Elsevier

Acta Psychologica

Volume 144, Issue 3, November 2013, Pages 571-582
Acta Psychologica

Semantic interference in language production is due to graded similarity, not response relevance

https://doi.org/10.1016/j.actpsy.2013.09.006Get rights and content

Highlights

  • Response exclusion and lexical competition accounts were tested.

  • We varied semantic relation and response relevance orthogonally.

  • Semantic similarity is the only critical determinant for semantic interference.

Abstract

There is an ongoing debate on the question whether semantic interference effects in language production reflect competitive processes at the level of lexical selection or whether they reflect a post-lexical bottleneck, occupied in particular by response-relevant distractor words. To disentangle item-inherent categorical relatedness and task-related response relevance effects, we combined the picture–word interference task with the conditional naming paradigm in an orthogonal design, varying categorical relatedness and task-related response relevance independent of each other. Participants were instructed to name only objects that are typically seen in or on the water (e.g. canoe) and refrain from naming objects that are typically located outside the water (e.g. bike), and vice versa. Semantic relatedness and the response relevance of superimposed distractor words were manipulated orthogonally. The pattern of results revealed no evidence for response relevance as a major source of semantic interference effects in the PWI paradigm. In contrast, our data demonstrate that semantic similarity beyond categorical relations is critical for interference effects to be observed. Together, these findings provide support for the assumption that lexical selection is competitive and that semantic interference effects in the PWI paradigm reflect this competition.

Introduction

Selecting words from the mental lexicon that express an intended message appropriately is a core component of the speech production system. This process involves a spread of activation within and between different levels of speech planning. For instance, upon naming an object, activation spreads to semantically related nodes at the conceptual level where the pre-verbal message is generated. These nodes in turn activate their corresponding entries at the lexical level. As a result of this multi-level spreading activation, the activation of the target word is flanked by concomitant activation of related words. Thus, even for basic and simple instances of speech production such as the naming of visually depicted objects (e.g., chair), semantically related concepts and their lexical entries (e.g., table, wardrobe) are concurrently activated.

In this paper we explore the consequences of lexical co-activation for word production. We ask whether lexical selection is characterized by competition from co-activated entries or unaffected by the activation status of potential alternatives. Inhibitory effects of semantic contexts on production latencies have long been taken to reflect competition at the level of lexical selection. For instance, when pictures of objects are named in the presence of visually or auditorilly presented distractor words in the picture–word interference (PWI) paradigm, a semantic interference effect is observed: naming is slowed in the presence of categorically related relative to unrelated words (e.g., Schriefers, Meyer, & Levelt, 1990). Likewise, repeated naming is slowed in blocks of trials consisting of categorically or associatively related objects (semantically homogeneous blocks) relative to heterogeneous blocks consisting of unrelated objects in the blocking paradigm (e.g. Abdel Rahman and Melinger, 2007, Belke et al., 2005Damian et al., 2001, Kroll and Stewart, 1994).

Many models of speech production account for semantic interference effects by assuming that lexical selection is a competitive process (e.g. Bloem and La Heij, 2003, Bloem et al., 2004, La Heij et al., 2006, Levelt et al., 1999, Roelofs, 1992, Roelofs, 2003). For example, according to Levelt and colleagues (e.g. Levelt et al., 1999;Roelofs, 1992) semantic interference effects in the PWI paradigm arise because semantic alternatives are co-activated by the target picture and distractor word at the conceptual and lexical level. Co-activated lexical entries compete with the target for selection, thus delaying the naming response. In contrast, when unrelated words are presented, activation spread by target and distractor word diverges onto different lexical entries, and lexical competition is reduced.

An alternative proposal suggests that lexical selection is non-competitive (Costa et al., 2005, Finkbeiner and Caramazza, 2006a, Finkbeiner and Caramazza, 2006b, Janssen et al., 2008, Mahon et al., 2007, Miozzo and Caramazza, 2003). According to the response exclusion hypothesis (REH) by Mahon et al. (2007) semantic interference effects in the PWI task are localized at the post-lexical stage of the articulatory output buffer. Distractor words have privileged access to the articulators, and the output buffer forms a bottleneck stage that can be engaged with only one process at a time. Thus, the distractor must be removed from the buffer before the target word can be articulated. The speed of this exclusion process is determined by response relevant criteria. Words that can be quickly dismissed as potentially relevant responses (e.g. unrelated words from different semantic categories) can be excluded faster than words that satisfy response relevant criteria (e.g. semantic category members: when naming a dog the distractor cat fulfills the response relevant criterion of belonging to the same broad category of animals).

