(Not) Confirming Horse-race Models of Word-Naming

Christopher D. Green
Department of Psychology
York University

Philip R. Groff
Department of Psychology
University of Toronto

The Phenomenon

One would expect almost all cognitive processes to slow down when carried out under a short-term memory (STM) load. Paap & Noel (1991) have found, however, that people name words that are low-frequency and orthographically irregular faster while holding five single-digit numerals in STM than when holding only one.

Results of Paap & Noel (1991)

(Notice that naming latencies increase with size of short-term memory load
for all but low-frequency, orthographically irregular words.)

The Theory

Their hypothetical explanation for this finding is that there are two routes by which words are named: phonological, in which the sound of the word is assembled phoneme-by-phoneme, and lexical, in which the pronunciation of the whole word is stored in long-term memory (LTM). These to routes are said to "race" with each other, the pronunciation candidate of the "winner" being sent on to the motor routine for actual pronunciation.

The lexical route is typically the faster of these because the pronunciation of the word is obtained all at once, whereas the phonological route uses more time converting each letter (or letter-cluster) into phonological form, holding it in STM, and then returning to the word for further processing. Past research has shown, however, that low-frequency (LF) words take longer to name, even if retrieved from the mental lexicon, thus making the phonological route more competitive for these words, often resulting in a "photo finish" which some "central executive" sorts out. If the output of both route is the same, there is little problem. Orthographically irregular (I) words, however, give rise to false output by the phonological route. Thus LF-I words cause a conflict between the correct output of the lexical route and the incorrect output of the phonological route. Resolution of the conflict takes additional time to sort out, leading to very long latencies for words of the LF-I class.

By occupying STM, Paap & Noel reasoned, the phonological route should be slowed down even further, thus allowing the lexical route, with its correct output, to win the "race" without the need for a "photo finish," thereby speeding up overall naming processing.

Replication History

Because the Paap & Noel phenomenon is so counter-intuitive, replication is a very important matter.

Herdman (1993) was able to replicate the effect several times, but showed that it did not occur if STM was occupied with graphical rather than numerical material. He concluded that the effect is one of verbal "crosstalk" rather than one of STM per se.

Pexman & Lupker (1993), were unable to replicate the effect at all, in five experiments, under various procedural modifications.

Bernstein & Carr (in press) were able to replicate the effect, including when the STM load was graphical, but only in subjects exhibiting an interaction between word-frequency and orthographic irregularity (Seidenberg, et al., 1984). They argued that these subjects were better readers, and it is only in these subjects that the Paap & Noel effect is present.

We decided to examine this hypothesis but, in contrast to Bernstein & Carr, we expected poorer readers to show the Paap & Noel effect because they will employ their phonological route mode often than good readers because they will be familiar with fewer words (i.e., due to smaller mental lexicons).

Experiment 1

Method. Twenty university students underwent a procedure nearly identical to Paap & Noel's. In each trial, the subject would be shown (1) either 1 or 5 numerals for 2 sec., which they had been instructed to remember until the end of the trial; (2) a word to be named, which was masked upon the naming response; (3) a single probe numeral. Subjects indicated whether or not they believed the probe numeral to have been among those shown in step (1). There were 80 trials. Afterwards, each subject was administered an AMNART, to test their vocabulary, and the Digit Span portion of the Wechsler Memory Scale.

Results. Low-vocabulary subjects showed significant effects for word-frequency (F(1,9)=8.88, p=.02), memory-load (F(1.9)=14.14, p=.01), and the standard frequency×regularity interaction (F(1,9)=13.86, p=.01). There was no significant effect for orthographic regularity, nor for any of the other interactions.

High-vocabulary subjects showed significant effects for frequency (F(1.9)=13.16, p=.01), regularity (F(1,9)=14.05, p=.01), and the frequency×regularity interaction (F(1,9)=16.69, p<.01). There was no effect for memory load, nor for any of the other interactions.

Discussion. Both groups showed frequency effects, as expected, and both showed the standard frequency×regularity interactions. Neither group showed evidence of the Paap & Noel effect, however. Surprisingly, the low-vocabulary group showed a general decrease in naming latency in the high memory-load condition. Paap & Noel found no effect for memory load.

Experiment 2

It was thought that perhaps the five numerals in the high memory-load condition of Experiment 1 did not sufficiently tax the STM system of the subjects to cause the Paap & Noel effect. Pexman & Lupker ran a condition with seven numerals, but since there are only ten numerals to choose from, it was hypothesized that clever subjects might hold in memory only those digits not presented on a given trial (viz., only three). Another possibility is that five numerals might be held in memory "iconically" as a set, rather than as five separate "chunks" of information. Consequently, we decided to rerun Experiment 1, using number-words (i.e., one, two, etc.) rather than numerals for the memory-load manipulation.

