Eye-tracking the ham sandwich:
The on-line processing of contextually instantiated metonymies
Steven Frisson1, Petra Schumacher2, & Andrea Krott1
1
University of Birmingham, UK; 2 University of Cologne
The vast majority of words have different interpretations or senses which are semantically associated,
i.e. polysemous. For example, the word school can refer to a place where children are educated, the
school building, the period of education, the time when children are studying (e.g. before/after
school), a part of a university, the school as an institution, etc. Previous research using polysemous
words (Frazier & Rayner, 1990; Frisson & Pickering, 1999; McElree, Frisson, & Pickering, 2006; but
see Lowder & Gordon, 2013) has shown that (adult) language users do not experience an early
processing cost for conventionalized senses that are metonymically related, such as the concrete
building sense and the building-for-institution metonymic sense of school. Concretely, …talks to the
school… is (at least initially) read as fast as ....walks to the school… These results suggest that
understanding a derived, metonymic sense can be as straightforward as understanding a literal sense,
at least as long as the different senses are already known to the reader.
However, metonymic interpretations can also be construed “on the spot” when someone or
something is referred to using a typical or contextually evoked characteristic of that person or object.
The best-known example is what Nunberg’s (1979) calls “reference transfer”:
(1) The ham sandwich is sitting at table 20.
In (1), the term ham sandwich is used by the waiter to refer to the customer who ordered such a
sandwich rather than the sandwich itself. As there is nothing in the lexical-semantic meaning of “ham
sandwich” that directly refers to customer, this novel interpretation is ad hoc and contextually
determined (e.g. in a different context, the ham sandwich could refer to a chef who made an awardwinning sandwich).
While the ham sandwich cases have attracted a lot of attention in the pragmatic literature (e.g.
Jackendoff, 1997; Nunberg, 1995; Recanati, 1995, 2010; Ward, 2004), it remains unclear how easy or
hard it is for language users to arrive at a novel metonymic interpretation. Frisson and Pickering
(2007), using eye-tracking, found that, for a rule-governed metonymy such as producer-for-product,
processing is more difficult for novel metonymies (e.g. she read Needham) than for literal
expressions, but that a prior context which could serve as input to the rule (e.g. she has all the books
by Needham), could negate this difficulty. However, these types of metonymies are more constrained
than ad-hoc, contextually determined metonymies.
Schumacher (2011, 2014) are the only experimental studies of ham sandwich metonymies and
serve as inspiration for the present studies. Using event-related potentials (ERPs), she found that ham
sandwich metonymies that are licensed by context (e.g. a restaurant scenario) are more computational
demanding than their literal counter-parts (reflected in a Late Positive potential). Without contextual
support, additional processing costs were observed (N400 for unexpected target) but the Late
Positivity effect remained, suggesting that context does not represent a necessary cue for processing
this type of metonymy.
We carried out 2 eye-tracking studies looking at (1) whether novel ad hoc metonymies are
more difficult to process than literal expressions and (2) whether the difficulty is attenuated by a
context that supports the metonymic interpretation. Concretely, we compared the processing profiles
in neutral contexts (which used proper names) to those in supportive contexts (which used a
description of the protagonists rather than the name, see Table 1). 40 items were constructed based on
Schumacher (2011, 2014). The experiment was a between-subjects design, with the neutral context
conditions tested in Experiment 1 (N=22) and the supportive context conditions in Experiment 2
(N=22). We discuss the results together.
During initial processing of the target region (first-pass reading), the metonymic condition
was harder than the literal condition, but only for the neutral context conditions. Following supportive
context, the metonymic interpretation did not differ reliably from the literal expression. Intermediate
reading measures (regression-path and first-pass regressions) showed the same pattern for the target
region. For the spill-over region, the metonymic expression took more processing effort for both types
of contexts. A late reading measure (total gaze) revealed increased processing on the metonymic word
for both context types.
These results indicate that (1) novel ad hoc metonymies are more difficult to process than
literal expressions and (2) that having a supportive context can make initial processing of this type of
metonymy easier. This fits with Frisson and Pickering’s (2007) findings for rule-governed
metonymies, though their results showed hardly any processing difficulty for novel metonymies in a
supportive context, which might indicate that extending a word’s meaning by applying a metonymic
rule is easier than understanding contextually instantiated and determined ad hoc metonymies.
The results are also nicely in line with Schumacher (2011, 2014) who distinguished between
contextual licensing (N400 effects) and later pragmatic meaning adjustment (Late Positivity effect).
The current study allows us to reconcile previous data from eye-tracking and ERP studies. It further
indicates that there are different types of metonymy (rule-governed vs. ad hoc) which are subject to
discrete processing demands.
(Note: we are currently running the experiments using ERP in order to determine whether the English
stimuli behave identical to the German ones from Schumacher (2011, 2014). We intend to include a
discussion of these findings in our presentation.)
Table 1. Sample stimuli
Condition
Example
Emma asks Jim what it is that's new on the menu.
Neutral context –
Jim replies that the espresso is new on the menu.
Literal
Claire asks Liam who it was that left without paying.
Neutral context –
Liam replies that the espresso left without paying.
Metonymic
Supportive context – The waitress asks her boss what it is that's new on the menu.
Her boss replies that the espresso is new on the menu.
Literal
Supportive context - The attendant asks her supervisor who it was that left without paying.
Her supervisor replies that the espresso left without paying.
Literal
Notes: target region in bold, spill-over region underlined.
References
Frazier, L., & Rayner, K. (1990). Taking on semantic commitments: Processing multiple meanings vs.
multiple senses. Journal of Memory and Language, 29, 181-200.
Frisson, S. & Pickering, M. J. (1999). The processing of metonymy: Evidence from eye movements.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 25, 1366-1383.
Frisson, S., & Pickering, M. J. (2007). The processing of familiar and novel senses of a word: Why
reading Dickens is easy but reading Needham can be hard. Language and Cognitive Processes,
22, 595-613.
Jackendoff, R. S. (1997). The architecture of the language faculty. Cambridge, MA: MIT Press.
Lowder, M.W., & Gordon, P.C. (2013). It’s hard to offend the college: Effects of sentence structure
on figurative language processing. Journal of Experimental Psychology: Learning, Memory and
Cognition, 39, 993-1101.
McElree, B., Frisson, S., & Pickering, M. J. (2006). Deferred interpretations: Why starting Dickens is
taxing but reading Dickens isn’t. Cognitive Science, 30, 115-124.
Nunberg, G. (1979). The non-uniqueness of semantic solutions: Polysemy. Linguistics and
Philosophy, 3, 143-184.
Nunberg, G. (1995). Transfers of meaning. Journal of Semantics, 12, 109–133.
Recanati, F. (1995). The alleged priority of literal interpretation. Cognitive Science, 19, 207-232.
Recanati, F. (2010). Truth-conditional pragmatics. Oxford: Oxford University Press.
Schumacher, P. B. (2011). The hepatitis called…: Electrophysiological evidence for enriched
composition. In J. Meibauer & P. B. Schumacher (Eds.), What is a context? Linguistic
approaches and challenges (pp. 33-53). Amstredam: John Benjamins.
Schumacher, P. B. (2014). Content and context in incremental processing: “the ham sandwich”
revisited. Philosophical Studies 168. 151-165.
Ward, G. (2004). Equatives and deferred reference. Language, 80(2), 262–289.