In the time window covering the N1 response, which was prominent for the manipulated words, the main effect of Duration was significant. An interaction between Duration and Replacement refined the finding by showing that the N1 was of equal amplitude for words beginning with a plosive (mean amplitude −0.59 μV) or a short cough (−0.55 μV). However, long coughs elicited a larger N1 (mean −1.92 μV) compared to the normal words beginning with a fricative (mean −0.84 μV; Tukey–Kramer p<0.05). The responses to long coughs were also larger relative to short coughs (p<0.001) and to normal words beginning with a plosive (p<0.01). Thus, the largest N1 response was elicited by words beginning with a long cough. The mean N1 peak latency to coughs was 146.2 ms with no significant difference between the two cough durations. Since connected speech was used, the normal words did not elicit a sufficiently reliable N1 for the analysis. An interaction of Expectancy×Replacement showed that the responses to the manipulated highly expected words were more negative (mean −1.71 μV) compared to the corresponding less expected words (mean −0.76 μV; p<0.05) and to the normal highly expected words (mean −0.38 μV; p<0.05). The responses did not differ significantly from the normal less expected words (mean −1.06 μV, n.s.). The responses to normal words did not differ significantly between highly and less expected words, although they were more negative to less expected words than to those of high expectancy (mean amplitudes −1.06 and −0.38 μV, respectively).#

A significant main effect of Expectancy was found in the time window of 180–280 ms. The interaction of Expectancy×Replacement further indicated that the responses to the manipulated words were, on the average, positive to both less and highly expected words (means 0.66 and 0.56 μV, respectively) and did not differ significantly from each other. For the normal words, the responses to less expected words were significantly more negative compared to the responses to highly expected words (means −1.82 and −0.13 μV; p<0.05). The responses to both the less and highly expected manipulated words were more positive compared to responses to normal less expected words (p<0.001 for both comparisons), indicating a P2 response to the manipulated words. Finally, an overall interaction of Expectancy×Duration×Replacement indicated that the probability effect was larger for normal words which began with a plosive (difference less minus highly expected −2.02 μV; p<0.05) whereas the difference did not reach significance for words beginning with a fricative (−1.16 μV, n.s.). The difference between highly and less expected words was not significant for manipulated words of either cough length. In addition, the responses elicited by the manipulated highly expected words were significantly more negative for the short coughs compared to the long coughs (means −0.50 and 1.63 μV; p<0.01), since the short coughs did not elicit a clear P2 response. The responses to manipulated less expected words did not differ significantly between short (−0.001 μV) and long coughs (1.32 μV).#

In the time window of 280–380 ms, covering the earlier part of the N400, a significant main effect of Expectancy and an interaction of Expectancy×Replacement were found. The responses were more negative to less expected compared to highly expected words, and although less expected words elicited more negative responses in the case of both normal and manipulated words, the difference reached significance only in the case of the normal words (less minus highly expected −2.11 μV for normal words, p<0.01, and −0.23 μV for manipulated words, n.s.). The responses to normal less expected words were also more negative than to the manipulated less (p<0.05) or highly (p<0.01) expected words. Finally, the time window of 380–520 ms, covering the later part of the N400 effect for the normal words, also showed a significant main effect of Expectancy. An Expectancy×Duration interaction indicated that the effect was more pronounced for words beginning with a plosive or a short cough (difference less minus highly expected −1.48 μV; p<0.001) as compared to a fricative or a long cough (−0.53 μV, n.s.). Furthermore, a significant Duration×Replacement interaction indicated that the words with a short cough elicited more negative responses (−0.86 μV) compared to the words with a long cough (0.22 μV; p=0.057), whereas the responses to normal words did not differ significantly between plosive and fricative beginnings (−0.13 and −1.01 μV). The difference between words beginning with a short cough and words beginning with a plosive also approached significance (p=0.060), the short coughs eliciting more negative responses. Although the averaged responses were more negative to words beginning with a fricative (−1.01 μV) than to those with a long cough (0.22 μV), the difference did not reach significance (p=0.11). No significant effects were found in the last time window of 520–620 ms.#

