Elsevier

Cognition

Volume 100, Issue 3, July 2006, Pages 464-482
Cognition

Cumulative semantic inhibition in picture naming: experimental and computational studies

https://doi.org/10.1016/j.cognition.2005.02.006Get rights and content

Abstract

We report an experiment in which subjects named 120 pictures, consisting of series of five pictures drawn from each of 24 semantic categories (and intermixed with 45 fillers). The number of intervening trials (lag) between successive presentations of members of the same category varied from two to eight. Subjects' naming latencies were slowed by 30 ms for each preceding member of the category. This effect was both cumulative and linear, and unrelated to the lag elapsing since the previous presentation of a category member.

These results definitively demonstrate the occurrence of cumulative interference for word retrieval by prior retrieval of other exemplars of the same semantic category—cumulative semantic inhibition. We claim that this inhibition effect could only occur if the spoken word production system possesses three specific properties (competition, priming, and sharing of semantic activation). We provide computational-modelling evidence in support of this claim.

We show that no current theory of spoken word production has all of these properties. In their current form, all these theories are falsified by these results. We briefly discuss the obstacles that may be encountered by current models were they modified to account for our findings.

Introduction

It is well established that prior presentation of stimuli related in meaning to a target affects the speed and accuracy of processing of that target. The nature of this effect varies according to the modality of presentation, task requirements and lag between the items. For example, if the target is to be read aloud, or is the subject of lexical decision, a semantically related prime will increase speed and accuracy of responses (see Neely, 1991 for a review). In contrast, a semantically related word presented simultaneously with or before a picture will slow the naming of that picture (e.g. Glaser and Dungelhoff, 1984, Schriefers et al., 1990). It is this effect of ‘semantic inhibition’ on word production that is the focus of this paper.

While ‘picture word interference’ tasks (where pictures and words are presented simultaneously or overlapping in time) have provided useful data, particularly regarding the time course of processing in word retrieval, it is widely acknowledged that they can be difficult to interpret and that slight methodological differences can produce marked differences in the results obtained (Damian and Martin, 1999, Nickels, 1997). Moreover the time interval over which a semantic distractor affects word retrieval is limited to only a few hundred milliseconds during the naming of a picture. In this paper, we are concentrating on a different semantic interference phenomenon, where producing the name of a semantically related word slows the retrieval of a target word on a subsequent trial seconds or minutes later.

Wheeldon and Monsell (1994) examined the effects on picture naming of having previously produced a semantically related word in response to a definition (e.g. subjects saw the definition Man's best friend, spoke the answer (“dog”) and then later in the experiment were presented with a picture of a fox to name). Producing a semantically related prime slowed picture naming latencies relative to an unrelated condition. This effect was more marked when two items intervened between target and prime (lag 2) than when the target immediately followed the prime (lag zero1) and was no longer apparent several minutes later (lag >38). Wheeldon and Monsell interpreted their results in terms of three distinct priming effects. The first was a short-lasting facilitatory effect from semantically related primes that is attenuated or eliminated if even one word intervenes between production of the prime and of the target: producing ‘dog’ (in response to a definition) transiently speeds the subsequent immediate naming of a picture of a fox. Second, there is a longer-lasting inhibitory effect from a semantically related prime: producing ‘dog’ slows the non-immediate naming of ‘fox’. This ‘semantic inhibition’ effect is most clearly evident once the short-lasting ‘semantic facilitation’ effect has dissipated, but is still relatively short-lasting in comparison to the third priming effect—repetition priming (e.g. Wheeldon & Monsell, 1992): producing ‘fox’ in response to a definition speeds naming a subsequent picture of a fox. This effect lasts at least 10 min or 100 intervening trials.

As each of these three ‘priming’ effects has a different time-course, Wheeldon and Monsell argue that they arise at different points in the word production process. Here we focus on only the second effect—inhibition from semantically related primes. This ‘competitor-priming’ effect is attributed by Wheeldon and Monsell to priming at the ‘lemma’ level which is the level at which lexical-syntax is represented. In word production, the lemma is activated subsequent to the activation of semantics but prior to activation of phonological form.2

If one semantic competitor slows a subsequent response and this effect is due to priming of the competitor, then sequences of semantically related items should result in cumulative slowing of word retrieval. Although Wheeldon and Monsell found that there was no slowing of naming latencies 4–8 min later, cumulative effects would be expected prior to the complete decay of competitor priming. While several studies have examined competitor effects using sequences of semantically related stimuli (e.g. Damian et al., 2001, Kroll and Curley, 1988, Kroll and Stewart, 1994), only one has examined whether naming latencies change as subjects run through a block of semantically related items (Brown, 1981; Experiment 2). Brown found that across eight items drawn from one category, there was progressive slowing of naming responses, with each item on average approximately 11 ms slower than the preceding response.

With sequences of items from a single semantic category, however, two effects potentially come into play that might ameliorate or even eliminate the effects of increasing numbers of active semantically related competitors. First, participants will be able to predict the category from which a picture is drawn in advance. Secondly, short-term facilitatory semantic priming from the preceding trial will aid retrieval of the target. Hence it is important that unrelated intervening items are added between the exemplars of each category. Brown (1981) used this design in Experiment 4, where members of a category were separated by seven unrelated pictures and found that naming latency showed a linear increase across successive presentations of pictures from the same category, very similar to that found in experiment 2 where items were blocked by category. While Brown's (1981) experiments suggest that there is a cumulative effect of inhibition over the exemplars of a category, in several respects the results remain unclear. His experiments were designed to examine the multiple retrieval of items from a semantic category, so before every stimulus item, participants were provided with the semantic category; the effects of this manipulation are uncertain, but at the very least one would expect that the category name would give some degree of semantic priming to every item in the experiment: when Wheeldon and Monsell 1994; (experiment 3) preceded each picture with a category name, they found that inhibitory priming was no longer greater for lag two as compared to lag zero responses. Furthermore, in Brown's experiment serial position in a list is completely confounded with position within the experiment (r=0.99). Hence the apparent cumulative effects may instead simply reflect general effects of slowing throughout the experiment unrelated to any semantic effects.

