimental associations, not unique labels for operations. Responses depend on context, task, and stimulus properties, and different manipulations can overlap. Infer a process from controlled comparisons and converging evidence.

Repair: “Early left activation proves language is completely hardwired”

Early organization supports developmental constraints but does not isolate genes from prenatal and postnatal experience. The measured response is also not a complete adult language system.

Apply: how would you test whether an ambiguous word activates two meanings?

Present sentences with varied context, probe words related to each meaning at different delays, and compare responses with suitable unrelated controls. Control meaning frequency and avoid assuming that the probe task leaves comprehension unchanged.

Review tomorrow and in a week. Score each explanation: 0 for a label, 1 for a definition, 2 for a mechanism plus example and limitation. Repair the weakest explanation before adding terminology.

One-minute explanation: Language combines structured forms with meaning and social use. Speech perception finds words in variable continuous input, aided by learned categories and context. Brain measurements reveal interacting networks and time-sensitive processes, not isolated meaning and grammar switches. Children generalize patterns rather than only repeat examples. Biological constraints, experience, and developmental timing jointly shape acquisition, while learning remains possible across life.

Sources and further reading

The supplied numbered citations did not include a bibliography. These primary studies support the ERP, infant-imaging, sign-language, and perceptual-development examples. The window, bank, and salt scenarios are teaching aids.

1. Kutas, M., and Hillyard, S. A. (1980). Reading senseless sentences: brain potentials reflect semantic incongruity.
2. Dehaene-Lambertz, G., Dehaene, S., and Hertz-Pannier, L. (2002). Functional neuroimaging of speech perception in infants.
3. Cerebral organization for language in deaf and hearing subjects: biological constraints and effects of experience.
4. Werker, J. F., and Tees, R. C. (1984). Cross-language speech perception: evidence for perceptual reorganization during the first year of life.