The Surprising Complexity of Scalar Implicature

This week’s papers both focused on language development in children, specifically regarding the concept of ‘scalar implicature.’  Scalar implicature refers to our ability to infer additional meaning from words like ‘none’, ‘some’, and ‘all’ that fall along a scale.  Adults interpret ‘weaker’ scaling words as negating ‘stronger’ ones – saying ‘some people have glasses’ implies that not all people have glasses.  Young children do not make this same inference, despite their skill at making linguistic inferences in a wide variety of other contexts.  Why do children develop scalar implicature later than other, similar skills?  Is the delay a result of lack of lexical knowledge, or does it mean a more fundamental logical mechanism is slow to develop? 

Stiller used three experiments to explore the source of delayed scalar implicature.  He used an implicature task with a non-standard scale to assess whether young children can use implicatures before they understand the subtext communicated by words like ‘some.’  The three- and four-year-old participants performed better on these simplified tasks, which relied the same kind of logical reasoning as scalar implicature.  However, their performance was still not comparable to that of adults.  Stiller further examined the importance of logical structure and context by adding additional features to the picture sequences used in the experiment, obscuring the ‘scale’ on which they were compared.  Adult participants no longer inferred scale from the image series, implying that scalar implicature depends on overall context rather than individual features. His third experiment confirms the power of context over implicature - we are more likely to construct scalar implicatures for rare features than for common features.  Stiller's results give us insight into the underlying logic of scalar implicature and the impact of feature rarity.  In the process, Stiller describes some interesting patterns that we unconsciously use to communicate more effectively.  For example, rarer features convey more useful information than common features, and people are more likely to notice and mention them.  

In his paper Accessing the unsaid: The role of scalar alternatives in children’s pragmatic inference, David Barner discusses several previously proposed sources of the relatively late development of scalar inference – it could be due to limitations on working memory, difficulty understanding and incorporating context, or limited ability to come up with scalar alternatives (like not thinking of ‘all’ as a logically stronger alternative to ‘some’).  Like Stiller, Barner’s results show the importance of context in scalar inference.  In his experiments, the presence of the word ‘only’ did not significantly modify children’s acceptance or denial of statements with context-independent scales (like ‘only some …’ vs ‘some …’), whereas it made a huge difference when questions with specific contextual alternatives were asked (like ‘the cat and dog …’ vs ‘only the cat and dog …’).  This aligns with Stiller’s results that the logical underpinnings of scalar inference are not causing the late development of this feature.  Neither are other factors like working memory to blame.  The main issue appears to be unfamiliarity with the implied context of extremely general scaling words like ‘some’ and ‘all.’

One thing I find amusing is that children are interpreting these statements more literally and logically than adults, who seem to be leaning on social convention (or general contexts and conventions) to convey a more specific meaning than they express.  This subtlety makes our communication more efficient, but seems to also come with drawbacks and a need for context that make learning slightly slower.