3.1 Definición del contexto operacional
3.1.1 Diagrama de flujo del proceso
We found that resyllabification was not as high a burden as previously assumed. In fact, Italian and Hindi resyllabify as low as 1% on average, meaning that it is not a redundant operation. English, on the other hand had a higher resyllabification rate. Italian has been shown to provide good evidence for storing syllable structure within the lexicon (Romani et al. 2011). As resyllabification is one of the reasons for not including syllabic information within the lexicon, do varying resyllabification rates indicate a difference in word-form encoding between languages? Could the evidence that was found for supporting the storing of syllable structure within the lexicon through Italian be found in Hindi but not in English? As English has a higher resyllabification rate relative to Hindi and Italian, would this change the nature of its lexical organisation to such an extent as to exclude syllable structure? To answer these questions we conducted a study with Hindi and English stroke patients with speech problems. There were three possible alternatives: 1) syllable structure is never preserved, 2)
179 syllable structure is always preserved across all languages, or 3) syllable structure is preserved in some languages but not in those that have high rates of resyllabification.
The errors were broadly categorised as word and nonword errors. Both Hindi and English patients made more nonword errors than word errors. One of the main differences between languages was that phonetic errors were not a clear predictor of the rate of simplification in Hindi. Whether it is measured by syllable structure or phonological markedness, Hindi patients always showed a clear tendency to simplify. Patients making more or fewer phonetic errors did not differ in this regard. English patients, on the other hand, were closer to the classifications made for Italian patients by Romani et al. (2011). HN could be classified as a non-fluent patient, JT as a fluent patient and CS as a mixed patient.
HN had the highest phonetic error rate while JT had the lowest, with CS falling in between the boundaries (5% - 10%). The difference between simplification and complication was significant for HN and CS but not for JT who had near equal simplification and complication rates with mostly neutral transformations.
A binomial regression on the effects of length and frequency on errors was performed with all of the patient responses. All of the Hindi patients showed a main effect of length. CS showed main effects of length and frequency while HN showed an effect of interaction between length and frequency. JT showed only frequency effects. The difference in length and frequency effects indicates that JT may have more centralised impairments compared to CS and HN.
The main question of syllable structure preservation was divided in three: 1) are consonants (non-peak positions) more vulnerable to errors than vowels (peak positions), 2) are core positions more vulnerable than satellite positions? and 3) are clusters preserved at syllable boundaries?
180 7.2.2.1. Consonant versus vowel errors
If there is a hierarchically organised syllable structure within the lexicon, then we should find the effects of that organisation in the errors made by the participants. As peaks are the most vital part of a syllable, segments that occupy that position (i.e., vowels) should be less vulnerable to errors than other syllable positions.
We found that in both Hindi and English, consonants were more vulnerable to deletion and substitution errors than vowels. All the Hindi patients except AS showed a significant difference between consonant and vowel substitution. Both CS and HN also showed significant difference while JT did not. All of the Hindi patients showed a significant difference between consonant and vowel deletions as did CS and JT among the English patients. This underlines the vital importance of vowels within the syllable hierarchy. Vowels did play a bigger part in insertion errors as would be expected in patients who simplify clusters with vowel insertions.
Vowel deletion is significantly higher than consonant deletion. Is this because vowel deletion leads to phonotactic irregularities? But it is possible for vowels to be deleted without violating phonotactic constraints by deleting 1) syllable initial vowels, 2) vowels in hiatuses, and 3) vowels between two consonants that can form a phonotactically legal sequence (e.g., /kala//kla/). However, we found that in both Hindi and English such deletions were rare. In English, /l/ and /r/ were not deleted in core positions while they were deleted in satellite position. /m/ and /n/ were not deleted in either core position or when in /s/+nasal clusters. It is possible that /m/ and /n/ should be treated as only occupying core positions as /s/ is usually treated as a pre-marginal. Further evidence needs to be collected in the future with a focus on these segments to clarify the issue.
181 7.2.2.2. Satellite versus core positions
With consonant errors, we found that Hindi patients consistently made more deletions in satellite positions as opposed to core positions. This was also found in English with CS who also made more deletion errors with premarginals. However, this pattern was not apparent with the other two English patients: HN and JT. Substitution errors in both languages did not show the expected asymmetry between core and satellite positions. Core positions were more likely to be involved in errors as opposed to satellites. However, as substitution of core positions do not result in the restructuring of the syllable structure, the absence of asymmetry does not disprove the head-licensing principle.
There is always the possibility that phonotactic constraints can explain the errors. The prevalence of errors in satellite position may be due to the fact that those particular segments are more vulnerable than others. But we found that such segments were not more vulnerable than other segments when they occur in simple onsets. This shows that it is the syllable position they occupy that is vulnerable rather than the segments themselves.
7.2.2.3. Errors at syllable boundaries
All patients in both Hindi and English showed a tendency to preserve the original syllable structure at syllable boundaries: heterosyllabic clusters remained heterosyllabic and homosyllabic clusters remained homosyllabic. In Hindi, the effects were not clear enough due to having a very small number of errors (<10). However, a probability analysis done with the larger set of data from CS showed that his error rates were lower than the probability of changes between different clusters. This provides the strongest evidence for the fact that a syllable structure template is available within the lexicon. It is possible that these preservations could be the result of phonemes being stored with pre-specified syllable positions, meaning that they could not move to other syllable positions when errors were
182 made (as in the Dell model). However, movement errors show that phonemes can move between different syllable positions making it difficult to endorse this view.
In summarising the empirical evidence from Hindi and English we arrive at the following conclusions: 1) deletions are more likely in syllabically weak positions, 2) substitutions at syllable boundaries tend to preserve the original syllable structure, and 3) movement between syllable positions are possible making syllable position-specific phonemes unlikely. The deletion of segments from syllabically weak positions supports the presence of a hierarchical syllable framework. However, the fact that the other two English patients (HN and JT) fail to fully conform to this pattern suggests that further investigation is necessary. The empirical evidence is not fully conclusive in establishing that syllable structure is the only organising principle within the mental lexicon. However, there is enough evidence to suggest that this may be a valid consideration and worthy of further investigation that may yield more definite results.
7.2.2.4. Phonological markedness effects
We saw in the previous analyses that patients (particularly Hindi patients) had a strong tendency to simplify syllable structure. Can this simplification at the structural level also be found at the segmental level? We assessed this using phonological markedness.
The change in phonological markedness during substitution errors was assessed for both Hindi and English patients. Within phonology, markedness is often defined in binary terms with unmarked items being simpler, more frequent or easier to articulate than marked items. However, phonemes cannot be classified as such as they are a bundle of distinctive binary (or unary) features whose markedness can differ depending on the values being considered. For example [g] may be considered less marked than [θ] when considering manner features (stops being less marked than fricatives). However, when considering place, [g] would be more marked than [θ] because velar consonants are considered more marked
183 than coronal consonants. If we look at voicing, [g] is again more marked. Considering these issues, an average markedness value was considered based on how the target and response phonemes ranked in terms of place, manner, voicing (for both Hindi and English) and aspiration (for Hindi). No markedness effects were found when we ordered phonemes according to place or manner. Some effects were evident when we considered single feature properties such as voicing and aspiration. While 3 of the Hindi patients showed markedness effects, none of the English patients showed any effect. The study does provide a justification for looking at markedness as single feature attributes (such as voicing) rather than as multi-feature blocks (such as place or manner).