As already mentioned, Marslen-Wilson and Zwitserlood (1989) did not find inhibition but degraded activation if the target word BIJ (bee) was preceded by the real word woning (dwelling), rather than by the semantically related word honing.
They reasoned that if lateral inhibitory links existed between words, woning would have inhibited honing, which in turn would have inhibited the semantically related word bij.
Further evidence against lateral inhibition at the lexical level came from a study by Marslen-Wilson and Warren (1994). Participants did lexical decisions on cross-spliced words. In a kind of identity condition, the initial portion of the word had been excised from another token of the same word (e.g., jo(b) ‡ (jo)b, W1W1 sequence). In a related word condition the initial portion was taken from another existing word (e.g., jo(g) ‡ (jo)b, W2W1 sequence), and in a related pseudoword condition it was taken from a pseudoword (e.g., *jo(d) ‡ (jo)b, N3W1 sequence). Thus the latter two conditions (W2W1 (jo(g)b) and N3W1 (jo(d)b)) contained subcategorical mismatches (Streeter & Nigro, 1979;
Whalen, 1984, 1991), in that the formant transitions in the vowel provided misleading information about the following consonant. The authors reasoned that if there was inhibition between lexical candidates, then misleading information that activates another existing word (W2W1 activating jog) should cause inhibition of the target word (job) and thus slow down reaction time. On the contrary, misleading information that does not activate a lexical entry of its own (N3W1 from nonexistent *jod) cannot directly affect lexical processing via such an entry. It probably weakly activates both, jog and job, but this should slow down responses less than in the W2W1 case. Results showed that the mean lexical-decision latencies to the W2W1 and N3W1 sequences did not differ from one another, whereas both were significantly longer than responses to the W1W1 sequences. This result was replicated by McQueen, Norris, and Cutler (1999) and is in line with the results of Marslen-Wilson and Zwitserlood (1989) reported above. Marslen-Wilson and Warren interpreted
the absence of a difference between the W2W1 and N3W1 conditions as evidence against models that incorporate lexical competition via lateral inhibition. However, Dahan, Magnuson, Tanenhaus & Hogan (2001) tested the strings used by Marslen-Wilson and Warren (1994) in a visual world paradigm and showed that participants were slower to fixate the target picture when the onset of the target word came from a competitor word (W2W1) than from a pseudoword (N3W1), speaking in favour of inhibitory links between lexical entries.
Further support for the notion of inhibition is provided by Radeau, Morais and Dewier (1989) in an auditory word priming experiment with lexical decision.
They used primes that deviated in the later part of the word from the target, rather than in the onset. Responses to a target word (e.g. parure, ‘plaster’) preceded by an identical prime (parure) were facilitated as compared to responses to the same target preceded by an unrelated prime word (e.g. rouler – parure, ‘drive’).
However, if the prime was identical with the target up to the second syllable (e.g.
palais – parure, ‘palace’) or even only in the onset consonant (e.g. poulet – parure,
‘chicken’), responses were slower than in the control condition. The amount of inhibition did not differ between the two deviating conditions. If the primes were pseudowords rather than real words, a large overlap between prime and target (e.g.
*rotu – roti, ‘roast’) elicited facilitated word-responses while less overlap (e.g.
*romu – roti or *ramu – roti) produced no significant effect when compared to the control condition (e.g. *muna – roti). The finding of inhibition seems to be dependent on lexical status of the prime. In terms of lateral inhibition among word candidates this is conclusive because a non-existing item cannot inhibit another word candidate. While Marslen-Wilson and Zwitserlood (1989) – following the same logic – did not find different effects for word (woning – (honing) - BIJ) and pseudoword (*foning – (honing) - BIJ) primes in cross modal semantic priming, auditory phonological priming appears to be sensitive to lexical status and gives evidence for inhibition between competitors.
Also Soto-Faraco, Sebastian-Gallés and Cutler (2001) reported inhibition in a series of cross modal phonological priming experiments in Spanish. They used pairs of tri- or four-syllabic words that did not differ in the first two syllables except in the vowel (e.g. abandono – abundancia, ‘abandonment – abundance’) or consonant of the second syllable. Those with a consonantal difference were further divided into pairs with deviations in only one feature (e.g. papilla – patilla, ‘baby food – sideburn’) and into pairs with differences in more than one feature (e.g.
bofeton – boletin, ‘smack – bulletin’). As expected, when a visual target word was preceded by its own first two syllables (e.g. aban – ABANDONO) it was classified as an existing word faster than if it was preceded by an unrelated syllable (e.g. ce –
ABANDONO). If it was preceded by the first two syllalbles of the pair member (e.g. abun – ABANDONO) reaction times were even longer than in the control condition. This inhibition was found equally for vowels and consonants and within consonants equally for one or many feature deviations.
However, inhibition is not an obligatory finding in the cross modal fragment priming paradigm. Friedrich (2005) employed a design very similar to that of Soto-Faraco et al. (2001) in an electrophysiological cross-modal fragment priming design. A German word (e.g. KANTE, ‘edge’) was preceded by either its’
own first syllable (kan-) or another word’s first syllable with a different vowel (kon-, taken from Konto, ‘account’). Unrelated target words (e.g. ZUNGE,
‘tongue’) were used in the control condition. Reaction times as well as electrical brain responses to the target words were recorded. Lexical decisions were fastest in the identical condition (kan-KANTE), intermediate in the deviating condition (kon-KANTE) and slowest in the control condition (kan-ZUNGE). The same pattern was true for error rates with least errors for the identical condition and most for the control condition. No inhibition was obtained, despite a quite similar design to that of Soto-Faraco et al. (2001).
In the event-related potentials (ERP), Friedrich (2005) observed two components: a P350 effect and an N400 effect. The P350 effect was described in earlier work (Friedrich, Kotz, Friederici & Gunter, 2004; Friedrich, Kotz, Friederici & Alter, 2004) as a correlate of lexical activation in a modality-independent mental lexicon. It is most pronounced over left (anterior) electrode sites and its amplitude is responsive to the relatedness between auditory prime fragment and visual target word. That is, peaking around 350 ms, the amplitude for a target that was successfully primed by its first syllable (e.g. am-AMBOSS,
‘anvil’) (Friedrich, Kotz, Friederici & Gunter, 2004) was more negative than that of an unrelated target (am – PENSUM, ‘pensum’). The same was true in the study by Friedrich (2005). The target word elicited most negative amplitudes in a 300-400ms time window if it was preceded by its own first syllable (kan-KANTE), intermediate amplitudes in the vowel-deviation condition (kon-KANTE) and least negative amplitudes in the control condition (kan-ZUNGE, ‘tongue’). If the size of the amplitude reflects the amount of lexical activation of the target word, then there is evidence for graded activation in that kan- activates KANTE more than kon- does, but the latter still leads to activation. Again there is no evidence for inhibition.
Amplitudes of the N400 component showed less sensitivity to segmental overlap. The identical as well as the vowel-deviation condition elicited less
negative amplitudes in a 350-500ms time window than did the control condition.
There was no difference in amplitude between the former two (Friedrich, 2005).
Thus, the N400 appeared to be insensitive to subtle differences between fragment and word, and it divided more superficially between (partial) segmental overlap and lack of segmental similarity.
3.1.4 Comparison of the Findings of Soto-Faraco and Colleagues (2001)