Serial position effects in graphemic buffer impairment: An insight into components of orthographic working memory

ABSTRACT

This study investigated the nature of graphemic buffer functioning and impairment, through analysis of the spelling impairment shown by GEC, a man with acquired dysgraphia and clear characteristics of graphemic buffer impairment. We discuss GEC’s error patterns in relation to different processes of orthographic working memory. This is the first study to show the contribution of these processes in one individual through performance on different spelling tasks. GEC’s spelling errors in writing to dictation showed a linear serial position effect, including deletions of final letters. These “fragment errors” can be explained as the result of information rapidly decaying from the buffer (reduced temporal stability). However, in tasks that reduced working memory demands, GEC showed a different error distribution that may indicate impairment to a different buffer process (reduced representational distinctiveness). We argue that different error patterns can be a reflection of subcomponents of orthographic working memory that can be impaired separately.

KEYWORDS:

• Graphemic buffer
• orthographic working memory
• dysgraphia
• spelling
• fragment errors

Acknowledgements

During the preparation of this paper, Trudy Krajenbrink was funded by an International Macquarie University Research Excellence Scholarship (IMQRES 2011045), Lyndsey Nickels by an Australian Research Council Future Fellowship (FT120100102), and Saskia Kohnen by a Macquarie University Research Fellowship (MQRF). The authors are grateful to GEC for participating in the study, and would also like to thank Brenda Rapp and Teresa Schubert for valuable feedback on a previous version of the manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In order to maximize the number of items for different error analyses we have used slightly different subsets of items from the total set of 705 words administered for writing to dictation which we will outline at the start of each analysis:

1. The overall error analysis (Table 7) was based on all words written to dictation (n = 705, of which 586 resulted in an error);

2. The regression analysis used the initial set of 705 words with the exclusion of double items and items without imageability ratings (n = 609 words remaining);

3. The buffer error analysis (Table 8) used the set of 609 words from the regression analysis and removed the correct, no responses, and semantic errors, leaving 570 errors for analysis;

4. The analysis of fragment errors considered all 586 errors from the total set of 705 words and counted the fragment errors according to the definition by Ward and Romani (1998).

2 In this analysis, the target and response were maximally aligned. Any letter that was present in the correct (relative) position was given one point. Relative position indicated the position of a letter in relation to the other letters in the word. For example: the response huse for house resulted in a score of 1 for the u, as it was still in the correct relative position in relation to its adjacent letters (even though after deletion of o it was no longer in the correct absolute (third) position). A letter that was deleted or substituted received 0 points. Any letter present but in the incorrect position received 0.5 point. When two adjacent letters were transposed, both were scored as 0.75. For example, the target ‘algae’ written as ‘agli’ received a score of 1+0.75+0.75+0+0 = 2.5, and therefore a letter accuracy of 2.5/5 = 0.5.

3 An error was classified as orthographically related when either at least 50% of target letters were in the response (task: trash), or at least 50% of response letters were target letters (hatred: hit; based on Nickels (1995) analysis of phonological errors in spoken production). We considered all segmental errors to be orthographic errors and not phonological as they often showed a strong orthographic rather than phonological relationship (e.g., yacht: yah).

4 It seems important for the discussion of fragment errors to agree to a clear definition of fragment errors and analysis of these errors. Ward and Romani (1998) categorized non-word responses that were two letters shorter than the target as fragment errors, including unrelated fragments (e.g., house as gib), which could be the result of a different impairment compared to correct fragments. Furthermore, as mentioned previously, it is important to include multiple letter errors, as some of these may in fact be fragment errors: Sage and Ellis reported that 20% of BH’s error involved omission of more than one letter, which often concerned the final part of a word (sledge: sle) (Sage & Ellis, 2004).

5 The analysis conducted on n=70 single letter errors for GEC found the same linear decrease of accuracy, with a significant effect of position (χ ² (4) = 37.79, p = < .001). Letters in initial position were always written correctly.

6 For 4 letter-words (e.g., flag), the initial letters were the first two (FL), medial the middle two (LA), final letters were the final two (AG).

7 We would like to thank Adam Buchwald for suggesting this task.

8 Schiller et al. (2001) argued that one could, for example, even in backward spelling, access the final letters in a forward manner by repeatedly working towards this position: for example when spelling chair, in order to retrieve the final letter, one could scan the word until R is reached (C-H-A-I-R) and then rescan to retrieve the next letter (C-H-A-I), and so on.

9 We would like to thank Brenda Rapp for this suggestion

Additional information

Funding

During the preparation of this paper, Trudy Krajenbrink was funded by an International Macquarie University Research Excellence Scholarship (IMQRES 2011045), Lyndsey Nickels by an Australian Research Council Future Fellowship (FT120100102), and Saskia Kohnen by a Macquarie University Research Fellowship (MQRF). The authors are grateful to GEC for participating in the study, and would also like to thank Brenda Rapp and Teresa Schubert for valuable feedback on a previous version of the manuscript.