Follow-up experiments on sentence production in Yucatec

Elisabeth Norcliffe, Katrina Housel, Juergen Bohnemeyer and I have been working on the next step in our investigations of Yucatec sentence production (see previous post). Some people have asked for more detail on the design and material and we definitely appreciate the feedback.

We started designing a series of experiments on Yucatec word order and voice choices, focusing on accessibility effects (such as the relative animacy of the agent and patient in a transitive event) trying to find evidence that distinguishes between alignment and availability accounts of accessibility effects (or provides evidence that both mechanisms are found). Due its typological features, in particular the available word orders and the way voice is marked (morphologically, without necessarily requiring word order changes), Yucatec seems perfect to test whether more accessible referents are order first (availability), aligned with the subject function (alignment, as expressed in Bock, 1986), or whether structural choices (such as a passive over an active) indicated marked or unmarked alignment between grammatical or semantic function with the relative salience of the involved referents.

Effects of accessibility on word and voice choice in transitives

Two example items for an experiment on animacy effects on word order in clauses with transitive verbs are given below, one for a CARRY event and one for a HIT/SMASH event (we also plan experiments on definiteness, and givenness). Speakers describe this transitive event to a listener (see below for detail on the task), thereby implicitly choosing whether they want to e.g. front the subject, or whether they want to use a passive to describe the event. Animacy has been shown to influence such choices in a variety of languages (Bock & Warren, 1985; Prat-Sala & Branigan, 2000, etc.). For example, English speakers are more likely to use a passive if the patient is human compared to if the patient is inanimate. English speakers are likely to produce the highest proportion of passive for the bottom-left still of the CARRY event (“The woman is carried by two horses” compared to active “Two horses carry a woman”).

The actual lexicalization (“carry” vs. “transport” or may “lift”) is likely to differ a bit between participants, and some piloting will be necessary to find sufficiently good videos that elicit the intended verbs.

Still from a CARRY event in four animacy conditions (agent = human or not; patient = animate or not)

During the pilots we will probably also see differences between Yucatec and English in terms of the lexicalization: while in English the verb “hit” can be used for punching and for smashing-into actions, this may not be the case in Yucatec. That this is feasible was shown in Wagner Cook, Jaeger, and Tanenhaus 2008. Producing Dispreferred Structure. 21st CUNY Sentence Processing Conference, Chappel Hill, NC.

Stills from a HIT/SMASH event in four animacy conditions

Each speaker will be paired with a listener, since pilots suggested that this leads to more fluent production and generally less confusion among the participants. Speakers are told that they can only use one sentence to describe the scene and that the listener has to be able to identify the scene from that description (see below). During the practice phase, speakers are reminded that they should only use one sentence to describe the scene. In our pilots, most participants followed these instructions most of the time.

At the beginning of each trial, the listener picks up a card that depicts stills of two scenes. The speaker knows that listener sees two scenes, but does not know which ones. The listener does not have visual access to the video that the speaker sees. On the listener’s card, the two stills are displayed right next to each other. Listeners are told to put the card face down onto a pile to their left, if they think the speaker described the left still, or onto a pile on their right, if they think that the speaker described the right still. After each experiment, the experimenter will record listeners’ responses, which will give us a measure of comprehension accuracy for the choices made by speakers.

The videos are created using the Poser software. The videos are pretty entertaining and participants usually love these experiments. This is probably also one of the reasons why Wagner Cook et al. had relatively low data exclusion rates (even specific verbs were elicited as intended over 80% of the time; compared to 40-60% data exclusion common for e.g. recall studies without listeners).

Effects of complexity on word order and voice choice in intransitives, transitives, and ditransitives

We’re also planning to run several experiments on the effects of constituent complexity (length) on word order (e.g. SVO vs. VSO) and voice choice (active vs. passive). We will use video descriptions following the paradigm (and adopting the stimuli) used in Wagner Cook et al. (2008). An example item for a 2 x 2 design eliciting contrasts (and hence longer descriptions) for either the recipient, theme, neither, or both is given below. Speakers describe a ditransitive scene to a listener and in doing so implicitly choose between the double object and the prepositional dative variant.

No contrasts

Simon give the bag to the guy. vs.      Simon gives the guy the bag.

Contrasts for the theme: leads to longer theme expressions

e.g. Simon gives the man the bag with stars on it.

Contrasts for the recipient

e.g. Simon gives the bag to the man with purple pants.

Contrasts for both the theme and the recipient

e.g. Simon gives the bag with the stars on it to the guy in the purple pants.

We will mix the stimuli from the complexity experiments with other video description experiments. We’ve also piloted the use of video descriptions as fillers in recall experiments, where participants first encode a message, then describe a video scene, and finally are prompted to recall the originally encoded message. One intervener trial worked pretty well for the recall studies we used (the intervening descriptions have to be long enough to let the immediate acoustic memory of the encoded message decay, while being not so complex that participants can’t recall the encoded message when prompted to).