In order to explore whether an N400 was elicited by the manipulation of the highly expected words, sentences with normal and manipulated highly expected words were compared in a separate analysis-of-variance (ANOVA) with factors Duration and Replacement (see Fig. 4B). A significant interaction of Duration×Replacement was found in two time windows, namely in 180–280 ms and 380–520 ms (F1,25=16.87, p<0.001, and F1,25=7.51, p<0.05). In the first time window, the result showed that more negative responses were elicited for normal fricative words compared to words with a long cough (difference −1.89 μV; p<0.01), indicating a P2 response to the manipulated words. The responses to the short cough words did not significantly differ from normal plosive words, although they were slightly more negative (difference −0.49 μV, n.s.). The responses were more negative to words with a short cough compared to a long cough (difference −2.13 μV; p<0.01). In the later N400 time window (380–520 ms) only two comparisons approached significance in post hoc tests. The responses for the normal fricative words were more negative compared to plosive words (means −0.49 and 0.90 μV; p=0.058). There was not a difference between words beginning with a long or short cough. The words with a short cough elicited slightly more negative responses compared to the normal plosive words, however, the difference was not significant (means −0.42 and 0.90 μV; p=0.086).#

Fig. 5 presents the topographical distribution of the early and late N400 effects in the four time windows for normal and manipulated words. For normal words beginning with a plosive (upper row) the negative effect (blue) is pronounced in posterior areas and begins in the time window of 180–280 ms, whereas for fricatives (second row) it is weaker and appears later. For manipulated words, the words originally beginning with a plosive and replaced with a short cough (third row) have a posterior negative distribution beginning in the time window of 280–380 ms. The effect is later than that for normal words, and further, it has two negative maxima. The topographical distribution did not seem to noticeably differ between the early and later effects. Only the amplitude was increased in the later time window.#

For normal words, as expected, no clear N1 was elicited (see Fig. 2). In connected speech, N1 response to word onsets is usually not elicited (Friederici et al. 1993; Hahne and Friederici 2001) because the words in fluent speech rarely have breaks between them. The manipulated words elicited a clearer N1 response, reflecting an automatic detection of the cough by the auditory system. This finding, together with the other results, supports the view that the phenomenon of phonemic restoration is a consequence of higher level semantic processing (Samuel, 2001). In Warren's experiments (e.g., Warren, 1970), the restoration may have occurred during the time delay between the perception of the cough and the behavioral response, i.e., indicating the cough's location in a written form of the sentence after hearing it. As the cough in Warren's experiments occurred in the middle of a sentence, it was followed by both a longer auditory input as well as reading of the sentence, which introduced a considerable time delay. As the auditory sensory memory for the stimulus decays rapidly, it is likely that the detection of the cough's location afterwards becomes more difficult.#

The longer coughs (145–200 ms) elicited a more prominent N1-P2 response complex whereas the N1 response to short coughs (80 ms) did not differ significantly from the responses to the normal words beginning with a short phoneme (plosive). Thus, both the N1 amplitude and the behavioral results indicated a greater salience of the long coughs and more severe disruption of the semantic processing. The N1 amplitude has been shown to vary as a function of physical properties of the stimulus, such as intensity and the slope of the energy change (Picton et al. 1970; Elberling et al. 1981; Bak et al. 1985; Kodera et al. 1979; Loveless and Brunia 1990). In the present experiment, there was no striking difference in the onset slope between the different cough lengths (see Fig. 6 for the average cough amplitudes). Thus, it is unlikely that the N1 amplitude had varied due to different onset slopes. If that had been the case, a more prominent N1 would have been elicited by the slightly sharper onset of the short coughs. Instead, another factor may have increased the N1 to long coughs, namely the temporal integration of stimulus energy with increasing duration. Temporal integration results in greater perceived loudness for a longer tone compared to a shorter tone with the same energy. The detection of tones improves with increasing duration up to 250–500 ms: in other words, short tones need greater power than long tones in order to be detected (Moore, 2003). The N1 amplitude has been found to increase with increasing stimulus duration up to 50 ms for sinusoidal tones (Onishi and Davis, 1968). Accordingly, the N1 amplitude of the present study may reflect the greater perceived loudness of longer coughs compared to short ones. In addition, when the average power was calculated for the first 50 ms of the coughs, the relative intensity was larger for long coughs (984 arbitrary units, a.u.) compared to short coughs (736 a.u.), indicating an intensity difference between the cough onsets.#

