Hi Florian, Ah, yeah turning it around to comprehension-to-production priming would definitely do the trick! I hadn't thought of that. Unfortunately right now I don't have that data because I haven't coded the interlocutor speech, so I only have production-to-production data (and I exclude observations where the target should be considered "primed" by some other interlocutor). Here's what I do have: 122 interviews from the Philadelphia Neighborhood Corpus, with /ing/-vs-/in/ coded for all types of ING (not just verbs, also nouns etc). There's an average of 39 tokens for each speaker, total N=4803, but some speakers have a lot more data than other. So for example if I exclude speakers with fewer than 50 tokens, there are 45 speakers with a mean of 83 tokens each.
Let me know if there's any other info I can provide or if you'd like me to send the data your way. I'm starting to think that even with what I originally thought was a lot of data (ditched Buckeye as insufficient a while ago!), it's still not going to be possible to do what I'm trying to do. In which case I may have to table the question until I get the interlocutor data coded to try out your suggestion... Thanks! Meredith On Sat, Jan 25, 2014 at 10:28 PM, T. Florian Jaeger <ti...@csli.stanford.edu > wrote: > Hi Meredith, > > ha, i love that question (yours). I recall that in the preparation of > Jaeger & Snider 2013-Cognition, we looked into whether we could answer that > question in the Switchboard, but there were too few speakers with > sufficiently many sentences. probably something like Buckeye would be the > corpus to use. And using -ing as you suggest probably gives you much more > data than the ditransitives we had. There's some preliminary evidence in > favor of a positive answer to your question. In Jaeger 2010 -CogPsych, I > discuss evidence from complement clause priming. I point to the finding > that complementizer presence primes more strongly in conversational speech > data (where it's rare) but less strongly in sentence production studies > (where it's the majority choice). If you search that paper for "syntactic > priming", you'll find this in the middle of my other ramblings. > > Ok, but back to your question.What type of data do you have? If you can > solidly estimate the production bias of a speaker (ideally for a variety of > a verbs), e.g., when not talking to anyone or when talking to interlocutor > A, you should be able to use that to predict how surprising primes during > *comprehension* are when the same speaker listens to interlocutor B. By > checking how much the speaker then deviates from her own baseline after a > comprehension prime, you should be able to tease things apart, no (i.e., in > this case, you'd be looking at comprehension-to-production priming)? This > might make sense anyway -- in Jaeger & Snider 13 it seemed like the > surprisal effects might originate in comprehension, though we discuss > alternative interpretations, too (see Section 5.4). Are you wondering > exclusively about production-to-production priming? > > Florian > > > On Fri, Jan 24, 2014 at 9:57 AM, Meredith Tamminga < > tammi...@babel.ling.upenn.edu> wrote: > >> Hi Florian, >> >> Thanks for the response. I'm reassured that you think a random by-speaker >> intercept will filter out the speaker baseline usage effect, since I've >> been using that approach in other portions of this project. I've done a bit >> of simulation testing this out and it does seem to be fine, at least for >> the distributions of between-speaker differences I'm working with. >> >> But, I'm still not sure how to then separate out and identify any >> *true*effect of the speaker baseline rate on persistence. In Jaeger & Snider >> 2007 >> you ask the very interesting question "probable given what?" about how >> surprisal affects the strength of the persistence effect. What I really >> want to ask here is whether there is an effect of an individual speaker's >> habitual usage in terms of whether /in/ or /ing/ seems more surprising, or >> whether the speakers generally share an evaluation where, say, /in/ >> provokes a larger persistence effect because it is nonstandard. In other >> words, could one aspect of the surprisal effect on persistence be "probable >> given the speaker's own preferences"? The problem, then, is to disentangle >> the uninteresting effect (that speakers with more extreme baselines have >> more apparent clustering as a matter of course) from the potential effect >> of interest (an effect where different speaker baselines produce greater or >> lesser degrees of persistence *beyond* the amount expected if there were >> no priming at work). >> >> I'm not aware of any other work in this area that has really grappled >> with this issue yet -- since it's a question that I don't believe has been >> asked. Your work on the preference of individual verbs for certain >> constructions doesn't face this problem, I don't think, because the prime >> and target aren't constrained to be drawn from the same distribution. >> >> One possibility I've been considering is some sort of transformation of >> the speaker baseline, so that in simulated data generated through a set of >> binomial trials with the weights of the empirical speaker baselines there >> would be no apparent relationship between baseline and amount of >> clustering. I haven't worked out quite what transformation that should be, >> and I'm not entirely convinced it's a good approach. >> >> Any further thoughts are much appreciated! >> >> Meredith >> >> >> >> >> On Wed, Jan 22, 2014 at 11:53 PM, T. Florian Jaeger < >> ti...