Suppose you and your favourite MoL-student are procrastinating on your homework assignment (very unrealistic, I know). You decide to play a game – not a very hard one, because your assignment is already difficult enough: one of your laptop screens shows a display containing multiple objects, each of a different shape and colour.
Your task is to think of one of the objects and have your teammate click on it. You are allowed to talk, but you can’t point to the shape or give any other kind of non-verbal hint. Suppose you choose the star in the upper left corner. What do you say to your teammate? Think about it for a minute, before you continue reading.
(Rubio-Fernández 2019, 5)
You probably came up with a number of possibilities for referring to the star. For instance, you could have said ’the star’, ’the upper left shape’, ’the blue shape’, ’the shape left of the rectangle’, and so on. Clearly, though, these different descriptions are not all equally suitable for the task at hand. You probably went for a very short one, ensuring your teammate selects the right object without overcomplicating things.
Perhaps, you chose ‘star’. If so, then Grice would be very happy for you. According to Grice (1975), rational and cooperative discourse participants normally choose their utterances in a way so as to be truthful (Maxim of Quality) and to be sufficiently informative while remaining as succinct as possible (Maxim of Quantity). Your description ‘star’ is as short as possible and refers uniquely to the object you had in mind. So, congratulations: according to Grice, you have just shown evidence of being rational and cooperative!
Hopefully then, you did not decide to say something like ‘blue star’. Although such an utterance satisfies the Maxim of Quality, it clearly violates the Maxim of Quantity. Why say ‘blue star’ if there is only one star? You could have saved yourself the trouble by just not mentioning the colour at all, in which case your teammate would still have understood which object you had in mind. The Gricean model assumes that if one is cooperative, one adheres to the maxims. You didn’t – ergo, you were being uncooperative.
That’s probably not very nice to hear. Luckily for you, however, you’re far from alone. In a series of experiments, Rubio-Fernández (2016); Rubio-Fernández (2019) let participants play precisely the game described above. Interestingly, rather than choosing to say ‘star’, most participants decided to go for ‘blue star’. Like you may have done, they violated the Maxim of Quantity. This would render them uncooperative in the Gricean model – but what does that mean?
If someone is uncooperative, we’d expect them to be so because they are lazy or unmotivated and won’t really bother to make clear what exact object they had in mind. However, here we call people uncooperative because they do the exact opposite: they give too much information. Why would we ever be uncooperative in this very unintuitive way? Clearly, there has to be a reason for our redundant behaviour – one that upholds our rationality as discourse participants.
To understand why we decide to be redundant, we should first ask ourselves in what kind of contexts we actually make this decision. In the same series of experiments, Rubio-Fernández (2019) found that increasing the number of objects on display caused the participants to mention the colour next to the shape of the target object more often. So, they were increasingly more likely to say ‘blue star’ (instead of just ‘star’) in displays with two, four and eight differently coloured shapes:
(Rubio-Fernández 2019, 5)
She additionally found that changing the colour of the other shapes to that of the target object decreases the chance someone still mentions the object’s colour. Thus, people are much less likely to say ‘blue star’ in the left display than in the right:
(Rubio-Fernández 2019, 5)
Degen et al. (2020) moreover discovered that it also matters how typical an object’s feature is for it to be mentioned redundantly. For instance, you probably wouldn’t mention a banana’s colour when that banana is yellow, whereas maybe you would when it’s, say, purple (see also Westerbeek, Koolen, and Maes 2015). Furthermore, not all kinds of features are equally likely to be mentioned redundantly. For instance, we much more often talk about an object’s colour than about the material it’s made of, or about its size.
Thus, in choosing to be redundant, we seem to consider several factors. We can group these factors into two broad categories. Factors having to do with the visual salience of the object, such as the observed effect of adding more objects to the display or changing their colours, can be termed bottom-up: they only arise given empirical data in a specific context. In contrast, we can say that other factors such as typicality and the kind of the object’s feature are top-down, since they relate to our world knowledge and previous experiences about objects and their features (Mitchell, Reiter, and Van Deemter 2013).