Finally, for the manipulated words, the N1 was larger for highly expected words compared to those of less expectancy. This result was not hypothesized since by the latency of the N1, there is not much information about the upcoming word fragment.¹1The N1 peaked at about 57 ms after the offset of the short coughs and 0–54 ms after the long coughs. Notice that the mean latency of N1 was measured from the grand averages over all long coughs, since averaging over separate cough durations was not reasonable due to a small number of certain durations. Therefore, the relative N1 latency varies according to the cough duration. The latencies compared to the two most frequently occurring cough duration (180 ms, occurred 62 times, and 150 ms, occurred 14 times) were 34 and 4 ms, respectively. In principle, an immediate coarticulation at the penultimate word could hint at an unexpected final word even in the case when the onset phoneme was obliterated. In that case, however, an opposite result would have been expected, i.e., more negative responses to less expected words (Praamsta et al., 1994). The sentences were recorded in their presentation form, i.e., the sentences were pronounced as such and not produced by splitting the less expected final word from another sentence. Therefore, it might be that other systematic differences between sentences with a highly or less expected final word, which were not covered by the parameters estimated so far, may still be present. Therefore, the explanation for the larger N1 to manipulated highly expected words remains open.#

In the case of normal words, the responses to less expected words were more negative compared to highly expected words between 180 and 520 ms in total. This expectancy effect was influenced by the type of the beginning phoneme. The effect was present as early as the time window of 180–280 ms for words beginning with a plosive, compared to those beginning with a fricative or all manipulated words, indicating an earlier onset of the negative response. In the next time window (280–380 ms), the effect was found for normal words with both phoneme types. As the duration of plosives is shorter, the next phonemes follow earlier and word recognition can be faster, resulting in earlier detection of the less expected word. For reaction times, the difference between phoneme types was around 100 ms, corresponding to the difference in phoneme durations. Further, it is possible that the more variable durations of fricatives have caused variability in the latency of the effect. This, in turn, would reduce its amplitude in the grand averaged responses. So far, most of the studies of the auditory N400 have used target words beginning with variable phonemes. Only one study (Van den Brink et al., 2001) reported to have used sentence final words beginning exclusively with a plosive. The authors also found an early negativity (N200) for incongruent compared to congruent words between 150 and 250 ms. They interpreted the response to reflect the assessment of the semantic features that are expected by the contextual specifications, thus, semantic processing based on partial information about the upcoming word. The present results showed that the duration of the onset phoneme has an effect on the early phase of the negative effect.#

The early negativity in the present study differs from the Connolly's PMN response in response latency and topography. The response for less expected plosive words was earlier than the PMN which occurred in the time range of 250–350 ms, with a mean latency of 275 ms (Connolly and Phillips, 1994). The later latency can partly be due to the fact that Connolly and his colleagues employed peak analysis. However, the assessment of peaks in individual responses can be rather arbitrary when no clear single peak can be defined. We, therefore, calculated the mean amplitude within different time windows. Additionally, since the final words in Connolly's studies were not restricted to begin with a certain phoneme type, it may have caused a large variability in the response amplitudes and latencies. Their response latency corresponds to our second time window, where the effect was elicited also by the fricative words. Finally, unlike the PMN response that was largest in right frontal electrodes (Connolly et al., 2001), the early effect observed here had a central posterior distribution (see Fig. 5).#

