When analysing spoken English, it doesn’t take long to encounter discourse markers, the single words or phrases that speakers commonly use to mark their stance or organise their message. Common discourse markers include well, now, you know and i mean. In the April 2014 issue of English for Specific Purposes, Francisco Javier Fernández-Polo examines the discourse marker i mean in conference presentations included in the ELFA corpus. This subcorpus includes 34 conference presentations in English by speakers of 21 different first languages. Recorded at universities in Finland, the data consist of naturally occurring English used as a lingua franca (ELF) in academic settings.
Fernández-Polo’s study is qualitative, involving a close analysis of a small number of cases toward determining the functions of i mean in context. There are only 56 occurrences of i mean in this conference presentation subcorpus (94,314 words1), and Fernández-Polo takes 48 of them into his analysis. He classifies these into four different categories – correcting mistakes and dysfluencies; enhancing clarity and explicitness; organising text; and marking certainty and salience (see Table 1 below). Examples of each are discussed in turn.
A striking finding from the paper concerns the wide inter-speaker variation in the use of i mean. Fewer than half of the 34 presenters use i mean at least once, with a single speaker producing 20% of the occurrences, and five speakers contributing two thirds of all hits. To see if a different distribution might be found in similar English as a native language (ENL) data, Fernández-Polo consulted the monologic lectures in the American MICASE corpus. He found that i mean occurs in the MICASE lectures with the same standardised frequency (5 per 10,000 words) and with similar inter-speaker variation – one speaker in MICASE produced 27% of occurrences, with 14 speakers producing 60% of hits. It thus appears that the choice of discourse markers varies a lot based on a speaker’s preference or habit.
The fluency connection: planned vs. unplanned speech
The largest category in Fernández-Polo’s taxonomy is “correcting mistakes and dysfluencies”, which he identifies in 20 of the instances of i mean. Considering that conference presentations are relatively well-planned, what could we expect to find in the ELFA corpus conference discussions? This isn’t discussed in the article, but I checked for i mean in the corresponding conference discussion subcorpus and found a statistically significant difference. While there are 6 occurrences of i mean per 10k words in the conference presentations, the frequency jumps to 19 hits per 10k words in the conference discussions. Interestingly, many of the same presenters are active in the discussion files as well, which were typically recorded in the same conference sessions.
This connection of i mean with fluency and unplanned speech also ties into my PhD research. I’m also using data from the ELFA corpus to investigate (dys)fluency features in spoken academic ELF, but my descriptive methodology employs Linear Unit Grammar, or LUG, in which written or spoken texts are analysed as a linear sequence of chunks (see my intro to LUG: part 1 and part 2). My corpus of fully LUG-annotated transcriptions currently stands at 32,000 words covering 3.5 hours of spoken ELF data, with roughly half of the data from monologues and half from polylogic interaction (with multiple speakers). In this data, I find 31 occurrences of i mean. Only five of them are found in monologues (all conference presentations), with 26 occurring in polylogic data. Of these, 23 hits occur in highly interactive student group work, where a presentation is being planned. How does i mean function in this unplanned, spontaneous speech?
Just another chunk in the chain
In LUG, i mean is treated as a single chunk that organises interaction (OI chunk), since it’s not a chunk that’s normally found in written texts (we might instead use something like “in other words…” or “that is to say…”). The benefit of working with LUG is that it is capable of fully parsing speech, revealing where these organising chunks occur in relation to the incrementation of meaning. The only hierarchical level above the single chunk is the Linear Unit of Meaning, or LUM. This clause-like unit contains each complete Message (M) chunk along with its optional supplementary chunks. Being able to see where i mean occurs in relation to these Linear Units of Meaning quickly illustrates how this organising chunk functions in managing fluency in unplanned speech.
Among the 31 occurrences of i mean in my LUG-parsed data, 19 of them occur between two Linear Units of Meaning. In some of these cases, i mean appears to function within Fernández-Polo’s category of “enhancing clarity and explicitness”. Consider the following example (you can hover the mouse over the chunk labels to view their full description):
M-in the netherlands
MSi come from
+Mevery single tree is protected
M-for every tree
+Mwould have to ask a permission
In this utterance, a PhD student explains that permission must be requested in order to cut down any single tree in the Netherlands. The i mean chunk falls directly between the two complete Message (M) sequences as a clear explicitation signal. In this respect, it all functions to create a fluent message, even if details of the chunks don’t conform to expectations of “well-formedness”. It was also clear for the listeners, as reflected in the yeah coming right after the first complete Message chunk – this came from one of the other participants2.
