Backchannel responses are the short signals a listener gives while someone else is talking, such as \u201cmm\u201d, \u201cyeah\u201d or \u201cokay\u201d. The term goes back to Victor Yngve in 1970. In speech these signals run alongside the speaker\u2019s turn; in a typed chat each one has to be sent as a message of its own.<\/p>\n\n\n\n
Most of the listening in a spoken conversation is audible and almost free. A typed conversation has no place to put it except the same column as everything else, which changes what those small signals look like on the record.<\/p>\n\n\n\n
The idea comes from a 1970 paper by Victor Yngve, \u201cOn getting a word in edgewise\u201d, given at the sixth regional meeting of the Chicago Linguistic Society. Jenny Yau-ni Wan\u2019s 2018 study in the Journal of Intercultural Communication<\/a> summarises it this way: the term back channel implies that two channels in a conversation operate at the same time. The main channel carries what the person holding the floor is saying. The back channel is where the listener gives information without claiming the turn.<\/p>\n\n\n\n Wan notes that Yngve did not limit the back channel to items such as \u201cyes\u201d and \u201cuh huh\u201d. He included short comments and short questions, as long as the listener was not taking over. Later writers moved the boundary; Wan notes that Duncan and Fiske, drawing on Yngve, extended the term to sentence completions, requests for clarification and brief statements. What they share is the position: a backchannel happens inside someone else\u2019s turn and leaves that turn with them.<\/p>\n\n\n\n The other names in the literature say the same thing from different angles. Marlene B\u00f6ttcher and Martina Rossi, writing in Frontiers in Communication in 2025<\/a>, list \u201clistener feedback\u201d, \u201cminimal responses\u201d and \u201cacknowledgment tokens\u201d as labels for the same thing, and define the backchannel as a response produced during the current speaker\u2019s turn to show active listening without interrupting.<\/p>\n\n\n\n Val\u00e9ria Krepsz and colleagues, in a 2022 corpus study published in Linguistica Silesiana<\/a>, analysed backchannels in 10 Hungarian conversations. Four forms made up a little more than 80 per cent of everything they found: a hummed \u201c\u00fch\u00fcm\u201d at 34.4 per cent, laughter at 24.1 per cent, \u201cigen\u201d (yes) at 11.7 per cent and \u201caha\u201d at 10.5 per cent. Most responses, 87.7 per cent, were a single item rather than a phrase.<\/p>\n\n\n\n They are also very short. In the same corpus a hum lasted 273 milliseconds on average and a backchannel \u201cigen\u201d 349 milliseconds. On average each speaker produced 27 of them per conversation, with wide variation between conversations. The study is narrow: 10 Hungarian conversations, recorded with fieldworkers taking part.<\/p>\n\n\n\n Krepsz and colleagues also cite an earlier analysis of 1,155 American English conversations in which 19 per cent of utterances were backchannels, and note that the frequency appears to depend on culture. Wan\u2019s own study compared Hong Kong Chinese and native English speakers in conversation with each other and found that the two groups leaned on different kinds of backchannel. Neither figure transfers to a chat, but both make the same point: in speech, listening produces a steady stream of tiny, cheap signals.<\/p>\n\n\n\n A single word can do several jobs. Krepsz and colleagues cite work by Gravano and colleagues showing that context decides which function a backchannel is performing, and that a word such as \u201cokay\u201d can perform more than one at once.<\/p>\n\n\n\n B\u00f6ttcher and Rossi pushed that further. In 39 minutes of German face to face conversation they found 634 short conversational items, of which 151 were backchannels. The rest were items the speaker used to hold the floor while planning what to say next, and the two groups shared words: \u201cja\u201d, \u201cokay\u201d and \u201cgenau\u201d turned up in both. What separated them was everything around the word. Two thirds of the backchannels, 66 per cent, were produced during the other person\u2019s turn and in overlap with their speech. Backchannels came with a head movement 30 per cent of the time, almost always a nod, against 9 per cent for the speaker\u2019s items. Duration and pitch differed too.