Lexical Planning in Sentence Production Is Highly Incremental: Evidence from ERPs

Li-Ming Zhao; Yu-Fang Yang

doi:10.1371/journal.pone.0146359

Abstract

The scope of lexical planning, which means how far ahead speakers plan lexically before they start producing an utterance, is an important issue for research into speech production, but remains highly controversial. The present research investigated this issue using the semantic blocking effect, which refers to the widely observed effects that participants take longer to say aloud the names of items in pictures when the pictures in a block of trials in an experiment depict items that belong to the same semantic category than different categories. As this effect is often interpreted as a reflection of difficulty in lexical selection, the current study took the semantic blocking effect and its associated pattern of event-related brain potentials (ERPs) as a proxy to test whether lexical planning during sentence production extends beyond the first noun when a subject noun-phrase includes two nouns, such as “The chair and the boat are both red” and “The chair above the boat is red”. The results showed a semantic blocking effect both in onset latencies and in ERPs during the utterance of the first noun of these complex noun-phrases but not for the second noun. The indication, therefore, is that the lexical planning scope does not encompass this second noun-phrase. Indeed, the present findings are in line with accounts that propose radically incremental lexical planning, in which speakers plan ahead only one word at a time. This study also provides a highly novel example of using ERPs to examine the production of long utterances, and it is hoped the present demonstration of the effectiveness of this approach inspires further application of ERP techniques in this area of research.

Citation: Zhao L-M, Yang Y-F (2016) Lexical Planning in Sentence Production Is Highly Incremental: Evidence from ERPs. PLoS ONE 11(1): e0146359. https://doi.org/10.1371/journal.pone.0146359

Editor: Hanjun Liu, Sun Yat-sen University, CHINA

Received: May 13, 2015; Accepted: December 16, 2015; Published: January 5, 2016

Copyright: © 2016 Zhao, Yang. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited

Data Availability: The minimal data set underlying the findings in this manuscript are uploaded as supplemental files. More detailed data are available from the data access of Institute of Psychology, Chinese Academy of Sciences, contacted at yangyf@psych.ac.cn.

Funding: This research was supported by the National Natural Science Foundation of China (http://www.nsfc.gov.cn/). Grant number is 61433018. YY received this funding. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Linguistic planning refers to the necessary retrieval of words and syntactic building in preparation for producing fluent utterances. Most models of language production assume that the linguistic planning of an utterance is processed incrementally [1–3]. One implication is that utterance articulation can be initiated before all of the constituent words and the entirety of its structure are planned. Based on this incremental hypothesis, a long-standing issue concerns how far ahead speakers plan before they start producing an utterance. This issue is related to a number of important central questions in psycholinguistic studies of language production, such as the horizontal information flow (how different processing levels interact) in spoken sentence production [4–5].

Most psycholinguistic models of language production distinguished three major processing levels involved in conceptualizing, formulating, and ultimately articulating an utterance [6–7]. Additionally, Garrett [6] and Levelt [7] hypothesize that the formulating process consists of two distinct processing steps: grammatical encoding and phonological encoding, and lexical retrieval is considered as an important sub-component of grammatical encoding. Thus the issue of planning scope can be considered in the context of each of these different processing levels. However, the focus of the present study is primarily on the issue of lexical planning scope.

The lexical planning scope in speech production, that is how far ahead speakers plan lexically before they start producing an utterance, has been a particular focus of prior research in this area. It has primarily been investigated using classical paradigms for word production and, based on the findings from this research, various candidates for the scope of lexical planning have been proposed, including the first content word, functional phrase, or clause, among others.

