The End of Language Embodiment

Motor Cortex and Monkeys are Responsive to Statistical Regularities of Letter Strings

Motor activity for nonwords

A cool new study in the Journal of Cognitive Neuroscience questions the notion that the premotor cortex response to action words is due to implicit motor simulation (de Zubicaray et al., 2013). Previously, the conceptual representation and/or simulation of action words in motor regions of the brain has been taken as evidence for embodied theories of language comprehension (Glenberg & Kaschak, 2002). These theories have been based on fMRI and EEG experiments showing that reading or listening to verbs that depict actions of the face, arm or leg activate somatotopically-specific regions of motor cortex (Hauk et al., 2004).

A review of the fMRI literature on the perception or production of nonwords (e.g., sonerge, rintoon) revealed peak activations in motor regions (shown above in Fig. 2 of de Zubicaray et al., 2013). Nonwords are devoid of semantic meaning and hence cannot elicit motor simulations. However, endings of words (e.g., –erge, –oon) can be highly predictive of grammatical category (verb and noun, respectively, for the examples above).

de Zubicaray and colleagues took advantage of these probabilistic sublexical cues when constructing stimuli for their own fMRI experiment. A set of nouns, verbs, noun-like nonwords, and verb-like nonwords were presented during a grammatical category judgment task (“identify whether a letter string on the screen seemed more likely to be a noun or a verb”). The noun stimuli were not related to body parts (e.g., cavern) and the verbs denoted manual actions (e.g., crumple). In brief, the results suggested that verbs and verb-like nonwords activated premotor cortex to a greater extent than nouns and noun-like nonwords.

The authors took this as evidence that the observed cortical responses to action verbs were due to ortho-phonological probabilistic cues to grammatical class and not to embodied motor representations. In other words, the spelling and pronunciation of word endings influenced activity in motor regions of the brain. Furthermore,  the verb-like activations overlapped with the motor hand area, as defined by a  localizer scan where participants viewed hand actions.

The results have challenging implications not only for theories of embodied language comprehension, but also for mirror neuron theories of action understanding. In monkeys, neurons in ventral premotor cortex (area F5) fire when they perform an action, and when they observe a similar action. fMRI studies in humans lack the spatial and temporal resolution of single unit recordings, but “mirror-like” activations in a parieto-frontal circuit have led to the hypothesis that the mirror system is responsible for comprehending the actions and intentions of others (Rizzolatti & Sinigaglia, 2010).1

If premotor cortex is sensitive to statistical regularities in nonwords, this presents a problem for grounding language in action areas of the brain.2

Speaking of statistical regularities in letter strings, an intriguing study in baboons suggested that they might be able to distinguish written English words from nonwords using an orthographic code and not just visual pattern matching (Ziegler et al., 2013). New evidence for this interpretation was provided by the transposed letter effect, where humans are able to read scrambled versions of words if the middle letters are reversed in a specific fashion (e.g., casinocaniso).3 This ability would trip you up when making lexical decisions, because caniso would be erroneously identified as a word.

Six trained ‘reading’ baboons actually showed the transposed letter effect, making more false positive errors for nonwords like caniso than for other types of nonwords. In the Discussion, Ziegler et al., (2013) put forth the bold interpretation that the monkeys were able to engage orthographic processes… which are not actually linguistic in nature:

In sum, word/nonword discrimination performance of monkeys is sensitive to a marker effect of orthographic processing—the transposed-letter effect—but not to the effects of visual similarity. This finding clearly suggests that monkeys use a truly orthographic code rather than a visual code.

. . .

Reading and writing are recent cultural inventions in humans. Although baboons do not have human-like language, they are sensitive to a classic marker of orthographic processing. These findings suggest that the front end of reading (Grainger & Dufau, 2012) is supported by neural mechanisms that are much older than the behavior itself and are not linguistic in nature (Platt & Adams, 2012).

Obviously, the monkeys are unable to translate orthography into phonology (i.e., read the words aloud) or perform grammatical judgments, so they wouldn’t show premotor activation for verb-like nonwords…


ADDENDUM (7/3/2013): I meant to link to Andrew D. Wilson and Sabrina Golonka’s Frontiers article, Embodied cognition is not what you think it is. They also have a nice new post on Grounded vs. embodied cognition: a (hopefully uncontentious) note on terminology.


1 Dr. Greg Hickok has been especially critical of this view.

2 One potential objection to the ‘end of embodiment’ argument is that verb-like nonwords primed semantic representations of actual verbs that are orthographically and phonologically similar.

3 But see If You Can Raed Tihs, You Msut Be Raelly Smrat for the limitations and boundary conditions of this effect.


de Zubicaray G, Arciuli J, McMahon K. (2013). Putting an “End” to the Motor Cortex Representations of Action Words. J Cogn Neurosci. Jun 27. [Epub ahead of print]

Glenberg AM, Kaschak MP. (2002). Grounding language in action. Psychonomic Bulletin & Review 9:558-565.