<\/p>\n\n\n\n Put plainly: in their recordings, a listener\u2019s \u201cokay\u201d and a speaker\u2019s \u201cokay\u201d were the same word and sounded different, arrived at different moments and came with different movements. Those are the differences the researchers measured between the two.<\/p>\n\n\n\n A typed chat has one channel. There is no way to say \u201cmm\u201d underneath someone else\u2019s message while they are still writing it. The listener either sends something or sends nothing, and whatever is sent appears as its own entry, in sequence, after the message it responds to.<\/p>\n\n\n\n That changes three things about the signal. It is no longer simultaneous, which was the defining feature in Yngve\u2019s picture. It is no longer cheap, because sending \u201cok\u201d takes a deliberate action where a spoken \u201cmm\u201d costs a fraction of a second. And it is no longer accompanied by anything: the timing inside the other person\u2019s turn, the pitch and the nod that B\u00f6ttcher and Rossi measured all fall away, leaving the bare word.<\/p>\n\n\n\n A reaction, such as a thumbs up attached to a message, is the closest a chat comes to a back channel, because it answers without adding a new line of text. How reactions appear once a chat is exported differs by platform, and this article makes no claim about it. The broader point about what a transcript keeps and loses is the subject of a separate piece on grounding in communication and what a chat keeps<\/a>.<\/p>\n\n\n\n Picture the same kind of exchange typed. One person tells a long story across several messages, and the other person\u2019s listening shows up as an \u201cok\u201d here and a \u201cyeah\u201d there, each on its own line between the story\u2019s messages. Listening that was continuous in speech becomes a handful of separate, countable turns.<\/p>\n\n\n\n Countable is where misreading starts. On the page, a short reply is the whole of the listener\u2019s visible response, with nothing alongside it to read it against. The reasoning that makes a one word answer seem to say more than it does is set out in the piece on Grice\u2019s maxims and what a short reply implies<\/a>. The research above adds a narrower point: in speech, the same word could be told apart as listening or as holding the floor by its timing, pitch and movement, and a chat removes all three.<\/p>\n\n\n\n Wan lists up to 13 functions researchers have attributed to backchannels, from encouraging the speaker to continue, through claiming understanding, to displaying a lack of interest or a lack of understanding. A typed \u201cok\u201d can be standing in for any of them, and the record does not say which. Time between messages is a separate signal again, covered in the piece on reply speed and how spoken turn taking works<\/a>.<\/p>\n\n\n\n One thing visible in your own side of a conversation is how much of it consists of these short acknowledging messages, and where they cluster. One step in Portraitor\u2019s flow, as its home page describes it, is picking the one person to analyse; that person\u2019s messages are read while everyone else stays context. If that person is you, the messages being read are your own side of the chat. The piece on analysing your own messages<\/a> explains what choosing yourself means, and Portraitor\u2019s home page<\/a> shows the full flow.<\/p>\n\n\n\n Every study cited here measured speech. None of them measured text, and the figures for frequency, duration and overlap describe spoken conversation in particular languages, with particular speakers, in recorded settings. They explain what was lost when a conversation moved into typing; they do not say how often people send \u201cok\u201d in a chat or what they mean by it.<\/p>\n\n\n\n The home page does not say how a one word reply is weighed against a long message in a portrait, or whether a run of short acknowledgements counts differently from a single one. That is a gap in what Portraitor publishes, and this article leaves it as a gap rather than guessing. What a chat analysis cannot settle about another person more generally is covered in the piece on what a chat analysis cannot tell you<\/a>.<\/p>\n\n\n\nWhat a spoken backchannel looks like when someone measures it<\/h2>\n\n\n\n
The same okay, from a speaker or from a listener<\/h2>\n\n\n\n
In a chat, the listener has to send a message<\/h2>\n\n\n\n
Reading a run of short replies in a written record<\/h2>\n\n\n\n
Where a transcript stops being able to tell you<\/h2>\n\n\n\n