Evidence that first noun phrase of an utterance forms the first unit of lexical planning scope in sentence production comes mostly from studies comparing the onset latencies of utterances with different structures that describe the relationship between two or more distinct objects. For example, Levelt and Maassen [8] asked participants to describe two moving objects on the screen, and found that onset latencies were longer for utterances that had a conjoined noun phrase as its subject (e.g., “The circle and the square move up”) than similar utterances that consisted of two conjoined sentences (e.g., “The circle moves up and the square moves up”). Using a similar task, Smith and Wheeldon [9] found longer onset latencies for sentences such as “The dog and the foot move above the kite” than sentences such as “The dog moves above the foot and the kite”. Existing evidence suggested that the phonological planning unit is limited to the first phonological word [10–12]. Consequently, the effects observed in these studies were unlikely to reflect this process. These researchers therefore claimed that conceptual and lexical planning scope were responsible for these effects, and that the planning scope for these component processes encompass the subject noun phrase rather than the whole sentence. However, this research was unable to establish if whole subject noun phrase or simply the head noun of the subject phrase served as the lexical planning scope, because in these studies only the head nouns were included in the subject noun phrase [9]. To clarify this question, Allum and Wheeldon [13] asked participants to describe two vertically presented pictures according to the color of the pictures, and compared the onset latencies for two types of sentences, one with a head noun modified by a prepositional phrase (PP) as the subject, such as “the dog above the flower is red” (henceforth “PP utterances”), and another with a conjoined noun phrase (CNP) as the subject, such as “the dog and the flower are both red” (henceforth “CNP utterances”). They found faster onset latencies for PP than CNP utterances, and took these findings to show that the conceptual and lexical planning scope encompass only the head nouns of the subject phrases.

Allum and Wheeldon [13] further tested the latency effects of CNP and PP utterances in Japanese, a head-final language, in which the modifier phrase of the PP utterances is produced before the head noun. Again, longer onset latencies were observed for the CNP utterances than for the PP utterances. As the initial phrase of PP utterances is no longer the head noun, they concluded that the conceptual and lexical planning scope is not the initial head noun or the subject noun phrase, and suggested that it instead encompasses a functional phrase. According to Allum and Wheeldon [13], the functional phrase represents a unit in the thematic representation of an utterance, but not necessarily the argument of the verb or the head of a verb argument phrase (p. 792). The subject noun phrase in a CNP utterance serves as the agent, while the subject noun-phrase of a PP utterance consists of two smaller functional phrases. Thus, the longer onset latencies for the CNP utterances could be attributed to the fact that its first functional phrase is longer.

This functional phrase hypothesis provides a novel definition of the lexical planning unit, but calls for further investigation of this issue. First, as utterance articulation can be initiated before all of the constituent words and the entirety of its structure are planned, the onset latency will include the time required for conceptual, lexical, and structural planning. Thus the evidence above coming from the latency effects of different utterance formats may not discriminate among these different planning scopes. Moreover, Wheeldon and colleagues [14] have suggested that the scope of lexical and structural planning are not always coincident, and that the latency effect of the initial phrase size may be attributed to structural planning only. However, some studies tried to solve this problem for the functional phrase hypothesis. For example, Zhao, Alario, and Yang [15] used a similar task to Allum and Wheeldon (2007) but asked participants to preview the pictures without a color cue before each stimulus was presented, which means that participants could retrieve both the names of the pictures but did not know which sentence structure would be used during the previewing period. The time required for the conceptual and lexical planning should therefore no longer influence onset latencies, and the latency effect between CNP and PP utterances was indeed absent in this preview condition. Thus the researchers confirmed that the latency effect previously observed should be attributed to conceptual and lexical planning rather than structural planning, but whether it is conceptual or lexical planning that drives the effect remains unclear. Allum and Wheeldon [16] also used a previewing task, in which the previewing effect for each sentence type was obtained by comparing performance when the bottom picture (the second noun) was previewed or not. The facilitation effect of previewing the bottom picture was found for CNP utterances only, suggesting that the second noun was planned in the CNP but not in the PP utterances. Nevertheless, it remains unclear whether this preview effect facilitated conceptual or lexical processing of the second noun in CNP but not PP utterances.

Another approach to testing lexical planning scope takes semantic context effects in lexical production as an index. Various candidates other than the first noun phrase have been suggested by this line of research [10, 17–18]. For instance, Griffin [10] asked participants to produce sentence with the structure “the N1 and the N2 are above the N3” in response to three pictured objects while manipulating the codability (the number of alternative names, such as ship, boat, or sailboat for a schooner) of each picture (N1, N2, and N3). Onset latencies were affected by the codability of the N1, but not N2 or N3. As codability influences lexical selection, it was concluded that speakers often incrementally select nouns (one by one) in fluent utterances. This claim is called the radically incremental hypothesis. With the similar logic, Zhao and Yang [19] tested the functional phrase hypothesis using a picture-word interference paradigm. In this task, two pictures were presented vertically with a distractor embedded in one of the pictures. Participants were asked to produce a sentence with the picture names like CNP and PP utterances while trying to ignore the distractor. A semantic interference effect was observed only for the first noun of both CNP and PP utterances. That is, onset latencies were longer when the distractor was semantically related (category coordinate) to the first noun than when they were unrelated. As the semantic interference effect is often interpreted as a reflection of difficulty in lexical selection during word production [20] (but see [21]), the results of Zhao and Yang [19] are generally taken to support the radically incremental hypothesis rather than the functional phrase hypothesis.

