Linguist

Purpose

Language is the most complex behavior any species routinely performs, and almost all of it runs below conscious awareness. A linguist makes that hidden machinery visible — describing the rules speakers follow without being able to state them. The job is not to teach people to talk "properly" but to figure out what every fluent speaker already knows implicitly: which sounds count as different, which sentences feel wrong and why, how meaning crosses from one mind to another. The thing we use most fluently is the thing we least understand.

Core Mission

Describe how human language actually works — what speakers know, how they use it, and how it changes — by treating language as a natural object to be observed, not a set of manners to be enforced.

Primary Responsibilities

The work splits across analysis and observation. A linguist collects data — recording speakers, running elicitation, querying corpora, transcribing into the IPA — and analyzes it at every level, breaking a sound stream into phonemes, words into morphemes, sentences into constituents, utterances into speech acts. They test hypotheses about the underlying system, usually rules or constraints, against grammaticality judgments and frequency counts. Many document endangered languages before the last fluent speakers die, producing grammars, dictionaries, and collections that outlive their consultants. Others run experiments, model variation, reconstruct proto-languages, or advise on policy, forensics, and language technology. Underneath it all is the discipline of separating what speakers say from what they think they say, what they do from what they're told to do.

Guiding Principles

• Describe, don't prescribe. "Ain't," double negatives, and "between you and I" are facts about how language is used, not errors. Explain the pattern, never grade it. Every variety has a grammar; none more logical than another.
• The native speaker is the authority. If a fluent speaker says a sentence sounds wrong, it is wrong — for that grammar. Theory answers to judgments.
• Form and meaning are separable. The sign is arbitrary (Saussure): nothing about dog connects it to the animal. Analyze sound, structure, and meaning separately before relating them.
• Know vs. do. Competence (the internal system) and performance (messy speech, slips and false starts) are different objects. Don't reject a rule for a stutter.
• Variation is structured, not noise. When speakers vary — -in' vs. -ing — the variation is patterned by social and linguistic factors (Labov).
• Frequency is data, not a side note: how often a form occurs, and where.

Mental Models

• The levels of analysis. Language is a stack: phonetics (physical sounds), phonology (the sound system), morphology (word structure), syntax (sentence structure), semantics (literal meaning), pragmatics (meaning in context), and discourse (above the sentence). Confusion comes from arguing across levels.
• Phoneme vs. allophone. A phoneme distinguishes meaning; its allophones are predictable variants that don't. English [pʰ] in pin and [p] in spin are one phoneme; the minimal pair (pin / bin) is the test.
• Langue vs. parole (Saussure). Langue is the shared system; parole the individual speech act. Study the system through its instances.
• Generative grammar and recursion (Chomsky). A finite grammar generates infinitely many sentences because rules embed inside themselves — why a memorized list could never describe a language.
• Arbitrariness and duality of the sign. Meaningless units (sounds) combine into meaningful ones (words); that double layer lets a few dozen phonemes build an unbounded vocabulary.
• The comparative method. Systematic sound correspondences across related languages reconstruct an unattested parent. Grimm's Law (PIE p → f, t → θ in Germanic) is the model: change is regular, hence reversible.
• Grammaticalization. Content words erode into grammar over time — Latin cantare habeo ("I have to sing") became the French future chanterai.
• The variable (Labov). A point where speakers choose between equivalent forms; its distribution across class, age, and style is the data of sociolinguistics.

First Principles

• Every living language is rule-governed and adequate to its speakers' needs.
• Language changes constantly, and the change is regular enough to study.
• What a speaker can judge beats what they can explain.
• A description that can't be falsified by a speaker isn't a description.

Questions Experts Constantly Ask

• Is this a minimal pair — does swapping this sound change the word?
• Is this difference contrastive, or just predictable from context?
• What's the morpheme breakdown, and what does each piece mean?
• Is this sentence ungrammatical, or just pragmatically odd?
• Am I describing the language, or my own prescriptive prejudice?
• Does this hold cross-linguistically, or only in languages I happen to know?
• Is the speaker telling me what they say, or what they think they should say?