Even though the decision mechanism on the response relevance of distractors as such yields discrete results (a distractor is a relevant response or it is not), what counts as a criterion for response relevance is not exclusively determined by semantic category membership. Depending on the goals of the task at hand, different item-inherent or task-related criteria can determine the response relevance of a distractor. This assumption is explicitly formulated in Mahon et al.'s work (2007; p. 512): “There are, in principle an indefinite number of response-relevant criteria, because such criteria are, in part, a product of task constraints decided by the experimenter”. Thus, constraints that determine the response relevance of distractors, rather than competitive lexical selection mechanisms, are assumed to be the major source of semantic interference effects.

Evidence in favor of the response exclusion hypothesis stems among others from observed exceptions from classic categorically induced interference effects in the PWI paradigm. For instance, semantically related verb distractors (Mahon et al., 2007), distractors that have a part-whole relation with the target (Costa et al., 2005) and associatively related distractors (Alario et al., 2000, Abdel Rahman and Melinger, 2007; but see Aristei, Melinger, & Abdel Rahman, 2010) induce facilitation, rather than interference. One common element between these types of distractors can be seen in terms of response relevant criteria: given the task at hand (naming whole objects by producing nouns; e.g., target: camel) verb distractors (e.g., ride) can quickly be excluded as potential responses based on their grammatical class membership; likewise, distractors referring to parts of objects (e.g. hump) can be excluded because the implicit task criterion is to name whole objects, and associates (e.g., pyramid) can be excluded because they are not semantic category members. Thus, exclusion times for these types of response irrelevant distractors should be comparable to unrelated words. However, because all of these distractors are semantically related to the target, facilitation due to semantic priming is observed (but see e.g., Roelofs, Piai, & Schriefers, 2012 for a critical review and alternative interpretations of these findings; Abdel Rahman and Melinger, 2009a, Abdel Rahman and Melinger, 2009b, Kuipers et al., 2006; for alternative accounts of semantic facilitation and interference effects that maintain the assumption of lexical competition).

According to the response exclusion hypothesis a semantic priming mechanism at the lexical level is assumed to facilitate naming. In contrast to the discrete response exclusion mechanism responsible for the interference effects, the priming mechanism is graded, varying with the semantic distance between target and distractor. Specifically, given equivalent levels of response relevance, semantically close words (e.g., target: horse, distractor: donkey) are assumed to yield stronger priming effects at the lexical level than more distant words (e.g., target: horse, distractor frog; Mahon et al., 2007). Three experiments run by Mahon et al. (2007) confirmed this hypothesis, although results were not always clear cut. In fact, across the three experiments there are internal discrepancies that were not further discussed by the authors. For instance, semantically close distractors did not always induce interference effects relative to unrelated distractors despite their response relevant status. Furthermore, an SOA manipulation yielded semantic distance effects only at a negative SOA (− 160 ms) but not at zero SOA, at which semantic facilitation has been reported for, e.g., part-whole relations (Costa et al., 2005). There is in our view no apparent common element that could explain both supportive and discrepant findings within the study, but these inconsistencies together with no available replication of the effects (see Lee and de Zubicaray, 2010, Vieth et al., 2012) invite caution in the interpretation of these results.

Additionally, Mahon et al.'s results contrast with the majority of studies on semantic similarity effects conducted so far, which demonstrated that close relations are associated with stronger interference than more distant relations in semantic blocking and PWI paradigms (Vigliocco et al., 2002, Vigliocco et al., 2004, Lee and de Zubicaray, 2010, Abdel Rahman et al., 2010; see also Aristei et al., 2010).

The discussed examples for different types of distractor words in different experimental settings (see above) suggest that response relevance is not solely determined on the basis of item-inherent features and coarse semantic information, but depends to a large degree also on experimental contexts and task constraints (Mahon et al., 2007). In its current formulations the response exclusion hypothesis does not provide explicit information about the individual contributions of response relevant criteria explicitly defined in task instructions and of those derived from the target stimuli (e.g. categories), nor about the dynamics of their potential interplay. Nonetheless, it is clear in the literature (e.g., Mahon et al., 2012, Mahon et al., 2007) that explicitly defined and implicitly derived rules are both driving forces of the response exclusion mechanism in terms of response relevance.