Hauk O, Johnsrude I, Pulvermüller F. (2004). Somatotopic representation of action words in human motor and premotor cortex. Neuron 41:301-7.

Rizzolatti G., Sinigaglia C. (2010). The functional role of the parieto-frontal mirror circuit: interpretations and misinterpretations. Nature Reviews Neuroscience 11:264-274.

Ziegler JC, Hannagan T, Dufau S, Montant M, Fagot J, Grainger J. (2013). Transposed-Letter Effects Reveal Orthographic Processing in Baboons. Psychol Sci. Jun 11. [Epub ahead of print]

Published on: July 3, 2013 @ 03:57

10 thoughts on “The End of Language Embodiment

  1. Playing Devil’s Advocate: the “verby” nonwords activate the motor cortex because the brain is trying to work out which verb it is (and failing, but still activating). For the same reason, I bet if you showed non-Greek speakers an assortment of Greek letters, it would activate their language areas more than just a set of lines. Even though no actual language content is being transmitted (they can’t read it), there will be “word-like” parts to it (some of them look like English letters). But it wouldn’t prove that language centers don’t actually do language.

    • I think Neuroskeptic as Devil’s Advocate hit the nail on the head here. In the task, dozens of real verbs — ALL hand actions — were presented to subjects. It’s not surprising that their brains, upon seeing a new verb-like nonword, would try to interpret it as a hand action. It isn’t necessary to posit activation via orthographic or phonological neighbours.

      • Are you positing that all the verb-like nonwords in this study activated the meaning of hand action verb(s) directly, bypassing any orthographic resemblance to non-hand verbs?

      • The only kind of verb subjects saw were hand verbs, which appeared on 1/4 of the trials. The subjects were primed to expect hand verbs to appear with high frequency. And when something verby flashed on the screen, it’s possible their brains tried hand verb interpretations first.

        The priming effect could have been strong enough to overwhelm any interpretation via orthographic similarity to non-hand verbs, which in this study had a frequency of *zero* after all, and therefore would not have been expected.

      • I’m a little curious about your model of attempted meaning activation, because the only way to know whether a nonword is verby or not is through statistical regularities at the ends of the stimuli, -erge vs. -oon. This still makes premotor cortex sensitive to orthography and phonology, which is problematic for grounded theories of language.

        Seems you’re getting at Greig de Zubicaray’s comment below? “…a roundabout appeal to context dependent meaning activation. One way of testing that task context hypothesis would be to run the nonwords first followed by the words in separate blocks.”

        So you’d like to see an experiment with nonwords run in separate blocks.

      • If manual word meanings (which ones in particular?) were activated contingent upon verb-like nonwords being identified as verbs via their orthographic cues, then the semantic activation would be epiphenomenal. You could try to attribute the activation to a post-lexical meaning integration/verification mechanism based on an expectancy produced by the ¼ trials in which manual verbs appeared, although it’s difficult to explain how this additional processing cost would be useful/strategic for performing the task. This latter explanation would be an example of context dependent meaning activation, as I wrote in my last post. We’ve invoked expectancy-induced priming mechanisms to explain results from go/no-go paradigms that have been interpreted as support for embodied language studies in the past, see:

        Otherwise, the proposal would be that semantic activation occurred prior to orthographic processing, as neurocritic points out, and I don’t think you’ll find much support for that in word recognition models.

  2. Yes, that’s related to the argument that verb-like nonwords can prime semantic representations of actual verbs that are orthographically and phonologically similar. You see the nonword sonerge, and and it activates the real verb converge. Making lexical or grammatical decisions for nonwords generally takes longer than if the stimulus was a real word.

  3. Hi. Thanks for the neurocomplimentary coverage of our paper. Re the argument that the overlapping result was due to the verb-like nonwords priming semantic representations of actual verbs that are orthographically and phonologically similar: this assumes that the verb-like nonwords were only activating orthographically and phonologically similar manual verbs and their meanings, which seems a little strong as the cues in the nonword endings were to grammatical class more generally, and we constructed word and nonword lists so the items had minimal overlap in their endings (< 10%). We'd have to determine how many disyllabic manual verbs in the average lexicon actually have those nonword endings to support the priming argument – I suspect this would be quite a small proportion. To circumvent this problem, you could appeal to the task context in which the nonwords were presented, by proposing that participants interpreted the task as only discriminating manual verbs vs non-manual nouns, yet that would be a roundabout appeal to context dependent meaning activation. One way of testing that task context hypothesis would be to run the nonwords first followed by the words in separate blocks.

    • Ah, of course, the potential orthographic neighbors would all have to be manual verbs. The example of sonerge/converge doesn’t even conform.

      Thanks for your comment. It was a fascinating (and somewhat surprising) finding, even though the collective nonword data have been out there for while.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s