Introduction
Discourse markers, small words and phrases such as well, so, like, you know, I mean, and basically, are a persistent feature of spontaneous spoken language. They do not add propositional content to a sentence the way nouns or verbs do; instead they organize discourse, signal a speaker's stance, manage turns between speakers, and help listeners interpret how one utterance relates to what came before it [1][2][3]. Deborah Schiffrin's 1987 study Discourse Markers, based on sociolinguistic interviews recorded in Philadelphia, established one of the field's founding frameworks, arguing that markers such as oh, well, and, but, so, and y'know supply contextual coordinates that help participants track where an utterance sits within an unfolding conversation [1]. A later edited collection by Andreas Jucker and Yael Ziv, Discourse Markers: Descriptions and Theory, extended this analysis across languages and theoretical frameworks, from relevance theory to conversation analysis, gathering thirteen studies on markers in English, Finnish, Hebrew, Korean, and Japanese [2].
Research on discourse markers has concentrated heavily on casual conversation: sociolinguistic interviews, friend-group talk, and adolescent peer speech. Comparatively less research has isolated how the same forms operate inside professional and workplace speech, where formality norms, turn-taking structures, and listener evaluation differ from casual talk. The sections below draw on corpus studies of business meetings, workplace interaction, and professional registers to describe how discourse markers are classified, how specific forms such as like, you know, basically, and I mean have been documented across registers, and how listeners judge speakers who use them. The discussion also proposes an organizing scheme for workplace discourse-marker use built around four functional categories, approximation, hedging, quotative marking, and turn-holding, and distinguishes that functional use from repeated, filler-like insertion that does not appear to serve any of these purposes.
What Counts as a Discourse Marker
Linguists differ on the precise boundaries of the category, but there is broad agreement on the core criteria. Bruce Fraser's widely cited paper "What Are Discourse Markers?" defines discourse markers as a pragmatic class of lexical expressions, drawn mainly from conjunctions, adverbs, and prepositional phrases, whose function is to signal a relationship between the segment they introduce and the prior segment of discourse rather than to contribute to that segment's propositional content [3]. Fraser groups markers into functional types, including elaborative markers such as and, inferential markers such as so, contrastive markers such as but, and markers that manage topic and turn-taking, such as well [3]. Schiffrin's earlier account similarly treats markers as operating on several discourse planes at once: textual organization, participant relationships, and information state [1].
Because discourse markers are semantically light and their interpretation depends heavily on context, the same form can carry different functions in different utterances. Like alone has been documented performing several distinct jobs in English, including approximating a following quantity or description, marking emphasis, introducing reported speech or thought as a quotative, and acting as a discourse focuser that highlights the informationally important part of an utterance. Gisle Andersen's book-length relevance-theoretic study of adolescent speech treats this functional range as central to how like is interpreted in context, with hearers relying on pragmatic inference rather than a single fixed meaning to work out which job the word is doing on a given occasion [4]. This functional multiplicity is a central reason a classification organized around what a marker is doing in a given utterance, rather than which word is used, is more useful for describing workplace speech than a simple word-frequency count.
A Workplace Classification: Four Functional Categories
Building on the general discourse-marker literature, workplace discourse-marker use can be organized into four broad functional categories. This grouping is not one that appears as such, under these labels, in the cited sources; it is a synthesis drawn from functions those sources independently document, applied specifically to the professional register.
Approximation
Approximating markers signal that a following quantity, category label, or description should not be taken as precise. Andersen's corpus-based analysis of London teenage speech identified like functioning as an approximator around numerical expressions, of the kind illustrated by his own frequently cited example of teenagers describing something being handed out "for free" [5]. His later book situates approximation as one of like's core relevance-theoretic functions, guiding the hearer toward a deliberately loosened interpretation of the phrase that follows [4]. In workplace talk, approximating markers plausibly cluster around estimates, timelines, and figures, where a speaker wants to flag that a number is not a firm commitment, though dedicated frequency counts of this specific function inside workplace corpora are limited.
Hedging
Hedging markers soften the force of an assertion, request, or evaluation. Karin Aijmer's corpus comparison of British English found that basically has shifted over recent decades toward a hedging function, appearing before a claim to signal that the statement is a simplification, a use present in a 2014 spoken corpus of British English but absent from a comparable corpus collected in 1994 [6]. Bengt Erman's study of you know in adult and adolescent conversation found a related but distinct split: adult speakers favored using the marker as what she calls a textual monitor, building coherence and smoothing transitions between parts of an utterance, a use consistent with softening or qualifying a claim, while adolescent speakers leaned more heavily on the marker's more socially and emotionally oriented functions [7].
Quotative Marking
The quotative function introduces reported speech, thought, or a paraphrased attitude, most visibly through the construction be + like ("he was like, we need this by Friday"). Variationist sociolinguistic research has tracked the spread of quotative be like closely since the 1990s. Sali Tagliamonte and Alexandra D'Arcy's study of Canadian youth speech documented be like as the dominant quotative form among young speakers, ahead of older forms such as say and zero quotation [8], and their later real-time comparison found that use of be like increased measurably within the same age cohort over a seven-year span, evidence against treating the form as a purely age-graded pattern that speakers simply abandon as they get older [9]. Isabelle Buchstaller's study of age-graded quotative behavior similarly found that be like use shifts across the lifespan in patterned ways rather than disappearing after adolescence, with different age groups showing distinguishable trajectories of uptake and retention [10].