In this study we go further in testing response relevance effects in speech production. Until now, relevant semantic item-inherent information was manipulated mainly by means of the selected target categories (e.g., Costa et al., 2005, Mahon et al., 2007), thus, its extraction occurred more implicitly and with dependence on individual response strategies. Here, we investigate response relevance effects by introducing relevant item-inherent semantic information in the task instructions. To do so, we employed the conditional naming paradigm in which picture naming is conditional on a classification of the object as belonging to a pre-specified category (Job and Tenconi, 2002, Mulatti et al., 2010). For instance, Job and Tenconi (2002) presented a series of living and non-living objects and instructed their participants to name only living things and to withhold the naming response when non-living objects were presented (and vice versa). Interestingly, the authors demonstrated that conditional naming, albeit including an additional semantic classification, is not associated with additional costs compared to unconstrained free naming of all pictures (but see Mulatti et al., 2010). While the specific mechanisms giving rise to this no-cost phenomenon are yet to be fully identified (e.g., Aristei et al., 2009, Aristei et al., 2007), the paradigm is well-suited to explore distractor response relevance.

Here, we combined the conditional naming procedure with the picture–word interference paradigm. Participants were instructed to name only those objects that are typically located in or on the water (e.g., canoe), and to refrain from naming objects that are typically located outside the water (e.g., bike), and vice versa. Object pictures were presented simultaneously with categorically related or unrelated distractor words, that can equally be located in or outside the water (see example below). Thus, by combining conditional naming with the PWI paradigm we can isolate the effects of categorical relatedness and the task-dependent response relevant status of distractor words. For instance, when naming is conditional on the object being typically found in or on the water (e.g. target: “carp”), only items that satisfy this criterion are potentially relevant responses, irrespective of their semantic category membership. For instance, the categorically related distractor herring is response relevant (only objects typically located in the water should be named), whereas the categorically related distractor gecko is not a potentially relevant response (because geckos are not typically found inside the water). In the unrelated condition the distractor word pier is response relevant, whereas train is neither related nor response relevant (see Fig. 1 and Appendix).

According to the response exclusion account, all distractors occupy the articulatory output-buffer and must be removed before the target name can be articulated. On a general level, removal is slower for distractors that fulfill response relevant criteria relative to distractors that do not.

A central assumption is that distractor response relevance is based on the affordances of the task at hand and thus, should be substantially affected by task instructions. This assumption is made explicit in Mahon et al.'s work (2007, p. 517): “…a mere change in task instructions can reverse the polarity of the distractor effect from semantic interference to semantic facilitation. In other words, the same materials that produce semantic interference under one set of task instructions produce semantic facilitation under a different set of task instructions.”

Accordingly, response relevance here is determined by the instructions for the conditional naming task, and semantic category membership is not diagnostic in terms of response relevance.

Therefore, when participants are instructed to name things that are located in or on the water, related and unrelated distractors that possess this attribute (they are likewise located in or on the water) are response relevant and should delay naming times relative to irrelevant distractors lacking this critical attribute (see Fig. 1). This task-dependent response relevance should be reflected in a slowdown of RTs irrespective of the categorical relation between target and distractor, and should therefore be seen for related as well as for unrelated distractors.

With respect to the effects of categorical relatedness there are two theoretical alternatives. First and most parsimoniously, the explicit task criterion in the conditional naming procedure should override otherwise response relevant item-inherent properties such as category membership. This is because category membership is not a valid and diagnostic criterion for determining relevant and irrelevant responses in the conditional naming task. Thus, only the conditional naming criterion, but not a broad categorical relation, would bear on the response relevant status of distractor words. Furthermore, because any semantic relation to the target is assumed to prime the target lexical entry within equal levels of response relevance (Mahon et al., 2007), we expect purely facilitatory effects for categorically related relative to unrelated words.

Different predictions could be derived if one were to assume that explicit task-relevant constraints cannot override the inherent response relevance of categorically related relative to unrelated distractors. In this case there would be two different types of response relevant criteria, one determining whether to respond or not (the conditional naming criterion), and a second one that has no relation to response decisions and the performance of the specific task at hand (categorical relatedness). However, this idea is in contrast to the description of the response exclusion hypothesis, stating that task instructions are responsible for the reversals of semantic distractor effects (Mahon et al., 2007; but see Kuipers et al., 2006, Kuipers and La Heij, 2008 for an alternative explanation in line with lexical competition).

Alternatively, one might also argue that the conditional naming criterion is applied only to the target pictures and has no influence on distractor response relevance, which depends on categorical relatedness by default. Such a principle is discarded by the REH itself, as it would have fatal consequences for the suitability of the exclusion mechanism. It would in fact imply that the response exclusion mechanism could be fooled by the semantic content of the input, and response relevance as such would be irrelevant.