But to reconcile discrepancies in findings and conclusions, some researchers have suggested that the lexical planning scope may be flexible rather than fixed. The reason for this is that a number of factors could modulate the planning patterns, such as time pressure [22], cognitive load and variable utterance formats [23], and the availability of the words and syntactic structures [4, 14]. However, according to the previous evidence, a certain planning unit may be preferred in certain speaking situation or conditions, even though the lexical planning scope is flexible. Moreover, an appropriate planning scope is critical for speech fluency. The current study aimed to test whether the functional phrase could be a preferred unit of lexical planning as expected according to the conditions in the circumstance, using the semantic blocking effect as an index of lexical selection and with EEG recordings.

The semantic blocking effect is a stable effect of semantic context in the blocked-cyclic naming paradigm, in which each block consists of several repeated cycles of a small set of pictures. In the semantically homogeneous condition, all the items in the block represent the same semantic category (e.g. animals), and in the semantically heterogeneous (or mixed) condition each item is from a different semantic category. It has been observed that in the picture naming task the naming latencies in the homogeneous condition are significantly longer than in the heterogeneous condition [24–25]. This latency effect, named the semantic blocking effect, is interpreted as an effect of lexical selection by some researchers [26–27] (but see [28]), and frequently used to investigate the time-course and neuro-anatomical regions of word selection [29–31]. Indeed, semantic blocking is associated with an ERP component around 200ms post picture onset [30, 32].

Current experiments

In the current study the semantic blocking effect was used to test the lexical planning scope. Two experiments are reported that examined effects of semantic blocking for CNP and PP utterances consisting of two nouns. The purpose of Experiment 1 was to test whether the semantic blocking effect was stable in sentence production. In this experiment, the first noun of each utterance was manipulated to be either homogeneous or heterogeneous in each block of trials, while the second noun was always heterogeneous in all blocks. As the semantic blocking effect is widely observed in the picture naming experiments, and the first noun of each sentence is considered to be completely retrieved before speech onset, it was expected that a semantic blocking effect would be observed in speech onset latencies and corresponding ERP components. If so, the semantic blocking effect would be considered to be a stable effect during sentence production, and therefore could be used to test lexical planning scope.

The aim of Experiment 2 was to test whether the second noun was lexically planned before speech onset. In this experiment, the second noun of the utterances was manipulated so that is was semantically homogeneous or heterogeneous in each block of trials, while the first noun was heterogeneous in all blocks. According to the functional phrase hypothesis, a semantic blocking effect in speech onset latencies should be observed for CNP utterances, but not PPs. By comparisons, as the radically incremental hypothesis would predict no semantic blocking effect for either CNP or PP utterances; while if the lexical planning encompasses the whole subject noun phrase, there would be semantic blocking effect for both CNP and PP utterances.

So far neurophysiological technology has been used broadly in word naming tasks in which participants name a single object in a picture, but quite little in sentence production. One of the main problems is that ERP data are very sensitive to the physical movement that takes place during speech production. Thus ERP data are only valid in the planning period, which limits the questions concerning speech production that can be studied effectively using ERPs. However, this was not a major issue for the present research because we were principally interested in the processes that take place during lexical planning, prior to the onset of speech. ERP data can provide insights into the neural basis of this lexical planning. Additionally, findings by Janssen et al. [32] had suggested that ERPs may show a different pattern from that observed in behavioral, and so we considered it important to study ERPs in addition to behavioral data to gain a fuller understanding of the nature of lexical processing. Finally, because ERPs provide a fine-grained account of activity that takes place over time, they have the potential to enable us to differentiate components that are associated with lexical planning and structural planning. Indeed, when a long utterance is going to be produced, there are many processes involved in the planning period, including lexical and structural planning, and so the ERP record will reflect the operation of these various processes. Therefore, to separate effects of structural planning and lexical planning in the present experiment, each block used only one sentence structure, CNP or PP. This repetition of sentence structure was intentional to reduce effects of structure planning. This approach was quite different from that adopted in previous studies of the functional phrase hypothesis [8, 15–16], in which sentence structures were deliberately varied. Moreover, according to the classic design of semantic blocking effect, each picture is repeated several times in each block. This was also different from the approach of previous studies, in which each picture appeared once in each block. The deliberate repetitions of structures and lexical items in the present study might well encourage speakers to adopt larger lexical planning scope [4, 14]. Thus it would not be surprising if we observed that the functional phrase or even larger unit was the lexical planning scope. However, if there was no semantic blocking effect on the second noun of both CNP and PP utterances, that will provide strong evidence for the claims of radically incremental hypothesis [10], and against the claim that the functional phrase as the preferred lexical planning unit.

