Speaking is a seemingly effortless task executed on a regular basis. Yet, speaking is a controlled action (e.g., Roelofs and Piai, 2011). As a broad term, control refers to regulatory/monitoring processes that ensure that our actions are in agreement with our goals, especially in the face of distraction (e.g., Posner and Petersen, 1990). We need to exert control over language production for various reasons. For example, speakers need to prevent interference from concurrent visual or auditory information present in their environment. Moreover, when retrieving words from long-term memory, associated information that is not relevant to the task at hand is also retrieved. Control must also be exerted over these interfering memory representations so the appropriate word can be selected (e.g., Badre and Wagner, 2007). Whereas the neuroanatomical characterization of core language-production processes has advanced considerably (e.g., Indefrey, 2011; Price, 2012), only recently has research been dedicated to the control aspects of production (e.g., Alario et al., 2006; Piai et al., 2013; Geranmayeh et al., 2014; Riès et al., 2014, 2015). In the present study, we examined the role of left lateral prefrontal cortex (PFC) in control over language production by testing stroke survivors with damage to left PFC on a control-demanding task. We varied the type of interfering information in order to increase the difficulty of picture naming, as we explain next.

A fruitful paradigm to investigate control functions implemented during speaking is picture-word interference (Rosinski et al., 1975). In this task, speakers are instructed to name pictures as fast and accurately as possible, while ignoring a distractor word superimposed on the picture. The distractor word can be manipulated to provide information (partially) congruent with the picture name (e.g., the picture of a carrot with the distractor “castle”, which overlap in their initial phonemes) or incongruent (e.g., pictured carrot, distractor “stamp”). Control functions are assumed to aid the production of the picture name in the presence of incongruent, competing information from the distractor word. In particular, naming pictures in the presence of distractor words likely requires interference control, an executive function involved in “suppressing a stimulus that pulls for a competing response so as to carry out a primary response, […] suppressing distractors that might slow the primary response […]” (Nigg, 2000, p. 222; for further discussion, see also Friedman and Miyake, 2004).

The type of relationship that the picture name bears with the distractor is an important determinant of people’s performance. For example, naming a picture with an unrelated distractor word (e.g., the picture of a carrot with the distractor “stamp”) is slower than naming a picture with a series of Xs superimposed (e.g., Glaser and Düngelhoff, 1984). In the neutral condition (XXX), only the picture name is activated by the picture stimulus, whereas with unrelated distractor words, both distractor word and picture name activate their representations, which then compete during word planning. We refer to this effect as the lexical interference effect. Furthermore, if the picture and distractor are from the same semantic category (e.g., pictured carrot, distractor: “radish”), picture naming is slowed down relative to an unrelated distractor. This effect is known as the semantic interference effect (e.g., Glaser and Düngelhoff, 1984). In this case, the semantic relation between picture and distractor makes the distractor a stronger competitor for the picture name relative to an unrelated word (e.g., Roelofs, 1992, 2003). By contrast, if the picture and the distractor share a phonological relationship (e.g., pictured carrot, distractor: “castle”), picture naming is faster relative to an unrelated distractor. This effect is known as the phonological facilitation effect (e.g., Schriefers et al., 1990). The phonological overlapping information between the distractor and the picture name pre-activates the picture name, facilitating its production.

Importantly, lexical and semantic interference possibly originate at different processing stages in language production. In particular, the semantic interference effect has been attributed to selection difficulty arising at the lexical level (e.g., Roelofs, 1992). Differently, the lexical interference effect could potentially reflect the operation of an early attentional blocking mechanism, which is more challenged in the presence of distractor words compared to the neutral condition (XXX; e.g., Roelofs, 2003; Roelofs et al., 2011). Whether the same brain region underlies the control mechanisms engaged in resolving these interference effects is unclear.

The study protocol was approved by the Institutional Review Board of the Northern California Health Care System, following the Declaration of Helsinki. Participants gave written informed consent and were paid for taking part in the study.

The semantic interference and phonological facilitation effects were calculated relative to the unrelated condition (i.e., semantic interference: related—unrelated; phonological facilitation: phonological—unrelated), whereas the lexical interference effect was calculated as unrelated—XXX. In this way, interference is indicated by positive values and facilitation by negative values.