Turn-Holding
Turn-holding, or floor-holding, markers signal that a speaker intends to continue and is not yielding the conversational turn, often functioning to buy planning time. Janet Fuller's comparative study of discourse-marker use across interview and casual-conversation settings found that markers including oh and well occurred at significantly different rates depending on the setting, functioning in part as reception markers that build coherence between turns and manage the interactional structure of who speaks next [11]. Janet Holmes and Maria Stubbe's large corpus-based study of workplace talk, conducted through the Language in the Workplace Project in New Zealand, documents how talk at work is continually shaped by the competing pressures of completing tasks efficiently and maintaining collegial relationships, a tension that plausibly motivates much turn-management and hedging behavior even though the study is organized around broader interactional strategies rather than quantified counts of individual markers [20]. In meetings specifically, where interruption carries a higher social cost than in casual talk, turn-holding markers plausibly play a more load-bearing role, though dedicated corpus studies isolating this function in workplace meetings specifically remain limited.
Frequency Patterns Across Registers
Direct, like-for-like frequency counts of individual markers across professional and casual registers are uneven in the published literature, but several corpus studies bear on the question. Michael Handford's analysis of the Cambridge and Nottingham Business English Corpus, a corpus of just over 900,000 words of spoken business discourse collected across twenty-six companies, found that business-meeting talk has its own distinct discourse-marker profile, shaped by the structured turn-taking of meetings and by the mixed first-language backgrounds of many participants [12]. Laura Di Ferrante's study of a workplace spoken corpus examined discourse markers specifically at the junctures where speakers shifted between small talk and task-focused "work talk," finding that these transitions were frequently marked explicitly and that shifts into work talk were flagged more often than shifts back into small talk, suggesting workplace speakers treat the boundary between the two modes as interactionally significant enough to signal overtly [13].
Register-based comparisons of individual markers also exist outside workplace corpora specifically. Fuller's interview-versus-conversation comparison found statistically significant differences in the rate of markers such as oh and well between the two settings, with both occurring more frequently in casual conversation than in the more structured interview setting [11]. Erman's comparison found that the overall frequency of you know did not decline in any simple way with formality; instead its functional distribution shifted with speaker age, adults favoring the coherence-building, textual use over the more socially oriented use more common among younger speakers [7]. Aijmer's corpus comparison found basically both increasing in frequency and broadening in syntactic position between 1994 and 2014, with a new preverbal hedging use emerging that did not appear in the earlier corpus at all [6].
How Listeners Interpret These Markers
Whether a discourse marker helps or hurts a speaker's perceived competence depends heavily on context and marker type, and several studies have tested listener judgments directly. Nicholas Christenfeld's experimental work on filled pauses found that they did not straightforwardly damage a speaker's perceived eloquence over short stretches of speech, and that listeners often failed to consciously register how many filled pauses a speaker had used unless specifically asked to count them [14]. This suggests some forms operate below the threshold of conscious listener judgment in brief exchanges, which complicates any simple claim that all disfluency-adjacent markers damage a speaker's image.
Attitudinal research on quotative and focuser like shows a more socially loaded pattern. Jennifer Dailey-O'Cain's study combined corpus analysis with a matched-guise perception experiment and found that be like was associated simultaneously with higher ratings on solidarity traits, speakers using it were rated as more attractive, cheerful, and friendly, and lower ratings on status traits, the same speakers were rated as less educated, less intelligent, and less interesting [15]. Isabelle Buchstaller's companion study of attitudes toward the newer quotative forms be like and go among British respondents used a matched-guise test and a social-attitudes survey, and explicitly compared its results with earlier American perception research on the same forms, extending the evidence for socially loaded evaluation of quotative markers beyond a single national variety of English [16]. Because professional evaluation in workplace settings, hiring decisions, performance reviews, promotion judgments, tends to weight status-linked traits such as competence and educated-sounding speech more heavily than solidarity-linked traits such as likability, this split in listener perception plausibly carries real professional stakes, even though the underlying experiments were not conducted in workplace settings specifically.
Charlyn Laserna, Yi-Tai Seih, and James Pennebaker's analysis of naturalistic speech samples recorded via a wearable audio device found that discourse markers such as you know, like, and I mean loaded onto a statistically distinct factor from filled pauses such as um and uh, and that this discourse-marker factor was more common among younger speakers, women, and speakers who scored higher on conscientiousness, while filled pauses occurred at comparable rates across age and gender [17]. The authors' interpretation was that discourse markers and filled pauses are not interchangeable members of a single "filler" category. They appear to index different underlying processes: filled pauses more tied to real-time cognitive load, and discourse markers more tied to social and personality factors, a distinction with direct bearing on the functional-versus-non-functional question addressed below.
