Syllable positions

While determining whether syllable structure was preserved or not, the primary criteria was that of licensing. Positions that license other positions would have to be less vulnerable to errors that effect syllable structure. As seen in ■ Deletion % of stim. ■ Target Substitution

% of stim.

Figure 25, we can see that such licensing positions (onset and coda cores) are less likely to be deleted in the target. However, positions that do not license (onset and coda satellites) are more likely to be involved in such errors. This reinforces the idea that syllable structure is organised as a hierarchy where some positions (peaks and core positions) have a stronger place than others (satellite positions). This pattern is not evident in deletion errors. This brings into question the vulnerability of satellite positions relative to core positions. However, substitutions of core positions do not lead to syllable restructuring in the same way that deletions do. Therefore, such errors may be occurring with greater freedom.

91

AS HK MJ

NC PK

■ Deletion % of stim. ■ Target Substitution % of stim.

Figure 25. Deletion and target substitution errors as percentage of occurrence in the stimuli

0.0%

92 3.5.4. Cluster errors

Cluster errors are the most interesting evidence for syllable representation within the lexicon. In the Dell model, phonemes are specified for syllable position so that a /b/ in onset is intrinsically different from /b/ in coda (i.e., they are in effect treated as two separately stored phonemes). This model would predict that cluster errors preserve their original structure. For example, a heterosyllabic cluster such as /n.t/ would be stored in the Dell model as /n_coda.t_onset/ with each phoneme with a pre-specified syllable position. Therefore, it is more likely that segments that are pre-specified for coda or onset positions would be replaced by in substitution errors by other similarly specified segments. However, for the same reason, the Dell model also predicts that errors where phonemes in homosyllabic clusters move to become part of heterosyllabic clusters (or vice versa) should occur very rarely.

The LRM model on the other hand stores phonemes for a particular word in serial order. Syllabification (and therefore assignment of syllable boundaries) occurs post-lexically.

If syllable boundaries are not assigned until after phonological retrieval within the lexicon, one would not find syllable boundary effects in all the patients. This means that if syllabification only occurs after the segments have been retrieved from the lexicon, then errors that occur within the lexicon will be immune to the restrictions imposed by syllable position and boundaries. They should have greater flexibility in which segments are deleted or which are substituted because there is no syllable structure within the lexicon to serve as a frame to keep segments in position.

Finally, cluster errors are a good indication of whether syllable structure plays a role within the lexicon, because they have more opportunities to restructure. As shown in the example in Figure 26, in a transformation such as pri .ʋiːpi . iː, a heterosyllabic cluster is changed into a geminate. In such a substitution, the only change that is required is to unlink /ʋ/ from the onset position and link the structure to the substituted / /. However, an error such

93 as ʧi. rəɳʧi . ən requires more substantial restructuring. Here the onset satellite position has to be deleted and a new coda position has to be created and linked with the remaining segment.

If syllable structure is stored, this requires greater exertion on the part of the speech production system as it means restructuring the initial syllable structure.

Figure 26. Examples for possible errors of geminates and clusters

What we do find is that geminates remain geminates while heterosyllabic clusters remain the same or become geminates. As there are a higher number of errors involving heterosyllabic clusters, the pattern is clearer with them as opposed to homosyllabic clusters or geminates. There the numbers are much lower (less than 4 in most instances).

Table 13 Errors involving Heterosyllabic Clusters in Hindi

Structure preservation Structure violation

het>het het>gem het>hom

N % N % N %

AS 10 6.7% 2 1.3% 0 0.0%

HK 16 13.0% 5 4.1% 0 0.0%

MJ 23 15.4% 15 10.1% 0 0.0%

NC 13 8.7% 2 1.3% 0 0.0%

PK 17 17.7% 7 7.3% 0 0.0%

94 Table 14 Errors involving Homosyllabic Clusters in Hindi

Structure preservation Structure violation structure. All patients appear to systematically simplify more often than complicate. The differences between simplifications and complications are significant for all of the patients.

Table 16 Syllable Structure changes in Hindi

Therefore, could the preservation of syllable structure be explained through other means such as syllable simplification? To check the validity of this claim, a chi-square was performed

95 against syllable restructuring and structural simplifications. It was found that restructuring was independent of syllable simplification or complication (χ²(1)=0.017, p=.897).

Another explanation would be that the effects are only a result of frequency. This would be consistent with the LRM model’s assumption of a mental syllabary where an error would result in higher frequency syllables replacing lower frequency syllables. This was tested by extracting all the individual syllables in the errors and aligning them with their equivalent syllable in the output. If frequency is the only explanation for the errors, then all patients should consistently replace syllables in the target with those that have a higher frequency.

Figure 27. Syllable replacement across patients

The results show that, in fact, three of the patients have a tendency to replace syllables in the target with lower frequency syllables. Table 17 shows that there is no significant difference between replacements by syllables of higher or lower frequency. The raw values show that the pattern is in the opposite direction to what one would expect (more

Replaced by higher freq syllables Replaced by lower freq syllables

96

If frequency is not affecting the errors in their distribution, could their effects be present in another domain such as syllable simplification? We saw earlier that satellite positions are being deleted more often than core positions. Could this be explained in terms of frequency?

For example, if a CCV syllable becomes a CV syllable it is not only simplifying in structure but also becoming a higher frequency syllable (see chapter 2 on the distribution of syllable types). We tested this hypothesis as shown in Table 18.

Table 18 Difference between Syllable Restructuring and Change in Frequency No restructuring Restructuring syllable restructuring and the change in frequency. PK comes closest to showing a significant difference. The results show that even though simplification is a significant factor in patient errors (see Table 16) it cannot be explained by frequency alone.