Figure 2 (left) shows a box-and-whisker plot of the participants’ median RTs. The right panel of Figure 2 shows how many SDs each patient’s performance is relative to the mean of the control participants for each effect of interest. Dashed lines indicate 2 SDs. It is clear from the figure that, for the lexical interference effect, all patients scored more than 2 SDs above the mean of the controls, suggesting relative impaired performance consistent across all patients. For the semantic interference effect, however, only two patients scored beyond 2 SDs, one above the mean and one below the mean of the control participants. Finally, for the phonological facilitation effect, five of the six patients scored more than 2 SDs below the mean of the control participants, suggesting that phonological distractors convey an additional benefit for the patients in picture naming relative to controls.

The results of the Bayesian estimation are summarized in Figures 3, 4. For each effect, a histogram is shown of 100,000 credible parameter values from the posterior distribution, given the data. The mean of the parameter values for each distribution is indicated on top. The 95% HDI is indicated for each histogram as a thick horizontal line below the distribution.

For the group effect (patients vs. controls), shown in Figure 3, the 95% HDI is clearly greater than zero (range: 337–438) and in fact, 100% of the credible parameter values of the difference in their performance is greater than zero. Thus, patients’ overall picture naming performance is credibly slower than in controls.

The next comparisons of interest are the distractor effects, shown in Figure 4. Interference effects are represented by positive values whereas facilitation effects by negative values.

The distractor effects in the control group are shown in the top row of Figure 4. For the lexical interference effect (unrelated vs. XXX, left panel), the 95% HDI clearly does not include zero (range: 57–100), with 100% of the credible parameter values greater than zero. Similarly, for the semantic interference effect (related vs. unrelated, middle panel), the 95% HDI is greater than zero (range: 12–57), with 100% of the credible parameter values greater than zero. Finally, for the phonological facilitation effect (phonological vs. unrelated, right panel), the 95% HDI does not include zero (range: −48 to −3), with 98% of the credible values smaller than zero. Thus, the lexical interference, semantic interference, and phonological facilitation effects are all credibly present in the data.

The distractor effects of the PFC patients are shown in the middle row of Figure 4. For the lexical interference effect (left panel), the 95% HDI clearly does not include zero (range: 227–349), with 100% of the credible parameter values greater than zero. By contrast, the semantic interference effect (middle panel) is not credibly present in the data: the 95% HDI includes zero (range: −143 to 108) with the credible parameter values almost symmetrically distributed below and above zero. Finally, for the phonological facilitation effect (right panel), the 95% HDI clearly does not include zero (range: −178 to −83), with 100% of the credible values smaller than zero. In summary, whereas the age-matched controls credibly showed all three classical distractor effects (lexical and semantic interference, and phonological facilitation), the PFC patients only showed credible lexical interference and phonological facilitation.

The comparisons of effect magnitude between patients and controls are shown in the bottom row of Figure 4. Decreased performance (i.e., larger interference and smaller facilitation effects) in patients relative to controls are represented by values smaller than zero. Increased performance (i.e., smaller interference and larger facilitation effects) are represented by values greater than zero. The lexical interference effect (left panel) is credibly larger in the patient group than in controls, with a 95% HDI between −273 and −143. In fact, 100% of the credible parameter values are smaller than zero. By contrast, the semantic interference effect is not credibly larger in patients than in controls, with the 95% HDI including zero (range: −77 to 178). Finally, patients have a credibly larger phonological facilitation effect than controls, with the 95% HDI between 52 and 157, and 100% of the credible values greater than zero.

The results of the ANOVA are presented in Table 2. These results largely converge with the results of the Bayesian analysis reported above. The only difference is the phonological facilitation effect in the controls, which was not significant in the ANOVA, but credible in the Bayesian analysis. However, with the Bayesian estimation we provide not only information on whether an effect is credible or not, but also a range of values for these effects that are most credible given the data.

This study is the first to examine picture-word interference in a group of patients with well-characterized left lateral PFC lesions, maximally overlapping in the ventrolateral PFC.

Patients were overall slower in picture naming than controls (for a similar finding in aphasic patients, see e.g., Biegler et al., 2008; Hashimoto and Thompson, 2010). Importantly, picture naming with a distractor word caused more interference for patients than for controls relative to a non-linguistic distractor (XXX). Furthermore, patients showed more phonological facilitation than controls. Finally, on a group level, patients failed to show a reliable semantic interference effect. These findings were evident from both the Bayesian estimation results and from the standardized scores of the effects for each patient. Below, we discuss each of these effects in more detail.