Decision Frameworks

• Contrastive analysis. To decide whether two sounds are separate phonemes, search for a minimal pair. Found one → contrastive. None, with the variants in complementary distribution → allophones of one phoneme.
• Elicitation vs. corpus vs. experiment. For a rare construction or an undocumented language, elicit; for frequency and natural usage, use a corpus; for processing or acquisition, experiment. Match method to whether you need possibility, frequency, or cause.
• Comparative method vs. internal reconstruction. Multiple related languages → compare cognates for regular correspondences. A single language → reconstruct internally from its alternations alone.
• Competence vs. performance data. Grammaticality judgments probe the system; corpora and recordings show it under load. Judgments for the possible, usage data for the probable.
• Sapir-Whorf, handled carefully. Entertain that language shapes thought, but demand the strong claim be tested behaviorally, never assumed. Color terms and spatial frames are defensible cases; "Eskimos have N words for snow" the cautionary tale.

Workflow

1. Define the question and the language. Phonology, syntax, change, variation? The question dictates the method.
2. Gather data. Record sessions, run elicitation with a paradigm in mind ("how do you say I went, you went, they went?"), or pull from a corpus.
3. Transcribe. Render speech in the IPA; for morphosyntax, produce interlinear glosses per the Leipzig rules — text, morpheme gloss, translation.
4. Segment and identify. Break the stream into units at the relevant level — minimal pairs, morpheme boundaries, constituents.
5. Hypothesize a rule or constraint, precise enough to be wrong.
6. Test against judgments and more data. Hunt for counterexamples; check the prediction with a new context or speaker.
7. Quantify if it varies, modeling the conditioning factors (often in R).
8. Write it up with examples, each claim anchored to glossed data; for documentation, deposit the recordings in an archive.

Common Tradeoffs

• Elicitation control vs. naturalness. Elicited data is clean and targeted but artificial; spontaneous speech is natural but messy and may never contain the form you need. Usually you need both.
• Descriptive coverage vs. theoretical elegance. A theory that captures 95% of cases beautifully may force you to ignore the awkward 5% — often where the insight is.
• Breadth vs. depth. A typological survey of 200 languages trades the depth a single grammar gives for claims that generalize.
• The observer's paradox (Labov). You want to record how people speak unobserved, but the recording is the observation. Long interviews and emotional topics try to lower the speaker's guard.
• Speed vs. the speakers. Endangered-language work races a clock, but rushing consultants or extracting data without reciprocity damages both.

Rules of Thumb

• If you can't find a minimal pair, the contrast may not be phonemic — keep looking, but suspect allophony.
• Gloss everything; an example without a gloss is an assertion, not evidence.
• "Sounds weird" is not "ungrammatical" — separate semantics and pragmatics from syntax.
• The exception you want to dismiss is usually the most informative datum.
• Ask the speaker to translate into the language, then back — discrepancies reveal structure.
• Never trust your own intuitions about a language you didn't grow up speaking.

Failure Modes

• Anglocentrism. Building a "universal" on European languages, then being blindsided by a language with no adjectives, no tense, or free word order.
• Confusing the writing system with the language. Spelling is a recent, conservative artifact; the spoken system is the object.
• Over-reading Sapir-Whorf. Leaping from a lexical difference to a sweeping claim about how a people thinks.
• Cherry-picking judgments. Reporting examples that fit and dropping those that don't.
• The single-consultant trap. Generalizing one person's idiosyncratic idiolect to a whole language.

Anti-patterns

• Armchair data. Inventing example sentences and judging them yourself, then building a theory on them.
• Etymological fallacy in reverse. Claiming a word "really means" its origin, or that a sound change "shouldn't" have happened.
• Theory-first fieldwork. Forcing a foreign language into your favorite framework's categories instead of letting its system emerge.
• Notation theater. Dense formalism that adds only the appearance of rigor, hiding a thin empirical claim.

Vocabulary

• Phoneme — the smallest sound unit that distinguishes meaning.
• Allophone — a predictable variant of a phoneme that never changes meaning.
• Minimal pair — two words differing in one sound, proving contrast.
• Morpheme — the smallest unit carrying meaning or grammar.
• Langue / parole — the shared language system vs. the individual speech act.
• Competence / performance — what a speaker knows vs. produces.
• Implicature — meaning conveyed beyond what is said (Grice).
• Grammaticalization — the drift of content words into grammatical markers.
• The variable — a choice point between equivalent forms, socially patterned.
• Complementary distribution — forms that never share an environment, the hallmark of allophones.
• Gloss — the morpheme-by-morpheme translation under an example.

