Discourse Markers

What discourse markers do

Take a slice of natural conversation and pull out every well, so, you know, like, I mean, and right. The propositional content is intact, but the talk falls apart. The hearer cannot tell where one topic ends and the next begins, which sentences are afterthoughts, which claims the speaker is committed to versus merely offering, and when to take a turn. Discourse markers carry that load.

Deborah Schiffrin's 1987 monograph Discourse Markers defined them as "sequentially dependent elements which bracket units of talk." Bruce Fraser, building on that work, characterises them as "lexical expressions which signal a relationship between the segment they introduce and the prior segment." Diane Blakemore, in a relevance-theoretic frame, treats them as procedurally encoded: they don't describe the world, they instruct the hearer how to integrate the upcoming utterance with their existing context.

How to recognise one

A typical discourse marker satisfies several diagnostics:

• Syntactic optionality. Removing it leaves a grammatical sentence (although the discourse becomes harder to follow).
• Loose syntactic integration. It usually appears at the edge of an utterance or as a parenthetical, not inside an argument structure.
• Prosodic separation. Often set off by a short pause or a distinct intonation contour.
• Truth-conditional invisibility. The literal truth conditions are essentially unchanged whether or not the marker is present.
• Procedural rather than conceptual meaning. The marker tells the hearer how to process what follows, not what is the case.
• Multifunctionality. The same marker can do several jobs across contexts (so can mark consequence, topic resumption, or floor-yielding).

Because these criteria are gradient rather than crisp, the boundary between discourse markers, modal particles, conjunctions, and interjections is debated. And, but, and because are conjunctions in syntax but often function as discourse markers when they introduce a turn.

A working inventory of English markers

Marker	Primary function	Example
well	Response marker; signals partial / mismatched answer	A: "Did you finish?" B: "Well, mostly."
so	Consequence; topic resumption; turn launch	"So, where were we?"
you know	Solidarity / shared ground appeal	"It was, you know, the usual thing."
I mean	Self-repair / clarification / strengthening	"It's expensive — I mean, really expensive."
like	Approximator / quotative / hedge	"He was, like, totally surprised."
right / yeah	Backchannel; confirmation	(Listener mid-turn:) "Right. Right."
anyway	Topic return after digression	"Anyway, as I was saying…"
actually	Counter-expectation; repair	"Actually, that's not what I meant."
oh	Information receipt	"Oh, I didn't know that."
okay	Transition; acceptance; turn close	"Okay. Moving on."

The inventory is open; lectures, court transcripts, social media DMs, and Slack channels each evolve their own short forms (tbh, ngl, lowkey) that play the same role.

A close look at "well"

Of all English markers, well may be the most studied. Schiffrin's chapter on it shows it consistently fronts utterances that don't perfectly match the prior turn's expectation:

• Indirect answer. A: "Where's the meeting?" B: "Well, the original room is double-booked, so…"
• Insufficient answer. A: "How was it?" B: "Well, it was something."
• Disagreement preface. A: "He's brilliant." B: "Well, he's certainly confident."
• Question redirection. A: "Will you come?" B: "Well, what time are you starting?"

The unifying procedural meaning, in Blakemore's analysis, is approximately: the upcoming utterance is the speaker's response, but the hearer should not expect a fully congruent or maximally informative answer. Removing well and saying simply "It was something" feels brusque; the marker softens the under-delivery.

Cross-linguistic typology

Every language studied so far has discourse markers, but their inventories, positions, and degrees of grammaticalisation vary widely.