A novel finding from the present study was the increased lexical interference effect for patients relative to controls. In fact, all six patients were consistently impaired by distractor words relative to neutral, non-linguistic distractors. It has been argued that interference control is a significant cognitive construct, involved in multiple tasks and related to other cognitive abilities (Unsworth, 2010). However, its functional neuroanatomy in relation to word production has remained unclear. Previous neuropsychological studies using verbal tasks have often used the color-word Stroop task (e.g., Tsuchida and Fellows, 2013; Geddes et al., 2014), in which participants name the ink color of color names (e.g., red printed in red ink or blue printed in red ink). However, this task requires multiple repetitions of a few words (typically, 3–5 color names) and as such, is not ideal for investigating word production processes. Evidence that word planning depends on attentional control is now substantial (for review, see Roelofs and Piai, 2011) and recently, it has also been shown that interference control (also termed selective inhibition) plays an important role in picture-word interference (e.g., Shao et al., 2015). Our finding that competing linguistic information is especially problematic for all our patients with lateral PFC damage not only strengthens the previous relation between interference control and word production, but also further specifies that this type of control necessarily implicates the lateral PFC. This finding is in line with the neuropsychological literature suggesting a critical role of the left lateral PFC (and the ventrolateral PFC in particular) for interference control in verbal tasks (Tsuchida and Fellows, 2013; Geddes et al., 2014).

The semantic interference effect in the patients was highly variable. Descriptively, three patients had longer RTs for the related than the unrelated condition (i.e., semantic interference, patients 3, 4, and 5, with only one patient showing interference 2 SDs above the mean of the controls) and three patients showed the opposite effect (i.e., semantic facilitation, patients 1, 2, and 6, with one patient showing facilitation 2 SDs below the mean of the controls).

Facilitation from semantically related distractors is not a common finding in the literature. Wilshire et al. (2007) reported a case study of an anomic patient showing semantic facilitation with simultaneous presentation of picture and distractor (i.e., 0 ms stimulus onset asynchrony). However, given that all our six patients were not aphasic, comparisons with Wilshire et al.’s patient remain difficult. Other studies reporting semantic facilitation in picture-word interference had additional manipulations such as brief distractor pre-exposure, distractor masking, or low co-activation of distractors, which decreased the strength of the distractor as a competitor (e.g., Roelofs, 1992, 1993; Piai et al., 2012). It could be the case that, for the patients experiencing semantic facilitation, the dynamics of processing the distractor vs. the picture is altered for some reason, and the distractor is able to prime the picture name but not compete with it (for further discussion, see Piai et al., 2012). Additionally, it could be that the amount of ventrolateral PFC damage or the specific portion of the ventrolateral PFC damaged are critical in determining if and how much control processes will be impaired in the presence of semantic competitors. Rather than being a functionally uniform region, it has been argued that different portions of the ventrolateral PFC subserve different cognitive functions (Clos et al., 2013). Also, it could be the case that a larger lesion is more likely to encompass portions of the ventrolateral PFC that are critical to resolving competition among semantic alternatives. Variations in the extent of ventrolateral PFC affected could potentially explain the variable behavior of our patients regarding semantic effects. However, assessing these hypotheses directly is not possible in the present study. In Figure 6, we show the lesion overlap of the three patients showing facilitation vs. the three patients showing interference, as well as the overlap in lesion location between the two groups. As can be seen, although the three patients showing semantic interference have lesions maximally overlapping in the ventrolateral PFC, the group showing facilitation does not have a converging lesion maximum. Moreover, there seems to be great overlap in lesion location between the two groups. Thus, the extent to which left lateral PFC lesions affect the semantic interference effect in picture-word interference remains unclear at this point, but will hopefully be clarified in future investigations.

