Fresh of the press: Preliminary Mayan results … [Norcliffe & Jaeger]

The orange juice is still warm, the cafe con helado barely melted, the Mexican music that has been playing on repeat for the last couple of hours still swings mind-numbingly in my fried brain (it’s VERY hot and humid here), and here we are: letting you, dear reader [sic], know what the world is waiting for: is there probability-sensitive morphosyntactic production in Yucatec Mayan (similar to English, cf. Frank & Jaeger, 2008-CUNY, 2008-CogSci; Jaeger, 2006-thesis, 207-LSA; Levy & Jaeger, 2007; Wasow et al., in press)? This is a follow-up on a recent post.

Over the last week, we (Elisabeth Norcliffe, Ling. Dept. Stanford University and I) collected data from somewhat over 30 Mayan speakers (most were students between the age of 18-23, but we also had some older participants from close by villages, and other folks who heard about the experiment), each participating in one of several 30-45 minute production experiments involving video/video description and recall tasks (of course, fully counter-balanced across lists, etc.; the whole psycholinguistic shebang). After an initially somewhat disappointing turn out, we could barely stop the floods of people who wanted to participate, once word got around. This was mostly due to the wonderful help of the UNO staff (in particular, the librarian, Marta Poot). While this worried me first (was our payment too high?), it seems that people were simply curious and eager to help. Actually, some participants didn’t even know that would get paid and it took a lot of convincing that they should really accept the payment because otherwise we would get into trouble.

We ran a couple of pilots trying a few different paradigms (repeating aloud or not repeating the to-be-recalled stimulus; the type of intervener task between the encoding and recalling of the stimui, e.g. video description, multiple picture naming, or spontaneous story telling based on picture input). Unsurprisingly, there was a lot of variation in terms of how easily participants found it to recall the sentences, how nervous they were (most didn’t seem very nervous), how bored they were (the speakers usually seemed pretty entertained by our hilariously funny video clips – thanks to our super RAs Katrina Housel and Irene); how fluent the speakers were, etc. Generally, it seems that female participants were MUCH better at this – the difference was striking in terms of fluency, willingness to get into the task, and the ability to successfully recall the stimuli.

We also started recording instruction in Mayan, but so far decided to go with instructions in Spanish, since we were still changing around to many of the details (all participants were fluent in Spanish). In the future we will use recorded instructions in Mayan and video clips of other participants (who have kindly agreed to be video filmed for that purpose).

In the version we ended up running most participants in, speakers were paired with a native speaker listener, which definitely helped because production were louder, somewhat clearer, and speakers seem to feel less confused by the task. The target trials contained a subject-extracted relative clause modifying the object of a transitive matrix clause, kinda like:

Táan  inw=il-ik   le=hiràafa  (k-uy=)akat-ik       le=máak=o
PROG  A1=see-INC  DF=giraffe  (IMPF-A3=)chase-INC  DF=person=D2

‘I’m watching the giraffe that is chasing the person.’

These types of transitive relative clauses have an optional piece of morphology. To be precise, most of the literature on Mayan relative clauses seems to assume that, for transitive subject-extracted relative clauses in Yucatec Mayan, the morphology is obligatorily omitted (leading to the so called agent-focus voice), but verrrry recent research suggests that they would be variation between the presence and absence of the k-uy, Gutierrez Bravo, ms.; Norcliffe, 2008-SSILA, 2008-LSA).

Our mission (which we accepted) was to test (a) whether one can run recall studies on Mayan and what challenges would have to be overcome to do so, and (b) more specifically, in this experiment, whether the optional piece of morphology is linked to redundancy-moderation as, for example, the English relativizer that (Jaeger, 2006-thesis). To test this, we manipulated the probability of a relative clause (and hence the information density at its onset). The NP modified by a relative clause either was an indefinite, definite, or universally quantified NP, and the original stimulus (to be recalled) either had the optional morphology or didn’t. In English these determiner/quantifier choices have been shown to correlate with the probability that the NP will be modified by a relative clause (Wasow, Jaeger, and Orr, in press), and since these correlations are assumed to be based on pragmatics (e.g. the uniqueness requirement of definite determiners, ibid), we hypothesized that the same correlation between relative clause probability and the determiners may hold for Mayan.

There were 24 target trials and 24 filler trials in this experiment. Each trial consists of three parts:

• The speaker hears a sentence and can -but does not have to- repeat it aloud. When they are ready to continue, the participant presses SPACE (sometimes this was done by the experimenter, if participants seem to be unfamiliar with computers)
• The speaker sees a video (for the target trials, this was a transitive event such as one guy punching another guy) and describes it to the listener “in one sentence”.
• The speaker hears a cue (a part of the to be recalled sentence) and recalls and says aloud the entire to-be-recalled sentence.

Participants were instructed that trials had this triplet structure and that they were 48 such trials.

A preliminary analysis of 19 of our subjects, suggests that we will find the effect we hypothesized based on English data on the connection between redundancy and reduction. In repeating the stimuli, participants seem to be more likely to insert the optional k-uy morphology in relative clauses that modify an indefinite NP than in relative clauses modifying a definite or universally quantified NP (of course, they are also more likely to repeat k-uy if it was present in the original stimulus, but these effects are independent of each other). This effect, if confirmed after the data from all participants is analyzed is predicted by theories of efficient communication (e.g. Uniform Information Density, Genzel & Charniak, 2002, 2003; Jaeger, 2006; Levy & Jaeger, 2006). Of course, there’s a LOT more work to be done before we can trust these results – as always in recall studies, we found a lot of variation beyond what we interested in an the current analysis simply excludes all those trials (hardly a desirable solution).

In any case though, we’re excited to see that the methodology seems to work and that there is variation, as claimed by e.g. Gutierrez Bravo and Norcliffe) and contrary to the traditional treatment of those structures. We also find that insertion of the optional morphology (k-uy) is actually more frequent than omission. This may be due to the slowly spoken (and over articulated) stimuli we used (something we plan to change in the future), but it is potentially interesting, as it seems from the literature on this agent-focus marking that omission is more frequent or even obligatory for transitive subject-extracted relative clauses.

Ideas and feedback on this post are especially welcome. Maybe you see flaws in our design, things we should change, or you have questions or want more detail on the design or the stimuli. Please let us know.