To bypass this problem, the response exclusion hypothesis assumes that, the decision mechanism is aware of the distractor status of the representation in the buffer (e.g. Mahon et al., 2007, p. 524), and word semantic content specifically relevant to the task is processed and affects the decision mechanism.

Therefore, as described above, the conditional naming criterion, and not semantic category membership, is predicted to determine the time required to exclude words from the output-buffer.

Besides these theoretical arguments against the idea of task-independent response exclusion criteria, we will add a more direct test by including a free naming task described in detail below.

By contrast, according to lexical competition models response relevance should not affect semantic interference effects (but see e.g., Kuipers & La Heij, 2008). Because lexical selection is competitive, categorically related distractors should yield slower RTs than unrelated words. Hence, we expect a classic semantic interference effect, and this should be of comparable magnitude for targets in the response relevant and irrelevant distractor conditions.

Due to the orthogonal manipulation of categorical relatedness and task-related response relevance, the target pictures were necessarily combined with different distractor words in the relevant and irrelevant conditions. Consequently, relevant and irrelevant distractors may differ systematically in their semantic similarity to the target. Because the conditional naming criterion is based on a semantic attribute, response relevant related words (e.g., herring) are likely to have a closer semantic relation to the target (e.g., “carp”) than response irrelevant related words (e.g., gecko), and the same holds for the unrelated condition (e.g., pier vs. train). This is because response relevant distractors share the semantic feature of being located inside/outside the water, and associated semantic attributes. It has be shown, for instance, that non-categorical relations such as thematic associations (e.g., associates of the thematic context apiary: bee, honey, honeycomb etc.; Abdel Rahman & Melinger, 2007) or even transient formations of ad-hoc categories (e.g., things that may be present on a fishing trip; Abdel Rahman & Melinger, 2011) can also induce semantic interference effects. Likewise, the categories of things that can be found in or outside the water (and associated attributes) may induce such interference effects. Thus, even though we orthogonally manipulated task-related response relevance and categorical relatedness of targets and distractors, we cannot exclude possible differences between response relevant and irrelevant distractors that are due to differences in semantic similarity.

Therefore, to disentangle response relevance and graded semantic similarity beyond categorical relations, we conducted a separate semantic similarity rating with all target and distractor pairs used in the conditional naming task. This rating serves two purposes. First, it should reveal any differences between relevant and irrelevant distractors in terms of semantic similarity. Second, it will be included as a predictor in the statistical analyses, providing further information about the relation between semantic similarity and response relevance and their individual contributions to the effects. To anticipate the results, the ratings showed in fact closer semantic relations between target-distractor pairs in the relevant than in the irrelevant condition.

Furthermore, we included an unconstrained naming task (from now on: free naming) as a control condition (see below).

In the free naming task, all objects were named without restrictions (the typical situation in the majority of picture–word interference experiments). When all pictures should be named, words that are classified as response relevant or irrelevant in the conditional naming task should not induce any differential effects on response latencies in the free naming task. Therefore, for free naming, response exclusion and lexical competition accounts predict the classic pattern of categorical semantic interference effects. Thus, according to the response exclusion hypothesis response relevance effects should emerge only in conditional naming (see above), whereas no such effect should be seen in the free naming task. In contrast, according to lexical competition accounts semantic interference effects should be equivalent in the conditional and free naming task.

Complementing the semantic similarity rating, the free naming task allows us to reveal any differences between the relevance conditions in terms of semantic similarity. As in free naming the response relevance dimension as defined by the conditional naming instruction is not relevant, any differences between the “response relevance” conditions reflect the influence of other factors. Most likely, they can be attributed to differences in semantic similarity, with “relevant” distractors having a closer relation to the target than “irrelevant” distractors.

Importantly, if differences between relevant and irrelevant distractors are due to graded differences in terms of semantic similarity, specific predictions can be derived. In this case we expect comparable latency differences between relevant and irrelevant words in the conditional naming task (where the property inside/outside the water is critical for correct task performance), and in the free naming task (where response relevance is not defined upon the typical location of the object). Thus, together with the semantic similarity rating, free naming enables us to disentangle task-related response relevance and graded semantic similarity.

