Filled pauses signal structure—even in writing!

In a few posts over a year ago, I explained how I did a replication and extension of Bailey and Ferreira's 2003 study looking at the influence of filled pauses on listener's interpretation of sentence structure. As a quick reminder, they observed that when a filled pause occurred at a non-clause boundary, listeners were less likely to judge the sentence grammatical than when the filled pause occurred at a clause boundary. I extended this work by comparing the influence of silent and filled pauses, and testing these with both native English and non-native English listeners (posts here, here, and here). Results replicated Bailey and Ferreira's with non-boundary pauses leading to lower judgments of grammaticality. Furthermore, silent pauses showed a larger effect than filled pauses.

That brings us to the present post which is actually a short report on a trip I made to the US in March. For the first part of my trip, I visited Michigan where I visited Debra Hardison at Michigan State University where I talked with her about my work on Fluidity. It was a nice visit and a great chat. I was pleased to get acquainted with her as she is well-known for her work in pronunciation which is exactly one direction where I think Fluidity needs work. I hope I will have a chance to consult with her again, or perhaps hear her speak at a conference, perhaps here in Japan.

But more related to the structural signalling work, I also had an opportunity visit Karl Bailey (yes, the first half of "Bailey and Ferreira") at Andrews University in western Michigan to chat with him about the work I've been doing. I really wanted to get his take on the results I've get so far, as well as ask a few questions about how he did the study originally.

We talked for quite a while and covered a lot of ground. He was quite pleased yet also not so surprised at the silent pause findings. And I guess in one sense it is a bit obvious: the fact that both silent and filled pauses show the same structural signalling effect is very reasonable since the extra time that each affords leads to certain inferences which in one condition (the non-boundary case) turn out to be wrong, while in the other condition (the boundary condition) turn out to be right. But it is still surprising (to me, at least) that silent pauses show a larger effect. One might suppose that filled pauses actually mislead more because the listener has to process the filled pause and recognize it first as a filled pause before realizing that it has no semantic value to the ongoing utterance, then subsequently draw various inferences. A silent pause does not incur any of that cost, so the extra time could be used to process the situation carefully: In the non-boundary case, they may be better prepared to recognize the ambiguity. But, they do not: It is the worse case.

I also wanted to confirm two other points with him. First, because their study was actually based on earlier work by Ferreira on the head noun effect, they constructed their stimuli with the "boundary" filled pauses actually one word inside the boundary: the uh uh deer ran into the woods. But they cast their argumentation in terms of it being at the boundary. For my work, I put the filled pauses at the boundary. He agreed that under their theory and analysis, that would yield exactly the same predictions about the signalling effect.

Another point I wanted to ask about was how they recorded the filled pauses. Because the filled pause in the boundary condition comes after an indefinite article (in many of their stimuli), it would most naturally be pronounced as thee uh uh (with a high front vowel rather than schwa). Indeed, he confirmed that that was the case with their stimuli. Now, that does mean that the boundary case actually had a highly marked disfluency compared to the non-boundary case. I don't think this is a problem for their study, because the participants' task was to judge grammaticality after the sentence was finished. But such a design could cause problems in other experimental paradigms, like self-paced reading.

Which brings me to my second main point for this post. After Michigan, I flew to Colorado to attend the CUNY Human Sentence Processing Conference. It was a great even where I was pleased to run into one of my former advisers from Northwestern, Katy Carlson, who is now at Morehead State University in Kentucky.

My talk is a continuation of the structural signalling effect work, but this time looking at filled pauses (only) in written texts. So, this was a pretty simple experiment I did to see if structural signalling would show up while readers were reading sentences with uh in them. This was a self-paced reading study, so that I could actually look at the effect of the filled pause in context. Sure enough, results were consistent with the listening tests: Readers slow down after the filled pause in the non-boundary condition, but maintain the same pace in the boundary condition. Furthermore, they also judge the sentences grammatical less often in the non-boundary condition.

So, it really seems that this structural signalling effect is real and robust, showing converging results with different pauses, different native language listeners, and different communicative modes.

[Note: This post was written in September, 2020. However, in order to preserve the chronology of the blog, it has been dated to reflect when the described events actually took place.]