The prevalence of ambiguity in language poses a problem for communication. Ambiguous expressions require listeners to infer which interpretation was intended, raising the probability of miscommunication. One fairly obvious solution to this problem would be for speakers to speak less ambiguously––but is this really what they do?
First, we need to define exactly what we mean by “speak less ambiguously”. For the purposes of this blog post, “speak less ambiguously” means avoid ambiguity. Thus, the question becomes: do speakers actively avoid using an expression with more than one interpretation, particularly when both interpretations are licensed by the context?
This question actually touches on more than just the issue of speaker conscientiousness. By examining which kinds of ambiguity speakers avoid, and when, we can better understand which parts of the language production architecture are more automatic, and which parts are more subject to intentional intervention. Additionally, understanding whether speakers actively avoid ambiguity, or whether the burden of interpretation is left to the listener, grants deeper insight into whether language evolution might be shaped more by speaker-centered or listener-centered cognitive pressures.
The case of the optional “That”
One well-known case of ambiguity is the so-called garden-path sentence, in which the beginning of a sentence leads the listener (or reader) towards an incorrect, initial parse of the sentence’s structure. For example, when hearing the sentence “He believed the coach was lying”, a listener might first think that the coach (a Noun Phrase) is serving as the Direct Object of the verb believe. Continued listening (or reading) reveals that the coach is actually the Subject of a new sentential complement. This is known to incur processing costs on listeners, usually indexed by increased reading time (Grodner et al, 2003) or particular brain electrical potentials thought to reflect difficulties in syntactic parsing (Osterhout et al, 1994).
These garden-path effects arise because of a temporary ambiguity. However, this ambiguity could be avoided altogether by including the optional complementizer “that”, e.g. “He believed that the coach was lying”. This modified sentence only licenses one interpretation of the coach: as the Subject of a sentential complement, not as a Direct Object. Thus, speakers could make things easier for listeners by including the “that”. Is this what they do?
Ferreira & Dell (2000) tested exactly this question. Specifically, they asked whether speakers were more likely to include the optional “that” when the sentence would otherwise be ambiguous. In Experiment 1, 96 participants were asked to remember and reproduce sentences of the form “The coach knew (that) X missed practice”. The original sentence presented to the participants either included the optional “that” or did not; potential ambiguity was manipulated by substituting in for X either a pronoun like “I” or “you”, e.g.:
- The coach knew (that) I missed practice. (unambiguous)
- The coach knew (that) you missed practice. (ambiguous)
Thus, there were two potential factors that could influence whether a participant would include the optional “that”. First, the inclusion of “that” in the original sentence could influence whether they chose to include it in their reproduction of that sentence. Second, the potential ambiguity of omitting “that” could influence whether they chose to include it.
Perhaps unsurprisingly, participants were much more likely to include the optional ‘that’ when it was included in the original sentence. But more importantly, their decision was unaffected by the potential ambiguity of the sentence (74.3% inclusion in unambiguous vs. 76.6% inclusion in ambiguous cases). In other words, speakers didn’t avoid producing particular sentence forms, even though those forms would be temporarily ambiguous. The decision to include the optional “that” was not predicted by the ambiguity of the sentence, but simply by whether the original sentence contained “that”.
Ferreira & Dell (2000) then report the results of another five experiments designed to further probe this effect and test for potential confounds. It turns out that the effect generalizes quite well––for example, even more extreme cases of syntactic ambiguity (e.g. The horse raced past the barn fell) fail to trigger substantial modulations in speech production (Experiment 3), supporting the hypothesis that speakers don’t avoid ambiguity. Speakers’ production decisions are much better predicted by the availability of particular lexical items, rather than the extent to which they’d disambiguate the meaning of an utterance (Experiment 4); that is, they change what they say on the basis of what is easy to produce, rather than what is ambiguous.
Together, these results, and many others (see Wasow, 2015) suggest that speakers don’t avoid ambiguity. Speakers produce utterances that are easy to produce, regardless of whether they are ambiguous. This raises the question: why don’t speakers avoid ambiguity?
Why don’t speakers avoid ambiguity?
If ambiguity presents such a problem for listeners, why don’t speakers avoid it?
Ferreira (2008) explores this question in detail, and arrives at a possible solution: avoiding ambiguity is hard, and the production system is organized around reducing the many difficulties involved in articulating what you want to say. Ferreira argues that avoiding ambiguity is tantamount to a “U-turn”:
This process describes a kind of “production U-turn”: To detect ambiguity, speakers must take a meaning, formulate a candidate expression, and compare it back against meaning (other meanings) to determine whether it is ambiguous. Like a U-turn in real life, executing such a U-turn would seem inefficient and difficult.