With our distinction between bottom-up and top-down factors in hand, let’s return to the question we raised before: why do we choose to be redundant? There are two popular answers to this question. The first could be called the Continuous Semantics Account or CSA for short (Degen et al. 2020). The CSA departs from a continuous view on semantics: all utterances are ’noisy’ to a certain degree. An utterance is said to be noisy when it’s not immediately clear what is meant by it, e.g. because it has more than one meaning or because it’s an unusual way of referring to that object. CSA sees being redundant as a way of potentially reducing this noise. Thus, giving more information than strictly necessary can help take away some of the confusion on what you intended to convey.
Another explanation is given by the Visual Efficiency Hypothesis or VEH (Rubio-Fernández 2020). The VEH says that being redundant helps the listener find the target object more quickly and efficiently. Being redundant doesn’t provide the listener with more information (contra the CSA), but it can still be very helpful for the listener’s visual search. Multiple empirical studies support this idea, showing that listeners respond faster to redundant utterances compared to minimally informative ones
\[Arts et al. (2011); paraboni_generating_2007; Rubio-Fernández (2020)\].
I don’t want to express a preference for either the CSA or VEH here. In fact, I believe we need both to fully understand our reasons for being redundant. It is no coincidence that the researchers interested in the top-down factors are proponents of the CSA, and those investigating bottom-up factors of the VEH. In a conversation I had with Robert Hawkins (one of the co-authors in Degen et al. (2020)), he explained to me how their notion of ‘semantic noise’ can perhaps best be seen as caused by a clash between the context and our world knowledge or expectations. Bananas are normally not purple, so just saying ‘banana’ to refer to a purple banana could lead to confusion. At the same time, the VEH can accurately account for the bottom-up factors we discussed: if more objects are added to a display, more time can be gained by redundantly mentioning an object’s colour and with it quickly eliminating all objects not matching that colour.
To close off, let’s zoom out and talk about more than shapes and colours. Redundant expressions cannot only be found in the context of reference games or purple bananas. Language is full of redundancy: we constantly tell each other stories that could have been told a lot quicker and give descriptions that could have been a lot more concise. What we have just learned, I think, is that the reason for this may be very allocentric: being redundant helps the people listening to us understand us more easily and quickly. So, the next time someone accuses you of TMI’ing, just know that you are a very rational, cooperative, and maybe even helpful human being.
Bibliography
Arts, Anja, Alfons Maes, Leo Noordman, and Carel Jansen. 2011. “Overspecification Facilitates Object Identification.” Journal of Pragmatics 43 (1): 361–74. https://doi.org/10.1016/j.pragma.2010.07.013 .
Degen, Judith, Robert D. Hawkins, Caroline Graf, Elisa Kreiss, and Noah D. Goodman. 2020. “When Redundancy Is Useful: A Bayesian Approach to ‘Overinformative’ Referring Expressions.” Psychological Review 127 (4): 591–621. https://doi.org/10.1037/rev0000186 .
Grice, H. P. 1975. “Logic and Conversation.” In Speech Acts, edited by Peter Cole and Jerry L. Morgan, 41–58. Brill.
Mitchell, Margaret, Ehud Reiter, and Kees Van Deemter. 2013. “Typicality and Object Reference.” Proceedings of the Annual Meeting of the Cognitive Science Society 35 (35): 3062–67.
Rubio-Fernández, Paula. 2016. “How Redundant Are Redundant Color Adjectives? An Efficiency-Based Analysis of Color Overspecification.” Frontiers in Psychology 7 (153). https://doi.org/10.3389/fpsyg.2016.00153 .
———. 2019. “Overinformative Speakers Are Cooperative: Revisiting the Gricean Maxim of Quantity.” Cognitive Science 43 (11): e12797. https://doi.org/10.1111/cogs.12797 .
———. 2020. “Redundant Color Words Are More Efficient Than Shorter Descriptions.” OSF. https://doi.org/10.31234/osf.io/gbpt3 .
Westerbeek, Hans, Ruud Koolen, and Alfons Maes. 2015. “Stored Object Knowledge and the Production of Referring Expressions: The Case of Color Typicality.” Frontiers in Psychology 6 (935). https://doi.org/10.3389/fpsyg.2015.00935 .