These differences support the argument that the early negative effect reflects semantic processing (N200) rather than phonological processing (PMN). If the early effect was only due to the unexpected phoneme, we would not expect a difference between fricatives and plosives because they must be differentiated by the duration of a plosive. Instead, if the early effect reflected semantic processing, more phonetic information than just the first phoneme would be needed. Given the shorter duration of the plosives compared to the fricatives, additional phonemes can come into the system earlier after plosives and therefore, semantic processing can begin earlier. Moreover, as the later N400 effect was also largest centro-parietally (as in Van den Brink et al., 2001), the topographical distribution did not differ noticeably between the early and later effects, only the amplitude increased. The similar topography suggests that it may reflect an early phase of the same (semantic) processing. Thus, as the early effect was modified by the duration of the onset phoneme and no difference was found in the topographical distribution compared to the later N400, we conclude that the response reflects early semantic processing (e.g., lexical selection as suggested by Van den Brink et al., 2001). The expectancy for the possible final word created by the sentence context was thus violated by the detection of the non-expected word, based on the word onset.#

The recognition of the less expected words was hypothesized to be more difficult when the word's onset was obliterated. In addition to an absent initial phoneme, the following word fragment did not fulfill the expectations created by the preceding sentence context. This hypothesis was confirmed by the delayed behavioral responses. In the EEG experiment, the manipulated words elicited the N400 effect, indicating increased integration demands with the context for less expected words compared to corresponding highly expected words. This, in turn, indicates that the words could be recognized in spite of the manipulated onset. However, the amplitude of the N400 did not increase for the manipulated words as hypothesized. Instead, the effect was elicited in a later time window. For the manipulated words the expectancy effect was present only in the time window of 380–520 ms, whereas for normal words the effect had its maxima in two earlier time windows (between 180–380 ms). The later onset of the N400 indicates that the word could only be recognized after some phonetic information following the cough had been presented. If the N400 amplitude had linearly reflected difficulties in semantic processing, it would have been larger to manipulated words compared to normal words, and could have been largest when the word's onset was replaced with a long cough. This was not the case. Actually, for words beginning with a long cough, as compared to a fricative, the effect was the reverse: responses to fricatives were more negative than they were to long coughs, although not significantly. The responses to the words beginning with a short cough tended to be more negative compared to normal plosive words, but mainly because the larger and earlier N400 for normal words was already decaying in this later time window.#

Considering the highly expected words, when their initial phoneme was replaced by a cough, the delayed behavioral responses indicated more difficult recognition of these words. This difficulty was not reflected as a significant enhancement of the N400 amplitude compared to normal words. The highly restrictive sentence context seemed to point to the expected final word candidate so efficiently that the word could be accessed without remarkably increased effort even though the phonetic information of the word's onset was eliminated. However, a tendency of an increased N400 was found for the highly expected words beginning with a short cough compared to the normal words beginning with a plosive. No such tendency was found for words beginning with a long cough compared to a fricative (see Fig. 4b). Of course, one could argue that the N400 was not enhanced (at least not for the long coughs) because no word recognition was possible any more. The behavioral results contradict this interpretation, as it was possible to repeat the highly as well as less expected manipulated words. For the degraded highly expected words the mean repetition time was even faster compared to normal but less expected words. If the beginning of the word neither confirms nor refutes the expectation created by the context, the remaining fragment of the word seems to be sufficient to complete the integration of the expected word with relative ease. Thus, both the behavioral and the electrophysiological results suggest easier semantic processing of the highly expected words, even when the initial phoneme is obliterated, compared to less expected words.#