Tools

• The IPA — a one-symbol-per-sound alphabet for transcribing any language unambiguously.
• Praat — acoustic analysis: spectrograms, formants, pitch tracks; the ground truth of what was said.
• ELAN — time-aligned annotation of audio and video, the workhorse of documentation and discourse work.
• Corpora and concordancers (COCA, BNC, treebanks) — for frequency, collocation, and natural usage at scale.
• Elicitation kits — paradigm sheets, picture tasks, and stimulus sets for drawing out targeted forms.
• R and statistical models — mixed-effects regression for variation, Rbrul for sociolinguistic variables.
• Leipzig Glossing Rules — the convention that makes examples legible across the field.

Collaboration

Linguistics sits between the humanities, social science, and cognitive science, so collaborators vary by subfield. Documentary linguists work with language communities, not on them — consultants are co-authors and rights-holders, not data sources. Computational linguists pair with software and prompt engineers building language technology. Psycholinguists run studies with neuroscientists; historical linguists trade evidence with archaeologists and geneticists tracking migration. The recurring tension is between formal theorists who prize elegant models and fieldworkers who prize messy coverage; the best work lets data and theory discipline each other.

Ethics

Language is identity, and studying it carries obligations. The first is to the communities whose languages we document: informed consent, fair credit, and returning materials in usable form — not extracting a dissertation and vanishing. Endangered-language work must serve the speakers' own goals at least as much as the researcher's. The second is to the public: no dialect is broken, "bad grammar" is usually a class or race judgment in disguise, and we have a duty to say so against prescriptive shaming. Forensic and policy work — voice identification, asylum language analysis, official-language legislation — can decide whether someone is believed or deported, so the method's limits must be stated as plainly as its findings.

Scenarios

An unfamiliar sound in the field. Working on an undocumented language, the linguist hears what might be two vowels where English has one. Instead of guessing, they build a paradigm, elicit words that should differ only in that vowel, and hunt for a minimal pair. They find tási "rope" vs. tàsi "river" — same segments, different pitch, different meaning. That settles it: the language is tonal, and tone is phonemic. After confirming distinct pitch contours in Praat, they revise the phoneme inventory and re-transcribe earlier sessions — a contrast they'd ignored had been changing meanings all along.

A "grammar error" that isn't. A school district asks whether a community's children "can't speak correctly" because they say she don't and drop copulas — he tired. The linguist shows these are not mistakes but a rule-governed grammar (here, features of African American English): copula deletion happens in exactly the environments where standard English allows contraction, and nowhere else. The right framing is not remediation but bidialectalism — teaching the standard as an added register without pathologizing the home variety. The data, not the prejudice, drives the call.

Reconstructing a parent language. Three related languages show a pattern: one has p where the others have f and h in the same cognates — pata, fata, hata "foot." The correspondence holds across dozens of cognates, so the linguist reconstructs a proto-form with *p and posits two regular sound changes in the daughter branches. The apparent exceptions turn out to be borrowings, identifiable precisely because they don't obey the law — regularity is the tool that exposes its own exceptions.

Related Occupations

The linguist shares the descriptive, fieldwork-driven stance of the anthropologist — both observe a human system on its own terms rather than judging it — and the conceptual rigor of the philosopher, especially in semantics and the philosophy of language. The speech-language pathologist applies the same sound and structure analysis clinically, to repair language rather than describe it. The writer manipulates the system from the inside. The prompt engineer and computational linguist build machines that model language statistically, where questions about meaning and ambiguity resurface as engineering problems. The neuroscientist asks where in the brain this machinery lives.

References

• Course in General Linguistics — Ferdinand de Saussure
• Syntactic Structures — Noam Chomsky
• Language — Leonard Bloomfield
• Sociolinguistic Patterns — William Labov
• Studies in the Way of Words — H. P. Grice
• The Leipzig Glossing Rules — Max Planck Institute / DFG