updated 18 March, 2013
In the 3rd Presidential Debate (Oct 22, 2012), one of President Obama's most talked about retorts to Governor Romney was the following:
Well, Governor, we also have fewer horses and bayonets — (laughter) — because the nature of our military’s changed. We have these things called aircraft carriers where planes land on them. We have these ships that go underwater, nuclear submarines.
Notably (to a linguist), Obama appeared to link two clauses via a resumptive pronoun. The two clauses are in bold, the resumptive underlined. Resumptive pronouns are a grammatical device found in many languages of the world (1), but they are reliably not judged as acceptable in English (2; despite seemingly being often produced).
A look at how Twitter users reported Obama's zinger provides some insight into how sentences containing resumptives are encoded. Although many users might copy and paste the text from a transcript, others will simply try to reproduce what they heard from memory. Verbatim memory is difficult to achieve in most contexts, since memory for exact form appears to be rather labile. Indeed, when people do reproduce exact form, it appears they mostly don't rely on any surface representation (i.e., anything like a 'mental transcript'); instead speakers reformulate based on the gist of what was heard and the recently activated lexical items (3). In other words, recollection of linguistic form from memory involves some reformulation of the underlying message. Thus seeing how ungrammatical sentences are repeated can give us insight into their underlying representation.
I looked at about 400 tweets (439, exactly), posted from 7PM PDT to about 8:30PM PDT. I tried to exclude obvious retweets. I classified what was said into four categories:
In the case of Obama's resumption, an overwhelming tendency (~40%) for Twitterers to desubordinate the resumptive-containing clause is consistent with Prince (1990)'s observation that resumptives are felicitous if they are part of an additional predication - as in a non-restrictive relative clause or an appositive. (It might also favor a view according to which resumptive pronouns in production do not reflect a production choice strictly about realizing the tail of an A-bar chain, but rather reflect a higher level decision about combinining clauses).
Although speakers preferred a desubordinating strategy to a gap strategy about 2:1, when speakers chose a gap, something interesting happened. In 80% of the gap-strategy tweets, speakers remembered a relative clause headed by the complementizer that (and not the relative pronoun where).
I then wondered whether speakers who tweeted the remark as a quotation were any more accurate. I looked for cases where Obama's remark was enclosed in quotation marks (this is necessarily heuristic, since quotation marks sometimes merely indicate attribution, and are not always a claim to exact reporting). The graph below shows that they are not. Notice the distribution of response categories is nearly identical.
In Potter & Lombardi (1990), the authors argued that verbatim recall of a target sentence reflected regeneration of the sentence because of a lexical lure effect: if words not contained in the original sentence were activated that were related to words in the original sentence, then those words would sometimes be inserted into the recalled sentence. In their experimental demonstrating, this incidental activation was achieved by interrupting the recall portion of the task with a word list recognition memory task, and this word list sometimes contained a lure (e.g., palace if the target sentence contained the word castle).
There is some evidence of lexical luring in the Obama tweets. In particular, many users replaced 'these things' with 'these ships' - where the word 'ships' actually appeared in the next sentence. The following chart shows that individuals mostly recall the words correctly, but occasionally inserted 'these ships' for 'these things [called aircraft carriers].' Less often they misrecalled 'these things' as 'this thing'; or they simply dropped the demonstrative. [The fact that 'ships called aircraft carriers' could substitute for 'things called aircraft carriers' is consistent with the view that a simple associative relation between target and lure words is not sufficient, and that conceptual relatedness also matters, as Lee & Williams (1997) suggested of the original Potter & Lombardi results].
I am contemplating further analyses, though have not yet performed them (including looking at different time slices; and comparing other remarks). However I think this simple initial analysis hints at the potential applications of Twitter for testing ideas in language production and memory.
"Resumptive pronouns and the not-quite-verbatim memory of Twitter users" by Matthew Wagers is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.