Experiment 1

As mentioned in the Introduction, the semantic blocking effect in onset latencies of single picture naming is consistently observed in the literature, and interpreted as a reflection of lexical selection. It is reasonable to use this effect as a proxy to test the lexical planning scope. But before that we need to know whether the semantic blocking effect can be extended to a sentence production task. The aim of this experiment was to clarify this issue.

As the first noun of each sentence is considered to be completely retrieved before speech onset, semantic blocking effect in the RTs (for speech onset) and ERP components would be expected when comparing the homogeneous and heterogeneous conditions for the first noun. To reduce the influence of structural effects, the sentence structure was kept unvaried in each block. In addition, to ensure there was no semantic effect due to the retrieval of the second noun, this was presented in heterogeneous conditions in all blocks.

Method

Participants.

Twenty-four undergraduate and graduate students participated in the experiment. They were all native Chinese speakers with normal or corrected-to-normal vision, and were paid for their participation. This study was approved by the Institutional Review Board of the Institute of Psychology at the Chinese Academy of Science. All participants provided written informed consents.

Materials.

Eighteen pictures were selected from the picture pool of Snodgrass and Vanderwart [33]. They were divided into 6 semantic categories, each consisting of 3 items (see S1 Appendix). Three categories (zoo animals, fruits, and furniture) served as the first noun in the produced utterances, and were presented always at the top position (called top groups); while the other three categories (transports, musical instruments, and body parts) were used as the second noun, and always presented at the bottom position (called the bottom groups). For the top group, items from the same category were combined to form three homogeneous sets; while items from three categories respectively were combined to form three heterogeneous sets. For the bottom group, items from different categories were combined to form three heterogeneous sets, and then recombined to form another three heterogeneous sets. Each of the six heterogeneous sets for bottom groups was paired with one of the top group to form six blocks (see S1 Appendix for a detailed arrangement). Each block was produced in CNP and PP utterances respectively, forming 12 blocks totally. In the blocks of CNP utterances, both pictures in each trial were presented in red lines; while in the blocks of PP utterances, the top picture was in white lines and the bottom picture was in red lines. Six filler items were presented for practice, and they were not from the experimental categories. There were two practice blocks, one in CNP utterances and the other in PPs.

Design.

Semantic context for the top groups (homogeneous vs. heterogeneous) and sentence type (CNP vs. PP) were manipulated as within-participant factors. There were twelve experimental blocks (three homogeneous blocks in CNP, and three in PP; three heterogeneous blocks in CNP, and three in PP). The homogeneous and heterogeneous blocks were presented in alternative orders, and the CNP and PP blocks were presented in an AABB design. For the first three trials in each block, each of the six target pictures (three pictures from the top group paired with three pictures from the bottom group) was presented once. For the rest trials in the block, each of the target pictures was presented six times, each top item paired with each bottom item twice, in a pseudorandom order, so that the same picture never appeared in consecutive trials.

Procedure.

Participants were seated in front of the computer screen, at a distance of about 60 cm, and tested individually. A fixation point appeared on the screen for 1,000 ms. Then the pair of pictures appeared for 4,000 ms. Participants were asked to name pictures with the required syntactic structures as accurately and quickly as possible. There was a blank interval of 2,000 ms between consecutive trials. Before starting each block, the six pictures used in this block were presented successively in a random order companied with their names, for participants to get familiar with them. Participants were asked to produce one sentence type (CNP or PP) through the whole block. The screen’s background remained black. The whole testing session lasted about 40 min, and participants had a short break between blocks.

ERP recording and analyses.