@csli.stanford.edu> wrote: >> >>> Hi Meredith, >>> >>> I think that a mixed model with a random by-speaker intercept should >>> successfully filter out the speaker baseline usage effect (though I haven't >>> simulated how well this works, depending on the distribution of >>> between-speaker differences). This approach is used in e.g., Jaeger (2006, >>> 2010) and Jaeger & Snider (2013, Study 1). A slightly different approach is >>> presented by Reitter (2006 and follow up) who focuses on the idea of >>> distance-based decay of priming (rather than a priming main effect). There >>> are variety of alternative approaches discussed in the literature. For >>> example, researchers from Frank Keller's lab have used a first half vs. >>> second half of conversation comparison. I forgot the reference to the >>> original proposal they cite, but if memory serves right Dubbey et al. 2010 >>> discuss and use this approach. >>> >>> You might also want into Gries 2005 and Szmrecsanyi 2005 though I don't >>> recall whether they in any way controlled for the baseline issue. >>> >>> HTH & sorry if I misrepresented anyone's work, >>> >>> Florian >>> >>> Gries, S. T. (2005). Syntactic priming: a corpus-based approach. Journal >>> of Psycholinguistic Research, 34(4), 365–99. doi:10.1007/s10936-005-6139-3 >>> Jaeger, T. F., & Snider, N. E. (2013). Alignment as a consequence of >>> expectation adaptation: Syntactic priming is affected by the prime’s >>> prediction error given both prior and recent experience. Cognition, 127(1), >>> 57–83. doi:10.1016/j.cognition.2012.10.013 >>> Reitter, D., Moore, J. D., & Keller, F. (2006). Priming of Syntactic >>> Rules in Task-Oriented Dialogue and Spontaneous Conversation. In >>> Proceedings of the 28th Annual Conference of the Cognitive Science Society >>> (pp. 1–6).Sturt, Patrick, Frank Keller, and Amit Dubey. "Syntactic >>> priming in comprehension: Parallelism effects with and without >>> coordination." *Journal of Memory and Language* 62.4 (2010): 333-351. >>> Szmrecsanyi, B. (2005). Language users as creatures of habit: A >>> corpus-based analysis of persistence in spoken English. Corpus Linguistics >>> and Linguistic Theory, 1(1), 113–150. doi:10.1515/cllt.2005.1.1.113 >>> >>> >>> >>> On Tue, Jan 21, 2014 at 11:54 AM, Meredith Tamminga < >>> tammi...@babel.ling.upenn.edu> wrote: >>> >>>> Hello R-lang, >>>> >>>> I have a dataset consisting of observations of the binary variable ING >>>> (e.g. workin' vs working) from conversational speech. I am interested in >>>> the priming effect on this variable: to what extent does the most recent >>>> prior observation (the prime) affect the outcome of the current observation >>>> (the target)? To assess this, I am using the dependent variable of "rep" -- >>>> does the target have the same value (/in/ or /ing/) as the prime? I intend >>>> to use this as the response variable for logistic regression. Available >>>> predictors are prime.var (is the prime a token of /ing/ or /in/?), >>>> target.var (is the prime a token of /ing/ or /in/?), same.word (are prime >>>> and target the same lexical item?), same.gram (do prime and target have the >>>> same grammatical status?), log.lag (log2 of the distance in seconds between >>>> the prime and the target), and spkr.mean (the mean rate of /ing/ use by the >>>> speaker who produced the token). There are 4879 observations (prime/target >>>> pairs) from 90 different speakers who have very different mean rates of >>>> /ing/ use. The prime and the target are always from the same speaker. >>>> >>>> So here's the problem: I can't figure out how to account for the fact >>>> that repetition of the variant is more likely as a matter of course (rather >>>> than as a matter of priming) when speakers have rates of /ing/ use near 0% >>>> or 100%. I am particularly interested in testing the hypothesis of an >>>> interaction between prime.var and spkr.mean: that for speakers who have low >>>> /in/ rates, /in/ is a stronger prime than /ing/, whereas for speakers who >>>> have low /ing/ rates, /ing/ is a stronger prime than /in/. If I include >>>> prime.var * spkr.mean in the model, though, the effect I am looking for is >>>> obscured by the trivial fact that speakers who use /ing/ a lot will >>>> naturally be likely to have another /ing/ target after an /ing/ prime, and >>>> speakers who use /in/ a lot will naturally be likely to have an /in/ target >>>> after an /in/ prime. What I'm trying to figure out is whether there is an >>>> interaction of prime.var and spkr.mean *beyond* what is expected just given >>>> that the prime and target share a bias. >>>> >>>> If anyone can make any suggestions for how to proceed here it would be >>>> much appreciated. Please let me know if there's anything I can clarify or >>>> add. >>>> >>>> Thanks! >>>> Meredith >>>> >>> >>> >> >