Matlock, T. (2004). Fictive motion as cognitive simulation. Memory And Cognition, 32(8), 1389–1400. doi:10.3758/BF03206329

Summary

In this paper, Matlock asks whether mental simulation is involved in processing sentences that involve motion words but which do not actually describe literal motion (the term for this type of motion is fictive motion). In a series of four quite elegant experiments, Matlock shows that people take longer to respond whether a sentence involving fictive motion was related to a story when the story described long/slow/difficult travel as opposed to short/fast/easy travel. The results suggest that people may be constructing a mental model of the scenario described by the story, and that then to understand the target sentence they simulate from this mental model. When the model actually involves motion that that is longer, slower, or more difficult, people accordingly run simulations that mimic those properties, and thus take a longer time to respond.

Methods

Experiment 1 used stories involving long-distance vs. short-distance travel. For example, the story might be about someone driving across the desert to visit a relative, and the desert is either large or small. The target sentence would then be something like “the road crosses the desert”. In the control study, they instead used target sentences like “the road is in the desert”. In this case they found participants took about 400ms longer to judge the target sentence in the long-distance stories.

Experiment 2 used stories involving slow vs. fast motion. For example, the story might be about someone walking or running along a path. The target sentence would then be something like “the path follows the creek”. In the control study, they instead used target sentences like “the path is next to the creek”. In this case they found participants again took about 400ms longer to judge the target sentences in the slow-motion stories.

Experiment 3 used stories involving difficult vs. easy terrain. For example, the story might be about someone driving along a coastline which is either jagged and hilly or straight and flat. The target sentence would then be something like “a road runs along the peninsula”. In the control study, they instead used target sentences like “there is a road along the peninsula”. In this case they found participants took about 350ms longer to judge the target sentences in the rough terrain stories.

Experiment 4 used stories that didn’t actually involve motion at all—for example, a story describing an earthquake fault along terrain that is either difficult or easy. The target sentence would then be something like “an earthquake fault runs across the valley”. There was no control experiment here, but I would expect the results to be similar. They found participants took about 200ms longer to judge the target sentences in the difficult terrain stories.

In all the experiments there were “filler” scenarios that did not involve fictive motion but did involve regular motion.

Algorithm

n/a

Takeaways

It’s not surprising to me that people might use something like mental simulation to help them interpret sentences, but I am surprised that the motion words have as strong of an effect as they do: i.e., that “the road crosses the desert” takes longer to read for a long-distance scenario than a short-distance scenario, while “the road is in the desert” is read at pretty much the same speed. I’m really curious to think about how this type of mental simulation serves interpretation of language from a computational point of view. Is it something like: we construct this mental model, and then when asked if “the road crosses the desert” is true, we look at our mental model and verify (perhaps visually) that it does in fact cross the desert (which would then take longer for longer distances)? Whereas, for “the road is in the desert”, all that is required is to verify that the road is in the desert, which should take the same amount of time regardless of how big the desert is?

I do wonder whether these results apply in the general case of interpreting language. For one, participants were explicitly told to imagine the scenario as they read it (and the first sentence of each story was of the form “imagine a road”). If they weren’t prompted to use imagery, would there still be an effect? Also, if the background story were described in much less detail, would there be an effect?

I also wonder if this is an effect just of something like a priming effect, versus being explicitly about running a mental simulation. In particular, they didn’t test reaction times on target sentences that were irrelevant to the scenario. For example, if the sentence was “the river crosses the road” (when the story was about a road and didn’t mention a river), would people still take longer in the long-distance condition? If so, then that might point to something more like a priming effect that is engaged when the person reads a motion word like “crosses”, rather than them running a mental simulation (because there is no reason for a mental simulation about a river crossing a road to take longer, as it is irrelevant to the long-distance aspects of the story).