Losing the sense of smell: Contrary to predictions, anosmics were better at remembering odor words, and rated odor and taste words more positively than control participants

Losing the sense of smell does not disrupt processing of odor words. Laura J. Speed et al. Brain and Language, Volume 235, December 2022, 105200. https://doi.org/10.1016/j.bandl.2022.105200

Highlights

• Anosmics and controls do not differ in their comprehension of odor words.

• Anosmics remembered more odor words than controls.

• Anosmics rated odor words as more positively valenced than controls.

Abstract: Whether language is grounded in action and perception has been a key question in cognitive science, yet little attention has been given to the sense of smell. We directly test whether smell is necessary for comprehension of odor language, by comparing language processing in a group of participants with no sense of smell (anosmics) to a group of control participants. We found no evidence for a difference in online comprehension of odor and taste language between anosmics and controls using a lexical decision task and a semantic similarity judgment task, suggesting olfaction is not critical to the comprehension of odor language. Contrary to predictions, anosmics were better at remembering odor words, and rated odor and taste words as more positively valenced than control participants. This study finds no detriment to odor language after losing the sense of smell, supporting the proposal that odor language is not grounded in odor perception.

Keywords: Mental simulationEmbodimentAnosmiaOlfaction

4. Discussion

We investigated the online processing of olfactory language in a large group of individuals with acquired anosmia. Following proposals that odor language is grounded in olfactory representations (González et al., 2006), we predicted participants with anosmia would be impaired in processing odor language compared to controls. However, we found no difference in lexical decision performance for odor and taste words in anosmics compared to controls, and no detriment in semantic similarity judgments for odor and taste words in anosmics compared to controls. We find no evidence that odor representations play a critical role in the comprehension of odor language. Although odor representations may be activated during explicit imagination or expectation tasks (Zhou et al., 2019), our results do not support the proposal that odor simulation is involved in odor language comprehension (Speed and Majid, 2018, Speed and Majid, 2019).

We also predicted anosmics would remember fewer odor and taste words than control participants. On the contrary, we found anosmics remembered more odor words. This suggests losing the sense of smell is not detrimental to odor language processing. The puzzling aspect here is that we find the opposite pattern to the one predicted. It is possible that odor words are more salient to anosmics because they are aware these words are related to the perceptual sense they have lost. In particular, we recruited participants via two anosmia charities, whose members likely attach more importance to the sense of smell and their loss. Reading odor words may have left an emotional trace for these participants. fMRI studies with patients with acquired anosmia have shown extensive activation in higher-order olfactory regions of the brain in anticipation of odor words, suggesting increased effort or attention (Han et al., 2019, Joshi et al., 2020). Such an explanation is also compatible with the better memory for odor words we found: anosmics may have effortfully processed odor words, simultaneously strengthening memory traces. On the other hand, the lack of difference in lexical decision response time to odor words between anosmics and controls suggests an explanation in terms of processing effort is unlikely.

There are other possibilities to consider. Perhaps the anosmic participants were more motivated to perform the study than the control participants who we recruited via Prolific Academic. This could be tested in the future by replicating the study in a lab environment where motivation may be better equated. Whilst we agree there are limitations with collecting control data from participants online, there are also some benefits. It has been shown, for example, that participants recruited online are more diverse than typical university samples (Burhmester, Kwang, & Gosling, 2011), and therefore may be a better comparison for the group of anosmic participants. Furthermore, research has shown that a number of classic psycholinguistic effects involving small differences in reaction time have been replicated with data collected online, and there are negligible differences in the quality of data between online studies and studies in the lab (Enochson, Culbertson, & Eriksson, 2015, Germine, Nakayama, Chabris, Chatterjee, & Wilmer, 2012). We are therefore confident in the quality of the control data we collected. Another factor to consider is that we followed our preregistered hypotheses and analyses, and did not correct for multiple comparisons when following up significant interaction and main effects (see Rothman, 1990). If we did apply a more conservative significance criterion, the difference in word recall and valence ratings between anosmics and controls for odor words would no longer be significant, whilst the difference in valence between anosmics and controls for taste words would remain. Future research should therefore aim to replicate such an effect.