Method. The same as in Experiment 1, but with number-words instead of numerals.

Results. Low-vocabulary subjects showed no significant effects whatsoever, although the effect for word-frequency came relatively close to approaching significance (F(1,7)=3.32, p=.11). The naming latencies of irregular words increased with increased memory loads (as expected, and contrary to the results of Experiment 1), though not significantly. Those for regular words were flat across memory load conditions.

High-vocabulary subjects showed only a significant regularity×memory-load interaction (F(1,7)=8.69, p=.02), and a significant three-way frequency×regularity×memory-load interaction (F(1,7)=5.73, p=.05). They also showed a near-significant frequency×regularity interaction (F(1,7)=4.00, p=.09). The general trend was toward increased latency with increased memory load. Unexpectedly, the HF-R words showed the opposite trend.

Discussion. The results in this experiment were not as clear as those in Experiment 1. Even some standard effects were absent. What was clear was that, again, neither group showed the Paap & Noel effect.

Experiment 3

Having failed to confirm Paap & Noel's findings using their own word-set, we decided to use the word-set of Herdman, who has been most successful in replicating Paap & Noel's results.

Method. Identical to Experiment 1, but using Herdman's words instead of Paap & Noel's.

Results. Low-vocabulary subjects showed the standard effects for frequency (F(1,9)=15.21, p<.01) and for regularity (F(1,9)=20-51, p<.01) They also displayed a marginal effect for memory-load (F(1,9)=4.20, p=.07), as they had in Experiment 1. There was also a significant frequency×memory-load interaction effect (F(1,9)=6.37, p=.03).

High-vocabulary subjects showed only a significant effect for frequency (F(1,9)=24.58, p<.01). No other effects were significant. There was also a slight trend toward a regularity effect (F(1,9)=2.92, p=.12).

Discussion. For a third time, neither group clearly evinced the Paap & Noel effect, though the graph for the low-vocabulary group showed trends in the right direction for all four conditions. Standard frequency and regularity effects were present, but Seidenberg's frequency×regularity interaction failed to appear.

Experiment 4

As in Experiment 2, we thought that memorizing number-words might tax the subjects' memories more than numerals, and finally reveal the Paap & Noel effect.

Method. Identical to Experiment 3, but with number-words instead of numerals.

Results. Low-vocabulary subjects showed a significant effect for word-frequency (F(1,9)=26.95, p=.00) . No other effects were significant.

High-vocabulary subjects showed significant effects for frequency (F(1,9)=39.48, p=.00) and regularity (F(1,9)=10.40, p=.01). No other effects were significant.

Discussion. Again, standard frequency and regularity effects were present. The Seidneberg frequency×regulatiry interaction was not, however. Most important, for the fourth time, neither group of subjects clearly showed the Paap & Noel effect, though this time there is a hint of it in the graph for the high-vocabulary subjects.

General Discussion

We found no compelling evidence for the Paap & Noel effect in any of the conditions we ran. A fortiori, we found no reason to believe, as per our hypothesis, that differences in vocabulary level had any effect of the strength of the effect. This result cannot be explained by a failure of our procedure because the standard effects for frequency and regularity were reliably found. Until it can be explained why Paap & Noel's word-naming effect is so inconsistent, it will remain of dubious validity, and as such, cannot serve as a buttress of any firmness to the dual-route model of word processing.


Bernstein, S. E. & Carr, T. H. (in press). Dual route theories of spelling to pronunciation: What can be learned from concurrent task performance? Journal of Experimental Psychology: Learning, Memory, and Cognition.

Herdman, C. (1993, June). Paper presented at the conference of the Canadian Society for the Brain, Behavioral, and Cognitive Sciences, Toronto.

Paap, K. R. & Noel, R. W. (1991). Dual-route models of print to sound: Still a good horse race. Psychological Research, 53, 13-24.

Pexman, P. M. & Lupker, S. J. (1993, June). Does memory load facilitate naming of low frequency irregular words? Poster presented at the conference of the Canadian Society for the Brain, Behavioral, and Cognitive Sciences, Toronto.

Seidenberg, M. S., Waters, G. S., Barnes, M. A., & Tanenhaus, M. K. (1984). When does irregular spelling or pronunciation influence word recognition? Journal of Verbal Learning and Verbal Behavior, 23, 383-404.