Age and Generational Patterns
Some of the clearest documented age-related patterns in discourse-marker research come from variationist sociolinguistics conducted on general population and adolescent-focused samples rather than from workplace-specific studies, and that distinction matters for how confidently the findings generalize to office settings. Tagliamonte and D'Arcy's apparent-time and real-time studies of Canadian English found that be like was the dominant quotative among adolescents and young adults, and that its use increased measurably within the same age cohort as they moved into their twenties over a multi-year window, a pattern the authors interpret as evidence of ongoing language change extending further into adulthood rather than a fixed feature of youth speech that speakers necessarily abandon as they age [8][9]. Alexandra D'Arcy's broader review of the popular narratives surrounding like argues that several widespread claims about the form, that it is meaningless, that it signals inarticulacy, that it is a recent invention, do not hold up well against the historical and empirical record, and that forms of like functioning to hedge assertions and introduce reported speech have a longer history than popular commentary generally assumes [18]. Buchstaller's age-graded diagnostics study found that quotative use shifts across the lifespan in patterned rather than random ways, with implications for how a single workplace spanning multiple generations of employees is likely to show layered rather than uniform patterns of use [10].
No workplace-specific corpus study identified in this research directly measured discourse-marker frequency broken down by employee generation. Applying these general sociolinguistic, age-graded findings to differences between younger and older colleagues in the same office is therefore a reasonable extension of a well-documented general pattern, not itself a directly workplace-measured finding, and should be treated with that caveat.
Functional Versus Non-Functional Use
The research summarized above supports a distinction between functional and non-functional discourse-marker use, a distinction implicit across the literature even where individual studies do not use these exact terms. Functional use covers cases in which a marker does identifiable pragmatic work: approximating a figure, hedging a claim, introducing reported speech, or holding a turn during planning. Fraser's classification of markers by the discourse relationship each one signals is fundamentally a functional taxonomy [3], and Fuller's finding that marker rates shift systematically with speaker role and setting indicates that speakers deploy markers in ways sensitive to interactional context rather than at random [11]. Jean Fox Tree and Josef Schrock's widely cited analysis of you know and I mean argues that both markers carry identifiable core meanings tied to distinct pragmatic functions, rather than serving as interchangeable, meaningless filler, reinforcing the case that at least a substantial share of discourse-marker use is functionally motivated rather than random [19].
Non-functional, filler-like use is harder to isolate empirically, since a marker's function is not always visible from a transcript alone and researchers differ on how to code borderline cases. The clearest empirical support for treating discourse markers and filled pauses as distinct categories, rather than collapsing all disfluency into a single undifferentiated "verbal filler" bucket, comes from Laserna, Seih, and Pennebaker's factor-analytic finding that discourse markers and filled pauses form separate statistical clusters tied to different demographic and personality correlates [17]. This supports treating repeated, apparently purposeless insertion of a marker, use that does not track any of the functional categories described above and appears driven by processing load or habit rather than communicative intent, as analytically distinct from the same marker used to approximate, hedge, quote, or hold a turn. In practice, listeners appear to respond differently to the two. Christenfeld's finding that filled pauses often escape conscious notice in short exchanges [14], set against Dailey-O'Cain's and Buchstaller's findings that marked forms like be like carry strong, consciously accessible social evaluations [15][16], suggests that non-functional filler use and functional marker use are not judged by the same perceptual mechanism, even when the surface form looks identical on a transcript.
FAQs
What is the difference between a discourse marker and a filler word?
Researchers generally treat discourse markers, forms such as well, so, you know, and I mean, as carrying identifiable pragmatic functions, including signaling a relationship between utterances, hedging a claim, or managing turns, while filled pauses such as um and uh are more closely associated with real-time word-finding and cognitive processing load. A factor-analytic study of naturalistic speech samples found that these two categories load onto statistically distinct factors with different demographic correlates, supporting treating them as related but separable phenomena rather than a single undifferentiated category of verbal filler [17].
Does using "like" or "you know" make a speaker sound less professional?
The research does not support a uniform answer. Listener-perception studies of quotative and focuser like have found a consistent split in which speakers who use these forms are rated higher on solidarity-related traits such as friendliness and lower on status-related traits such as perceived education and intelligence, a pattern documented in both American and British studies [15][16]. Because professional judgments often weight status-linked traits heavily, this split plausibly disadvantages speakers in high-stakes evaluative settings, but the same markers can also serve clear, listener-recognized functions, such as hedging or signaling shared understanding, that are not inherently unprofessional.
Do younger and older employees use discourse markers differently?
Sociolinguistic research documents real age-related and generational differences in discourse-marker use, particularly for quotative like, which shows both age-graded patterns and evidence of ongoing generational change measured in real time. This evidence, however, comes primarily from general-population and adolescent-focused corpora rather than from studies measuring discourse-marker use inside workplace settings specifically, so applying these broader findings to differences between colleagues of different generations in the same workplace is a reasonable extension of the general pattern rather than a directly documented workplace finding [8][9][10].