The continuous EEG was recorded using an elastic cap equipped with 64 Ag/AgCl electrodes according to the International 10–20 system. The EEG data was amplified via AC amplifiers, at a sampling rate of 500 Hz (band pass filter = 0.05 Hz–100 Hz). The right mastoid electrode served as the online reference. An electrode placed between Fz and Cz served as the ground. The horizontal EOG was measured by placing two electrodes at the outer canthi, and the vertical EOG was measured with two electrodes placed above and below the left eye. All electrode impedances were kept below 5 KΩ during the experiment.

NeuroScan software 4.3 was used to preprocess the raw ERP data. Ocular artifacts were corrected by NeuroScan software [34]. The data were offline re-referenced using the algebraic average of two mastoids, and passed through 0.1–30Hz band (24dB/oct) filters. The data were segmented from 100ms before to 600ms after the stimulus onset (the onset of the picture pairs). A baseline correction was applied from 100 to 0 ms preceding stimulus onset. The artifact rejection criterion was ±100μV. The first occurrence of each picture (the first three trials) in each block was not included in further analyses. Finally, we excluded the ERP data of the trials on which the participants made production errors (see the types in details in the behavioral analysis), or the speech onset latencies were less than 600 ms. On average, 17% of all trials were rejected. Trials were then averaged in each condition for each participant, and this average was used for further statistical analysis. For the CNP utterances, there were 1121 out of 1296 (86.5%) trials in the homogeneous condition, and 1044 out of 1296 (80.6%) trials in the heterogeneous condition. For the PP utterances, there were 1075 out of 1296 (82.9%) trials in the homogeneous condition, and 1047 out of 1296 (80.8%) in the heterogeneous condition.

Results

Behavioral results.

Three types of responses were scored as production errors: (1) using unexpected content words, including picture names and the adjectives; (2) using incorrect syntax; and (3) fluency problems (repairing, stuttering, hesitation, and production of nonverbal sounds that triggered the voice key). Outliers were defined as latencies less than 300ms, more than 3000ms, or exceeding more than three standard deviations from a participant's mean latency. Latencies on error trials, outliers, and recording failures amounted to 4% of the data and were excluded from the latency analyses. As the semantic blocking effect is targeted, the first occurrence of each picture (the first three trials) in each block was also excluded. The mean onset latencies in each condition were shown in Fig 1.

Download:

Fig 1. The mean onset latencies in each condition in Experiments 1 and 2 (Session 1).

https://doi.org/10.1371/journal.pone.0146359.g001

Two separate analyses were carried out with participants (F₁ / t₁) and items (F₂ / t₂) as random factors. The factor of sentence type (CNP vs. PP) was analyzed within participants and between items. The factor of semantic context (homogeneous vs. heterogeneous) was analyzed within participants and items.

In the analyses of onset latencies, the main effect of semantic context was significant (F₁(1,23) = 7.30, MSE = 3,461, p < .05; F₂(1,16) = 14.07, MSE = 644, p < .01). The main effect of sentence type was significant for participants, but not for items (F₁(1,23) = 6.40, MSE = 1,246, p < .05; F₂(1,16) = 1.17, MSE = 2,832, p = .3). There was no interaction between semantic context and sentence type (F₁(1,23) = 2.10, MSE = 677, p = .16; F₂ < 1). The error rates were low (1% on average) and not analyzed further.

Electrophysiological results.

For ERP results, the overall repeated measures ANOVAs were conducted on mean amplitudes of consecutive 50 ms time windows, from the onset of the picture pair to 600 ms after picture onset. This analysis included Sentence Type (CNP vs. PP) and Semantic Context (homogeneous vs. heterogeneous) as within-participant factors, and other two factors relevant to clusters of electrodes. That is, Anteriority with three levels (anterior, central, and posterior) and Hemisphere with two levels (left and right). As a result, six regions were considered: Anterior-Left (FP1, AF3, F1, F3, F5, F7), Central-Left (FC5, FC3, C5, C3, CP5, CP3), Posterior-Left (P5, P3, P1, PO5, PO3, O1), Anterior-Right (FP2, AF4, F2, F4, F6, F8), Central-Right (FC4, FC6, C4, C6, CP4, CP6), and Posterior-Right (P2, P4, P6, PO4, PO6, O2). The results were summarized in Table 1.