As the N400 amplitude is considered to reflect either lexical access, or relative difficulty of the semantic integration, the non-existing N400 for the manipulated highly expected words (or small in the case of short coughs) and the non-enhanced N400 for the manipulated less expected words can be interpreted as follows. Concerning lexical access, some word recognition models (e.g., Marslen-Wilson, 1987) suggest that lexical access is initiated by the word's onset. A cough in the word's beginning cannot initiate activation of any lexical candidates because no word begins with such a sound. Therefore, there should be no activation related to the initial lexical processing if that was solely based on bottom-up information of the speech signal and no N400 should be elicited if it reflects lexical access only. Since in our study successful word recognition took place even when the word's onset was manipulated, the semantic processing cannot be solely based on bottom-up analysis. Also, concerning selection, competition is not possible between different lexical candidates having an identical onset, as no phonetic information was present, and again, no N400 would be expected.²2These considerations can also explain the smaller N400 effect to manipulated words compared to normal words in the early time windows. As word access or selection was not possible when the word's onset was obliterated, no early N400 (N200) was elicited. Finally, concerning contextual integration, the online processing of the context may occur in parallel to the analysis of the current auditory speech input and the integration may occur after the analysis. If the speech input is interrupted by a cough, no immediate lexical activation takes place. Lexical reconstruction must take place on the basis of the available speech input prior to and following the cough. The context analysis, however, can be continued and possibly completed during processing of the cough, supporting activation of the most expected word. When the remaining word fraction follows and fulfills this expectation, no N400 is elicited. In the case of short coughs, however, the time might have been too short for complete contextual analysis, and therefore, a small N400 would appear for highly expected word fragments. Thus, it seems that the processing of contextual information was not completed by the end of the short coughs. No effect was elicited after long coughs, suggesting that the context processing is more advanced given the additional time.#

In normal semantic processing the N400 amplitude reflects increased demands of word recognition or semantic integration. Processing of the semantic properties of the word and information about the preceding linguistic context may contribute differently to the overall N400 response (Federmeier and Kutas 1999; Kutas and Federmeier 2000). The N200 and N400 responses have been suggested to reflect separate processes: the earlier N200 recognition of the word, and the N400 integration into the context (Hagoort and Brown, 2000). When the final word's onset is replaced, the semantic processing of words may differ from the normal. The non-enhanced N400 suggests that the underlying processes of word recognition may differ in latency or distribution of activation, both influencing the N400 amplitude. For the manipulated words, there is no difference between the less and highly expected words during the cough replacement. Thus, no information is initially available to indicate that the upcoming word will not be the expected one, and consequently, the manipulation affected the early expectancy effect. Later, the N400 effect was nevertheless elicited, although it was delayed. The delayed N400 may reflect either delayed word recognition, which takes place on the basis of the remaining word fragment, or difficulties in the integration process. As the effect is not enhanced and word recognition is still effective, it may be that the influence of context takes place during the cough processing, as discussed earlier. In the case of short coughs (80 ms), the N400 onset was approximately 220 ms after the beginning of the word fragment. Van den Brink et al. (2001) have suggested that the sentence context has an effect on word recognition at 220 ms, and possibly already at 140 ms after the onset of words which begin with plosives. The present results suggest similar timing in the processing of the remaining word fragment. An obliterated onset does not completely hinder the word recognition even when the following input does not match the expectations created by the context. The remaining fragment seems to provide sufficient phonetic information for the word recognition. In summary, the present findings can be interpreted according to a view that different sub-processes underlying the N400 were differentially activated by the manipulated words. The earlier part (N200), possibly reflecting the recognition of word or lexical access based on the initial phonetic information, was diminished. The later part (N400), indicating integration with the linguistic context, seemed to be completed in the same time window as for normal words. If the word beginning is obliterated, the listener may rely more on context based predictions in word recognition. As the N400 was not increased for the manipulated words, these predictions seem to offer considerable help for the processing of a word's meaning.#