Using Bayesian estimation, we established that a zero difference in semantic interference in the patient group, on the group level, is a credible value given the data². Thus, we can conclude that our patient group was not consistently impaired by semantic competitors. However, all six patients showed an increased lexical interference effect relative to controls. The dissociation between lexical and semantic interference in the patients’ behavior may indicate different types of control functions necessary for resolving lexical vs. semantic interference, which are differentially dependent on the lateral PFC. In the introduction, we related performance in picture-word interference to a control function enabling participants to suppress distractors that might slow down picture naming. However, the reason why distractors may hinder picture naming can be diverse. For example, in the picture-word interference literature, distractor effects exist that are caused by the distractor words themselves, independent of their semantic relation to the picture. The distractor-frequency effect is such an effect, in which pictures paired with low-frequency distractors are named more slowly than with high-frequency distractors (Miozzo and Caramazza, 2003). It has been shown that the semantic interference and distractor-frequency effects reflect different underlying operations (Scaltritti et al., 2015; for a computational model formalizing the parameters associated with these effects, see also Roelofs, 2003). The semantic interference effect is thought to emerge due to competition between the picture name and the distractor word during lexical selection (e.g., Roelofs, 1992; La Heij et al., 2006; Rahman and Aristei, 2010). By contrast, the distractor-frequency effect is explained by a reactive blocking mechanism that blocks the distractor word so that processing of the target picture is favored (Roelofs, 2003). The speed with which the distractor can be processed determines the speed of blocking (e.g., Roelofs et al., 2011). The blocking mechanism is an early attentional mechanism that operates independently from resolving semantic competition (e.g., Roelofs, 2003).

The lexical interference effect conceivably measures the engagement of this early attentional, blocking mechanism. The mechanism is always necessary in conditions with a distractor word, such as the unrelated condition, so that processing the picture is favored over distractor processing. In the neutral condition, no blocking is necessary because no distractor word is presented. So the lexical interference effect maximally captures the workings of the blocking mechanism. The semantic interference effect is not informative about the involvement of early attentional blocking demands, because the semantic and unrelated conditions both require blocking of the distractor word.

In our study, resolving lexical interference was impaired in all six left PFC patients whereas resolving semantic interference was not. This dissociation implies that the left PFC is critically involved in attentional distractor blocking so that picture processing is prioritized despite distracting information. But the left PFC is not necessarily required for resolving semantic competition between picture name and distractor word during response selection. This latter proposal is in line with recent evidence that the left lateral PFC is not always necessary to resolve interference from semantically related competitors (Riès et al., 2015). In this recent study comparing the performance of left PFC patients in two production tasks inducing semantic interference, Riès et al. (2015) found that left PFC patients had a larger semantic interference effect than controls in blocked-cyclic naming, a picture-naming task that requires proactive interference control due to blocked design and picture repetition. By contrast, in a continuous picture naming task in which pictures are not repeated (Howard et al., 2006), the same patients and controls had an interference effect of similar magnitude.

An additional finding of our study was that the phonological facilitation effect was larger in the patients than in the control participants, similar to the findings of Hashimoto and Thompson (2010) in their aphasic patients. It is well known that patients with word finding difficulties benefit from phonemic cues (e.g., Li and Williams, 1989). Although our patients performed within normal limits on the WAB and thus, were not aphasic as the patients in Hashimoto and Thompson (2010), it is common for this patient group (i.e., within normal limits-WAB) to still have subclinical language impairment beyond what would be expected for their age (e.g., Riès et al., 2015). In fact, such a suggestion finds support in our own data. As reported above, patients were overall slower than the controls. Perhaps even more importantly, in the XXX condition, in which the distracting information is neutral, patients were still slower (250 ms on average) than the controls. This finding is consistent with residual word-finding difficulties these patients often report. It is thus possible to explain the hyper phonological facilitation in these patients in that they might have developed a greater reliance on phonemic cues in comparison to controls when having word-finding problems. However, this account remains speculative at this point. Crucially, all six patients showed an increased lexical interference effect, providing evidence that any damage to the left lateral PFC will affect the ability to carry out picture naming in the presence of competing linguistic information.

In conclusion, all six left PFC patients showed an increased lexical interference effect. The impaired performance of the patients in naming pictures in the presence of a competing distractor word (i.e., linguistic competing information) provides evidence that the left lateral PFC is a necessary structure in providing control over word production processes. This finding corroborates the evidence that word production requires attentional control and further specifies that interference control in word production necessarily implicates the lateral PFC.

Conceptualized and designed the experiment (DS); acquired the data (DS); analyzed the data (VP); wrote the article (VP, SKR, DS). All authors have approved the final version of the article and agree to be accountable for all aspects of this work.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.