In other words, avoiding ambiguity requires at least two steps:
- Recognizing that an expression is potentially ambiguous;
- Reconstructing an expression to avoid that ambiguity.
Performing (1) likely requires “simulating” how a listener would interpret a sentence. That is, having selected the words The horse raced past the barn fell, a speaker would have to pass that message through their own comprehension system; if the resulting interpretation is ambiguous, the speaker would then reformulate the message, iterating until their “simulation” arrives at an unambiguous interpretation (Pickering & Garrod, 2013; Ferreira, 2018). Although this kind of audience design may be implemented in certain cases, such as recalibrating an utterance so as to avoid hurting someone’s feelings (Ferreira, 2018), the evidence (Ferreira, 2008; Wasow, 2015) suggests that it’s rarely deployed to avoid syntactic ambiguity.
We started with the question: do speakers avoid ambiguity? At least in the case of syntactic, ambiguity, it appears that they don’t (Ferreira, 2008). One possible explanation for this is that avoiding syntactic ambiguity is hard, and speech production is already hard enough. Speakers are primarily concerned with getting their message across the easiest way they can.
How, then, is the problem of ambiguity addressed? Ferreira (2008) and others (see Levinson (2000) and Piantadosi et al (2012)) suggest that the onus of interpretation falls primarily on listeners. That is, in the so-called “division of labor” across speakers and listeners, speakers are responsible for efficiently producing utterances “even if they’re not optimally understandable”, while “addressees do what they need to to understand their speakers” (Ferreira, 2008).
Critically, this might also have implications for how languages evolve. Wasow (2013) theorizes that language evolution is shaped by constraints around speech production, consistent with Levinson’s (2000) proposal that languages evolve to minimize the effort involved in speaker articulation.
But as always, open questions remain. Notably, there are other kinds of ambiguity––lexical ambiguity (e.g. homophones), pragmatic ambiguity (e.g. sarcasm, etc.), and more. Do speakers avoid these ambiguities, or, as with syntactic ambiguity, do they produce them all the same? Alternatively, do speakers supplement their ambiguous expressions with disambiguating cues, such as gestures, eye gaze, or prosody (e.g. the tone/rhythm of speech)?
Ultimately, understanding when, and why, speakers avoid (or fail to avoid) ambiguity, and how listeners interpret ambiguous utterances, gives us deeper insight into how conversation manages to occur at all, despite such a noisy communication channel.
Ferreira, V. S., & Dell, G. S. (2000). Effect of ambiguity and lexical availability on syntactic and lexical production. Cognitive psychology, 40(4), 296-340.
Ferreira, V. S. (2008). Ambiguity, accessibility, and a division of labor for communicative success. Psychology of Learning and motivation, 49, 209-246.
Ferreira, V. S. (2018). A Mechanistic Framework for Explaining Audience Design in Language Production. Annual review of psychology, (0).
Grodner, D., Gibson, E., Argaman, V., & Babyonyshev, M. (2003). Against repair-based reanalysis in sentence comprehension. Journal of Psycholinguistic Research, 32(2), 141-166.
Levinson, S. (2000). Presumptive meanings: The theory of generalized conversational implicature. MIT Press.
Osterhout, L., Holcomb, P. J., & Swinney, D. A. (1994). Brain potentials elicited by garden-path sentences: evidence of the application of verb information during parsing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(4), 786.
Piantadosi, S. T., Tily, H., & Gibson, E. (2012). The communicative function of ambiguity in language. Cognition, 122(3), 280–291. https://doi.org/10.1016/j.cognition.2011.10.004
Pickering, M. J., & Garrod, S. (2013). An integrated theory of language production and comprehension. Behavioral and Brain Sciences, 36(4), 329-347.
Wasow, T. (2013). The appeal of the PDC program. Frontiers in psychology, 4, 236.
Wasow, T. (2015). Ambiguity Avoidance is Overrated. Ambiguity: Language and communication, 29.
 Note that the minimization of ambiguity could also mean that speakers provide additional cues with which to disambiguate their intentions (e.g. prosodic marking, gesture, eye gaze, etc.), strategically or automatically––rather than avoiding the ambiguous expression altogether. In other words, it could be that speakers don’t avoid ambiguous expressions, but that they usually produce these expressions in a disambiguating context, or with supplemental cues with which a listener can disambiguate their intent.