The ageing neighbourhood: phonological density in naming

Abstract

Ageing affects the ability to retrieve words for production, despite maintenance of lexical knowledge. In this study, we investigate the influence of lexical variables on picture naming accuracy and latency in adults ranging in age from 22 to 86 years. In particular, we explored the influence of phonological neighbourhood density, which has been shown to exert competitive effects on word recognition, but to facilitate word production, a finding with implications for models of the lexicon. Naming responses were slower and less accurate for older participants, as expected. Target frequency also played a strong role, with facilitative frequency effects becoming stronger with age. Neighbourhood density interacted with age, such that naming was slower for high-density than low-density items, but only for older subjects. Explaining this finding within an interactive activation model suggests that, as we age, the ability of activated neighbours to facilitate target production diminishes, while their activation puts them in competition with the target.

Keywords:

• word retrieval
• phonological neighbourhood density
• ageing

Acknowledgements

This project was supported by Grant Number R03-DC007072 from the NIDCD of the National Institutes of Health. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIDCD or NIH. The authors also wish to acknowledge the valuable work of many students and research assistants, including Holly Kavalier, Megan Slater, Tracy Ball, Stephanie Leeper, Stephanie Cain, Zhen Chen, Ling-Yu Guo, Dawna Duff and Nichole Eden. Thanks also to Gary Dell, Amanda Owen, Karla McGregor and many anonymous reviewers for reading many earlier drafts of the paper.

Notes

1. Although Dell and Gordon (2003) simulated aphasia and we are examining normal ageing in the current study, we do not assume that word retrieval in ageing and aphasia are identical. However, we do think it likely, and a parsimonious hypothesis, that the mechanisms contributing to processing in both populations overlap.

2. Although previous studies have used summed probabilities, summing across phonemes and biphones exacerbates the extent to which these measures are confounded with length (see Storkel, 2004), so we used mean probabilities.

3. To ensure that manually measured and automatically measured RTs were similar, we compared manual and automatic RTs for a subset of 500 pseudo-randomly selected items. They showed a high positive correlation (r = 0.979), with an average discrepancy of 37 msec. In addition, paired t-tests across participants showed that mean naming response times were not significantly different with or without manual measurements for any of the 71 participants (ps from 0.278 to 1.0).

4. This possibility was suggested by an anonymous reviewer. As recommended by Wagenmakers, Krypotos, Criss, and Iverson (2012), we conducted the RT analyses using several scales (raw RTs, log-transformed RTs and z-score RTs). Results were the same, except that the interaction of Age and Initial Phoneme Probability was significant for RTs and log RTs, but not z-score RTs (and of course the main effects of Age and the Intercept were not significant for z-score RTs).

5. It is worth noting, as one of our anonymous reviewers pointed out, that this process of automatically spreading activation creates different dynamics within the lexicon than activation arising from external sources, such as through phonological priming. Although such studies can inform each other, there are many factors, such as the modality and timing of presentation, the goal of the task and features of the prime itself, that are likely to influence whether primes might facilitate or inhibit access to a target word (see Wheeldon, 2003). These are beyond the scope of the current paper, which deals only with internally generated spread of activation.

6. It is also important to keep in mind that what Chen and Mirman modelled was not a change in activation levels within a neighbourhood, but a comparison of neighbourhoods in which neighbours were relatively strong or weak compared to the target by virtue of the neighbourhoods’ different structures. For example, a target from a tightly knit semantic neighbourhood would face competition, whereas a target with only distant semantic neighbours would benefit from facilitation. Because such structural characteristics are not expected to change with normal ageing, these relative effects should be maintained (i.e. closer semantic neighbours should still generate more competition than distant semantic neighbours). However, further simulations are required to determine what the net effect of weakened connections would be in such a model.