It has been suggested that if odor simulation is unavailable during language comprehension, emotional simulation may become relevant for odor language (Speed & Majid, 2019). Our emotional ratings provide first support for this idea. Odor and taste words had more positive valence associations for anosmics than controls, but this was not the case for vision words. It is possible, then, that anosmics rely more strongly on emotional associations for odor-relevant words since they can no longer rely on odor experience. It is also possible, however, that odor and taste words are rated as more positive because the anosmics are aware their experience of the word referent is limited after losing their sense of smell. The words could be emotional because individuals know what they have lost. Another possibility is that in the absence of odor simulation, participants rely more heavily on linguistic co-occurrences (see, e.g., Connell, 2019, Reilly et al., 2021). For example, it has been shown that odor-related words occur in more emotional parts of the English lexicon than vision-related words (Winter, 2016). Anosmics may rely on the emotional content of odor-word neighbors to support word meaning. A similar argument has been made in the context of blind language processing and is a matter of on-going debate (Kim et al., 2019a, Kim et al., 2019b, Lewis et al., 2019, Ostarek et al., 2019).

The present investigation was limited to acquired anosmics, i.e., individuals who have previously been able to smell, and are likely to still possess memories and semantic associations to odors. So, in principle, these anosmics could mentally simulate odor via memory traces of previous olfactory experience, although it has been shown that acquired anosmics have weaker olfactory imagery than control participants (Flohr et al., 2014), suggesting this is unlikely. To shed more light on the issue, the same study could be repeated with individuals who have congenital smell loss (i.e., people born without a sense of smell). Moreover, due to the online nature of the study, we were unable to conduct any physical testing of olfactory ability which could provide additional validation of the anosmic group. We also note the inherent difficulty in drawing conclusions based on null results. An obvious question is whether we had adequate statistical power to detect effects. In terms of sample size, we have twice as many participants as previous studies that have observed action language processing deficits in individuals with Parkinson’s disease (Fernandino et al., 2013b, Fernandino et al., 2013a). In addition, such studies performed statistical analyses by participants only, whilst we used linear mixed effects models that take into account both participant- and item-level variance, and are therefore more powerful (Brysbaert & Stevens, 2018). At a minimum this suggests even if there are simulation effects they are incredibly small in size. It is possible, however, there are other contexts in which odor simulation is more relevant, such as when reading a menu or recipe. Critically, while there is no indication of impairment in odor language processing in anosmics, in some tasks the effects were even in the opposite direction to predictions (e.g., word recall, emotional valence).

It could be argued the tasks used in the present study were semantically shallow, meaning they could easily be completed without necessarily activating semantic information. However, previous studies using a lexical decision task have found that patients with Parkinson’s disease are impaired comprehending action verbs compared to controls (Fernandino et al., 2013a). Furthermore, Reilly et al. (2021) did not find a difference between their anosmic patient and a group of controls in the narration of an olfactory event, which presumably requires deep semantic processing, lending further credence to our conclusions.

Taken together with previous studies of participants with an intact sense of smell (Pomp et al., 2018, Speed and Majid, 2018), as well as studies with anosmic patients (Han et al., 2019, Joshi et al., 2020, Reilly et al., 2021), the evidence so far does not support the proposal that mental simulation of odor occurs during odor language processing. The existing findings suggest the connection between language and olfaction is symmetric, with equally weak connections (for English speakers) between language and olfactory areas in comprehension as production (Majid, 2021, Speed and Majid, 2018). Instead, the evidence suggests odor language may involve high-level representations, such as hedonic information (Pomp et al., 2018, Speed and Majid, 2018). This contrasts with other behavioral paradigms that suggest single words can activate sensory odor information. Olofsson et al. (2012), for example, found responses to odors were facilitated after participants were presented with matching labels, suggesting labels activate odor templates. In their study, however, participants were familiarized with the odors and their labels first, and then presented with the same odor four times in the experimental trials. It is possible, then, that a label can activate an odor representation via short-term odor memory with repetition, but in everyday language processing olfactory activation is not automatic. This requires further exploration.

To conclude, we provide evidence suggesting that odor language is comprehended using high-level odor representations, rather than low-level simulations of odor. Embodied theories of language processing should be fine-tuned to account for differences in mental simulation across the sensory modalities.