Download:

Table 1. Time Course of the Main ERP Effects in Experiment 1 with scalp distribution in brackets.

https://doi.org/10.1371/journal.pone.0146359.t001

There was a main effect of Semantic Context in the window of 250–350 ms, with a larger negativity in the heterogeneous condition compared to the homogeneous condition, and with an anterior and central distribution (see Fig 2). The ERP effect of Sentence Type was presented in Fig 3, taking the CZ electrode for instance.

Download:

Fig 2. Semantic blocking effect of ERPs.

Average ERPs from 100 ms before up to 600 ms after the onset of the picture pair, for homogeneous and heterogeneous blocks in CNP and PP utterances respectively, at the 9 electrodes as an example in each area (anterior, central, and posterior multiply with left, central, and right). Positivity is plotted upwards to compare with the results of Janssen et al. [32]. The scalp distribution of the semantic blocking effect (homogeneous condition minus heterogeneous condition) is presented in the time window of 250–350ms.

https://doi.org/10.1371/journal.pone.0146359.g002

Download:

Fig 3. Sentence type effect of ERPs.

Average ERPs from 100 ms before up to 600 ms after the onset of the picture pair, for CNP utterances in heterogeneous blocks and PP utterances in heterogeneous blocks, at the CZ electrode as an example. Positivity is plotted upwards. The scalp distribution of the sentence type effect (CNP minus PP) is presented in the time window of 100–150ms and 300–400ms.

https://doi.org/10.1371/journal.pone.0146359.g003

Discussion

Experiment 1 tested whether the semantic blocking effect in RTs and ERPs observed in single word production could be extended to overt sentence production.

Firstly, a semantic blocking effect in onset latencies was observed for both CNP and PP utterances as expected. Secondly, the semantic blocking effect in ERPs was also observed for both CNP and PP utterances. The ERP component distributed at anterior and central electrodes, and in the time window of 250–350 ms, which is consistent with previous results [29, 32]. The result showed that the semantic blocking effect in onset latencies and ERPs observed in single word production can be extended to overt sentence production, although the utterances are much more complicated and the syntactic processing for sentence structures are involved in the speech planning. Moreover, the semantic blocking effect of ERPs showed a larger negative wave for the heterogeneous condition compared to the homogeneous condition, which is in line with the findings of Janssen et al. [32], but different from Aristei et al. [29] in the polarity of the semantic blocking effect.

The effect of sentence type was observed in onset latencies, and the ERP component distributed at anterior and central electrodes, pronounced in the time window of 100–150ms and 300–400ms. From visual inspection, the ERPs in the time window of 100–150ms revealed a P1-N1 complex that is typically associated with the presentation of visual stimuli. This may result from the different visual displays (cued by color) of these two types of utterances. It might be that the ERPs in the time window of 300–400ms reflected the syntactic processing of the sentence structures. However, the lexical planning scope was not tested yet in this experiment. According to the functional phrase hypothesis, the second noun is lexically selected before speech onset for the CNP utterances, but not for the PP utterances. Thus the ERP effect of Sentence Type in the time window of 300–400ms may also reflect the retrieval of the second noun. This is tested in the Experiment 2.

Experiment 2

The goal of Experiment 2 was to test whether the second noun in CNP and PP utterances was lexically selected before speech onset, that is, to test the functional phrase hypothesis on the lexical planning scope. If the second noun is lexically retrieved before speech onset for certain sentence types, the semantic blocking effect as in Experiment 1 should be observed for this type of sentences. To be specific, according to the functional phrase hypothesis, there would be semantic blocking effect in onset latencies and ERPs only for CNP utterances, similar to the one observed in Experiment 1. However, according to the radically incremental hypothesis, a semantic blocking effect in onset latencies should not be observed for either CNP or PP utterances. Moreover, if lexical planning encompasses the whole subject noun phrase, there should be a semantic blocking effect for both CNP and PP utterances.

So far the semantic blocking effect in ERPs was expected to coincide with that in onset latencies [29–30]. However, according to the interpretation of Janssen et al. [32], the ERP effect of semantic blocking reflects input processes rather than the lexical selection of the output processes. In light of this assumption the ERP component of any semantic blocking effect would not coincide necessarily with the semantic blocking effect in onset latencies. To be specific, if the corresponding ERP effect reflects input processes, and the latency difference between CNP and PP utterances results from input processes for the second noun, a semantic blocking effect in ERPs would be observed for CNP utterances only, even though the semantic blocking effect in onset latencies is absent for both CNP and PP utterances.