Zwitserlood (1989) showed that multiple word candidates are active as long as the information in their lexical entries is consistent with the information in the input. In a cross modal fragment priming design using semantically related auditory prime fragments and visual target words, a Dutch word fragment like kapi- that could be taken from either kapitein (captain) or kapitaal (capital) activated both semantically related target words, SCHIP (ship) and GELD (money). When the fragment was longer and contained disambiguating acoustic information (e.g.
kapita-), only the respective semantic associate stayed active (e.g. geld), while the activation status of its competitor word (e.g. schip) was comparable to that of a completely unrelated control target. Apart from segmental information, also biasing sentence context could reduce or deactivate the non-intended target word despite matching phonological information. These results confirm that all items in the mental lexicon that match the input signal are initially activated but lose their activation as soon as the signal deviates from their stored representation.
Using Dutch, Marslen-Wilson and Zwitserlood (1989) investigated the effects of competitors on lexical activation. They compared the impact of phoneme mismatches that created other real words to mismatches that created pseudowords.
In a semantic cross-modal priming design with lexical decision, the visual target BIJ (bee) was recognized faster when it was preceded by the word honing (honey) than by pakket (parcel). Further, responses to BIJ after honing were also faster than those to BIJ after the real word woning (dwelling) and after the pseudoword
*foning, both differing only in the onset consonant. There was no difference in reaction time between the woning- and the *foning-condition. If lexical competition had played a role (e.g. via lateral inhibitory links), then the real word woning should have primed the target BIJ less strongly than the non-existing word
*foning, because the latter has no lexical entry that could inhibit honing. That is, Marslen-Wilson and Zwitserlood (1989) confirmed the existence of graded activation in the mental lexicon, in that woning and *foning activated the target BIJ, but to a lesser extent than honing. Note, however, that this was only the case if
the cohort size was limited to two alternatives (e.g. honing and woning), while no priming was found if there were more than two close competitors.
Having established reduction of lexical activation due to deviating segments, researchers aimed to find out whether activation differences are sensitive to information at the sub-segmental level. Connine, Blasko and Wang (1994), for example, showed that several alternatives are activated in the mental lexicon by sub-segmental ambiguity. When listeners heard an auditory prime with an ambiguous onset consonant in terms of voicing (e.g. a voice onset time intermediate between dime and time), both semantically related visual targets, PENNY and CLOCK were identified faster than when preceded by control primes (e.g. ambiguous glad-clad or unambiguous war). However, the size of priming was reduced as compared to priming with perceptually unambiguous words (e.g. dime – PENNY). In contrast to Zwitserlood (1989), Connine et al. (1994) did not find effects of biasing sentence contexts. Both targets, PENNY and CLOCK obtained equally facilitated responses after d/time in a context that favoured only one interpretation.
Bölte and Coenen (2002) examined gradual activation by means of cross-modal priming experiments with lexical decision tasks, once using semantic priming and once direct phonological priming. In the semantic priming, a word prime (e.g. tomato) activated its target word (e.g. PAPRICA) as strongly as a prime pseudoword that deviated from the word prime in only one onset feature (e.g.
*domato – PAPRICA). In contrast, if the prime deviated from the real word in several phonological features (e.g. *zomato), response speed in target (PAPRICA) lexical decision was reduced, but still faster than in the control conditions (*sibrary/library – PAPRICA). That is, the phonological difference between a deviation by one feature as compared to several features was reflected in response latency and hence in amount of lexical activation. Results looked different in the phonological priming design. The identity priming condition (paprica – PAPRICA) was faster than the conditions with onset-deviation or with control words. There was no difference in reaction time between targets primed by pseudowords that deviated in only one feature (*baprica) versus those that deviated in several features (*zaprica). However, these latter conditions were still faster than control conditions.
In sum, semantic priming did produce some evidence for multiple activation of lexical representation as a function of the goodness-of-fit between the input and the information stored in the mental lexicon (Zwitserlood, 1989;
Marslen-Wilson & Zwitserlood, 1989; Connine, Blasko & Wang, 1994; Bölte &
Coenen; 2002). Semantic priming experiments also revealed that phoneme mismatch might block lexical activation (Zwitserlood, 1989), but they did neither report evidence for inhibition of partially mismatching words nor did they show differences in activation caused by word versus pseudoword primes (Marslen-Wilson & Zwitserlood, 1989). Furthermore, the results of Bölte and Coenen (2002) suggest that semantic priming might provide a more fine-grained measure of lexical activation than phonological priming, which did not prove sensitive to the number of mismatching features between prime and target.
Gradual decrease in activation of lexical representations has received some further support from results of phoneme monitoring experiments. Connine, Titone, Deelman and Blasko (1997) asked participants to listen to words and pseudowords and press a button whenever they heard the stop consonant /t/, /k/ or /p/. In the crucial conditions, the target phoneme was word finally. The assumption is that a target phoneme is easier and hence faster to recognize if it is part of a word than if it is embedded in a pseudoword. Therefore reaction time can be used as an indicator of lexical activation. Indeed, the /t/ in cabinet was spotted faster than the /t/ in *shafenet. The crucial conditions were those that changed one or several features of the word initial segment, leaving the rest of the word intact. Results revealed that the /t/ in cabinet was spotted faster than the /t/ in *gabinet (one feature changed) which was again faster than the /t/ in *mabinet (several features changed), while complete pseudowords (*shafenet) yielded slowest monitoring responses. The same was true for monitoring of /p/ but curiously not for /k/.
Frauenfelder, Scholten and Content (2001) further extended these results in French by placing the phoneme target not only at the end but also word medially and by introducing the mismatch word-initially, word-medially and word-finally.
Again phonemes were monitored fastest in undistorted words (e.g. vocabulaire,
‘vocabulary’). If there was a mismatch in word-onset, lexical activation in terms of faster monitoring times could only be seen in conditions with a word final target (e.g. *focabulaire) but not for word-medial targets (e.g. *focabulaire). This suggests that the lexical item needed time and considerable supporting input to overcome an initial mismatch. Also, when the mismatch occurred word-medially, shortly before the target phoneme (e.g. *vocabunaire), there were no decreased response times. Thus, the results by Connine et al. (1997) support the assumption that features rather than whole segments are considered in lexical access while Frauenfelder et al. (2001) further emphasize the impact of the amount of additional supporting information and the distance between the mismatch and the target phoneme in the word.
In sum, there is ample evidence for graded lexical activation and for processing on a feature-based rather than phoneme-based level from a wide variety of experiments employing semantic priming (Zwitserlood, 1989; Marslen-Wilson
& Zwitserlood, 1989; Connine, Blasko & Wang, 1994; Bölte & Coenen, 2002) and phoneme monitoring (Connine, Titone, Deelman & Blasko, 1997; Frauenfelder, Scholten & Content; 2001). However, results on inhibitory effects for gradually mismatching words are still inconsistent, as outlined below.