As to the polarity of semantic similarity effects, the two theoretical approaches predict opposite results. According to the response exclusion hypothesis a semantically close distractor from the same superordinate category should prime the target more than a semantically distant distractor from the same category (e.g., target: horse; distractor: zebra vs. whale; Mahon et al., 2007; see discussion above). Therefore, with the response relevant status being equal (e.g., in the free naming task, within the class of related and within the class of unrelated words), close distractors should facilitate naming compared to distant distractors.

In contrast, according to lexical competition models strongly activated concepts should yield stronger competition, and thus longer naming latencies, than weakly activated distractors (e.g., Aristei et al., 2010, Roelofs, 1992; see also Abdel Rahman and Melinger, 2009a, Abdel Rahman and Melinger, 2009b). Therefore, semantically close distractors should yield stronger interference effects than distant distractors.

Importantly, for the present purpose, the polarity of graded similarity effects is not the most critical aspect. If any effects are present in both tasks, we can conclude that they reflect differential effects of graded semantic relations between target and distractor, rather than distractor response relevance.

Section snippets

Participants

23 females and 7 males, aged from 20 to 45 years (mean age = 26.3 years), were paid for their participation in the experiment or received partial fulfillment of a curriculum requirement. All participants were native German speakers and reported normal or corrected-to-normal vision.

Materials

The target picture set consisted of 100 color photographs of common objects, scaled to 3.5 × 3.5 cm. Half of the objects are typically located in or on the water (e. g. yacht), and the other half are typically located

Results

Mean response times (RTs) for correct trials, standard error of means, and mean percentages of errors in the experimental conditions are presented in Table 1. Trials with incorrect naming, stuttering, mouth clicks, or vocal hesitations, and trials with voice key failures or malfunctioning were discarded from the RT analysis (across all participants the percentage of naming responses in the no-go condition, i.e., “false alarms”, was 0.69%). Furthermore, trials with RT latencies faster than 330 ms

Discussion

In the present study we tested a central assumption of the response exclusion account according to which, distractor-induced interference effects are determined by the response relevant status of distractor words, rather than their semantic relation with the target. Often, these two factors are confounded. To disentangle the individual contributions of item-inherent categorical relatedness and task-related response relevance effects, we combined the picture–word interference paradigm with the

Acknowledgments

This work was supported by grants AB277 4 and 5 from the German Research Foundation (DFG) to Rasha Abdel Rahman. We thank Wido La Heij, Nicolas Dumay and two anonymous reviewers for their comments on a previous version of the manuscript.

References (53)

  • C. Mulatti et al.

    Picture-word interference and the response-exclusion hypothesis

    Cortex

    (2012)
  • A. Roelofs

    A spreading-activation theory of lemma retrieval in speaking

    Cognition

    (1992)
  • H. Schriefers et al.

    Exploring the time course of lexical access in production: Picture–word interference studies

    Journal of Memory and Language

    (1990)
  • P.A. Starreveld

    On the interpretation of phonological context effects in word production

    Journal of Memory and Language

    (2000)
  • G. Vigliocco et al.

    Semantic distance effects on object and action naming

    Cognition

    (2002)
  • G. Vigliocco et al.

    Representing the meanings of object and action words: The featural and unitary semantic space hypothesis

    Cognitive Psychology

    (2004)
  • R. Abdel Rahman et al.

    Semantic distance effects. Presentation to the workshop on competition effects in language production

    (2010)
  • R. Abdel Rahman et al.

    When bees hamper the production of honey: Lexical interference from associates in speech production

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2007)
  • R. Abdel Rahman et al.

    Dismissing lexical competition does not make speaking any easier: A rejoinder to Mahon and Caramazza (2009)

    Language & Cognitive Processes

    (2009)
  • R. Abdel Rahman et al.

    Semantic context effects in language production: A swinging lexical network proposal and a review

    Language & Cognitive Processes

    (2009)
  • R. Abdel Rahman et al.

    The dynamic microstructure of speech production: Semantic interference built on the fly

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2011)
  • F.-X. Alario et al.

    Semantic and associative priming in picture naming

    The Quarterly Journal of Experimental Psychology

    (2000)
  • S. Aristei et al.

    High and low spatial frequencies role in visual categorization during picture naming: ERP double dissociation

  • S. Aristei et al.

    Category-specific visual processing: ERPs and conditional picture naming paradigm

  • S. Aristei et al.

    Electrophysiological chronometry of semantic context effects in language production

    Journal of Cognitive Neuroscience

    (2010)
  • L.W. Barsalou

    Ad hoc categories

    Memory & Cognition

    (1983)
  • Cited by (0)

    View full text