In summary, the literature provides clear evidence that semantic competitor primes result in slower latencies for word production. However, it remains unclear the extent to which this priming is cumulative depending on the number of preceding semantically related stimuli and whether there is any interaction between this cumulative priming and the number of intervening (unrelated) stimuli. Here we present an experiment which orthogonally contrasts effects of lag and number of competitors, and which demonstrates a clear effect of number of competitors but no effect of lag on response latency for picture naming. We will show, using a computational model, that accounting for our data requires three mechanisms: spread of activation to semantically related items, repetition priming, and competition between words for lexical selection. While a variety of existing models possess one or two of these mechanisms, none incorporates all three.

Section snippets

Participants

There were 24 participants aged 18–38 (17 women, 7 men) drawn from students and staff of Macquarie University. All had English as their first language and normal or corrected-to-normal vision.

Procedure

Participants were each presented with a sequence of 165 pictures for naming. Embedded within this sequence, there were five exemplars from each of 24 categories (see Appendix A). Pictures from a category were separated by 2, 4, 6 or 8 intervening items (which we call lag); each interval was present once

Reaction times

In analysing the reaction time data, responses with latencies of less than 250 ms (2.9% of responses), no response within 2000 ms (2.6%), and voice key errors (caused by throat clearing etc; 2.4%) were excluded. We also excluded naming errors (14.4%), but included for the RT analyses “acceptable alternatives” (3.0%; see Appendix A for a listing of acceptable alternatives). Our design minimised but did not eliminate confound between ordinal position within a category, and serial position of the

Discussion

This experiment is the first to examine the extent to which there are cumulative effects of semantic competitor priming, controlling for position within the experiment. We found a clear effect, where picture-naming latency is slowed by an additional 30 ms (95% CI±8.2 ms/item) for each preceding semantically related item (but the error rate does not change). Moreover, this effect is identical whether these semantically related exemplars are separated by 2, 4, 6, or 8 intervening unrelated items.

We

Computational data: simulating cumulative semantic inhibition

In order to show that the three properties listed above are sufficient for producing the cumulative semantic inhibition effect, we constructed a minimal model that incorporates all three of these properties. The structure of the model is shown in Fig. 2. There are 165 S-level units (because there were 165 targets in our experiment) each of which has an excitatory connection to the corresponding L-level unit whose weight is specified by the parameter slconnection. Here each unit at the S level

Competition, priming and shared activation: sufficient and necessary

Our computational modelling work has shown that the three properties described above—competition, priming and shared activation—are sufficient to produce results in a simple model that simulate our findings with normal subjects. This is the case regardless of whether a decompositional or non-decompositional view of the semantic system is adopted.

In order to determine whether each of these features is necessary for cumulative semantic inhibition to occur, we ran the first model (with

Conclusions

The existence of the cumulative semantic inhibition effect we report in this paper does not, of course, refute existing models of speech production, but it cannot be explained by any of them in their current forms. Ways in which they might be modified to accommodate our results are apparent, but this will be a non-trivial exercise for modellers of the speech production process for two reasons: (a) will such modifications harm the models' abilities to simulate data which they can currently

Acknowledgements

We would like to thank Anna Woollams for considerable help setting up the experiment and James Moran for testing some of the participants. This research was carried out while David Howard was a visitor to MACCS. Lyndsey Nickels was supported by an Australian Research Council (ARC) QEII fellowship and Max Coltheart by an ARC Federation Fellowship.

References (43)

  • M.T. Bajo et al.

    The locus of semantic interference in picture naming

    Psicológica

    (2003)
  • B. Biedermann et al.

    The representation of homophones: Evidence from remediation

    Aphasiology

    (2002)
  • A.S. Brown

    Inhibition in cued retrieval

    Journal of Experimental Psychology—Human Learning and Memory

    (1981)
  • V. Bruce et al.

    Perceptual priming is not a necessary consequence of semantic classification of pictures

    Quarterly Journal of Experimental Psychology Section A—Human Experimental Psychology

    (2000)
  • A.M. Collins et al.

    A spreading-activation theory of semantic processing

    Psychological Review

    (1975)
  • M.F. Damian et al.

    Locus of semantic interference in picture-word interference tasks

    Psychonomic Bulletin and Review

    (2003)
  • M.F. Damian et al.

    Semantic and phonological codes interact in single word production

    Journal of Experimental Psychology—Learning Memory and Cognition

    (1999)
  • G.S. Dell

    A spreading activation theory of retrieval in sentence production

    Psychological Review

    (1986)
  • G.S. Dell et al.

    Lexical access in aphasic and nonaphasic speakers

    Psychological Review

    (1997)
  • K.I. Forster et al.

    DMDX: A windows display program with millisecond accuracy

    Behavior Research Methods Instruments and Computers

    (2003)
  • W.R. Glaser et al.

    The time course of picture–word interference

    Journal of Experimental Psychology: Human Perception and Performance

    (1984)
  • Cited by (0)

    View full text