• Japanese sentence-final particles. Spoken Japanese requires a marker at the end of most utterances. Ne invites or assumes agreement ("It's hot today, isn't it?"); yo asserts information the speaker believes the hearer lacks ("It's hot today — for your information"); sa is informal-assertive; na is reflective; kana wonders. Combinations like yone are common. These particles are obligatory enough that learners producing bare propositional sentences sound robotic.
• Mandarin Chinese. 啊 (a), 吧 (ba), 嘛 (ma), 呢 (ne) are sentence-final particles with overlapping but distinct functions: ba signals a soft suggestion or guess; ma emphasizes obviousness; ne can mark contrast or invite engagement.
• German modal particles. doch, ja, halt, eben, mal, schon sit in the middle field of the clause and adjust speaker stance. Doch roughly means "as you know / contrary to what you might think" — an answer to an unspoken expectation. They are notoriously hard to translate into English without three or four extra words.
• French. donc, quoi, bah, en fait, ben, tu vois form an inventory comparable to English's, with quoi sentence-final functioning roughly like you know.
• Spanish. pues, bueno, vale, o sea, en plan (newer, similar to English like).
• Arabic. yaʿni ("I mean") is one of the most frequent items in spoken Arabic across dialects, doing extensive repair, hedging, and turn-launching work.
• Greek. de, loipón, dhiladhí ("that is to say"), ára ("therefore"). The clitic de in Ancient Greek shows how deeply such items can be grammaticalised over time.

Markers tend to grammaticalise from full lexical items: English well began as the manner adverb meaning "in a good way"; like began as a comparative; I mean began as a literal speech-act verb. The grammaticalisation cline (Hopper & Traugott 2003) traces how items move from open-class to closed-class with bleached, procedural meanings.

A functional taxonomy

Schiffrin organises markers along five "planes of talk." Most markers operate on more than one plane simultaneously.

• Ideational structure. How propositions cohere. And, but, so, because.
• Action structure. What speech acts the speaker is performing. Well hedges responses; now opens a new act.
• Exchange structure. Turn-taking. Right, yeah, mhm as backchannels; so as turn-launchers; okay as turn closers.
• Information state. Tracking shared knowledge. You know appeals to common ground; oh registers receipt of new information.
• Participation framework. Speaker / hearer roles. Vocatives, address forms.

Acquisition and variation

• L1 acquisition. Children acquire propositional content earlier than discourse markers. Adult-like use of well, actually, and you know is among the latest pragmatic skills, often not consolidated until adolescence.
• L2 acquisition. Markers are notoriously slow to acquire and high in error. Mismatched marker use is one of the fastest indicators of non-native speech, even when grammar and accent are excellent.
• Sociolinguistic variation. Quotative like spread through American English in the late 20th century — D'Arcy and Tagliamonte's work documents the diffusion. So-prefacing in academic and tech speech ("So the thing is…") is a more recent shift.
• Genre variation. Written formal prose suppresses most markers; written informal prose (chat, social media) reintroduces them; broadcast journalism uses now and well as turn signals; courtroom transcripts foreground well as a hedge.

Why discourse markers matter

• Conversation analysis. The minute structure of turn-taking is built around these items; CA transcription notates them with as much care as content words.
• Pragmatics. Markers are the cleanest empirical handle on procedural meaning — the kind of meaning relevance theory pinpointed.
• Forensic linguistics. Authorship attribution and deception detection use marker frequency as a stylistic fingerprint.
• NLP and dialogue systems. Conversational AI must produce and interpret markers convincingly; their absence makes assistants feel stilted, their misuse makes them feel uncanny.
• Language teaching. Modern communicative pedagogy treats marker fluency as a real teaching target rather than a stigma.

Common pitfalls

• "They're just filler." Some are; most aren't. Compare a transcript with markers stripped to one with hesitation fillers stripped — the former is much harder to follow.
• Treating markers as random or interchangeable. Each has a specific procedural profile. Well and so are not synonyms.
• Stigmatising "like" and "you know" as bad speech. Empirically, frequency of these markers correlates with conversational fluency, not poor language. The stigma is sociolinguistic.
• Ignoring final-position markers in non-English data. English-trained analysts often overlook Japanese ne/yo, German particles, or French quoi.
• Treating markers as conceptually meaningful. Their meaning is procedural — instructions to the hearer about how to process the surrounding talk.