In addition to this kind of function, i mean chunks can be found in this inter-M position playing a role in turn-taking. In one example, yeah | i mean is used to take the floor and start a new Message chain. Alternatively, the chunk is found in the sequence of at least OI | so OT | i mean OI | yeah OI | (laughter) OI | at the end of a complete Message sequence, part of a series of signals that the speaker was relinquishing the floor. With 19 occurrences, this inter-M position is the most the common position of i mean in my data. The other 12 cases, however, are more interesting from a fluency perspective.
Signalling a break in the chain
Unlike conventional grammars, which parse a sentence in a hierarchical tree of nested constituents, Linear Unit Grammar parses a string of text (spoken or written) as a linear chain of chunks, reflecting the way we produce and process speech in real-time. Each of those chunks predicts or prospects what is likely to follow in order to complete the Message chunks that have just come before. When planning an utterance in real-time, we sometimes have to “break the chain” of prospection and start off in a new direction. As a courtesy to our listeners, we usually signal this break in the chain. And here we can also find i mean, within the chain of Message chunks, before a point of completion has been reached.
Among my 12 cases of i mean that appear after prospection is started – but before a point of completion is reached – seven of these are self-interruptions. That is to say, the chain of prospection is interrupted “mid-stream” and a new chain is established:
M-do you use poly-
M-are polypores used
+Mas indicator species
This is a straightforward rephrasing, but the i mean serves as a signal of the interruption of what has come before. This is usually performed on one’s own stream of speech, but in one example a speaker uses i mean in a chain that interrupts another speaker and takes the floor with but OT | maybe OI | maybe OI | it’s MF | i mean OI | before continuing.
As with the above example, other instances of pre-completion i mean can be found with a rephrasing or revising function. In these cases, i mean might precede a Message Supplement (MS) or Message Revision (MR) chunk. There’s a clear example of this in Fernández-Polo’s Example 3:
M a district has one or two hospitals
M-everything else is either a hospital
MRa health centre
Here the speaker has mistakenly used the term “hospital” where he meant to say “health centre”, and i mean again signals this brief break in the chain before linking back up with the completing chunk, “a dispensary”. So, even though this extract can be interpreted as a “repair”, it still shows a fluent management of linguistic resources with mindfulness of the listeners’ expectations based on what has just been uttered. None of these observations are intended to be criticism of Fernández-Polo’s analysis, which gives a good functional account of these discourse markers. I’m suggesting that structural accounts of these chunks, based on a linear analysis of meaning-making, opens up additional observations of how speakers manage the task of creating fluent, intelligible speech.
Update 8.5.2014: For further analysis of the functions of I mean in the ELFA corpus conference data, see Mauranen 2012: 186-191. Among the 67 instances of I mean analysed there, the majority of instances (67%) can be seen as “tack changes”, or signals of a new direction in the discourse.
1 Fernández-Polo cites a word count of “over 97,000 words” without mentioning how this figure was derived. My count in AntConc (with markup excluded) is 94,314 words for the conference presentations (56 hits for i mean) and 74,039 words for the conference discussion subcorpora (139 hits for i mean). These figures were used for a log-likelihood test yielding a result of 59.42, or p<0.0001.
2 LUG doesn’t treat the utterance boundary as necessary information for performing chunking, since a great deal of interaction is co-constructed between participants in the form of collaborative completions (“finishing each others’ sentences”). Thus, more than one speaker regularly contributes to a chain of prospection.
Fernández-Polo, Francisco Javier (2014) The role of I mean in conference presentations by ELF speakers. English for Specific Purposes, 34, 58-67. DOI: 10.1016/j.esp.2013.09.006.
Mauranen, Anna (2012) Exploring ELF: Academic English shaped by non-native speakers. Cambridge Applied Linguistics Series. Cambridge: Cambridge University Press.