Systematicity-driven linear parameter impact values

The following Figures show the linear impact values of the phonological parameters in the cvcv and the cvccv (Figure 6.3) and the cvcvcv (Figure 6.4) word groups, in the ‘syntax’ and ‘no syntax’ conditions. The bars represent the covariance of each parameter with the phon-sem systematicity. (Note that these bars represent the effect of the joint application of all parameters to the phonological space calculation.)

cvcv, syntax

Figure 6.3: Linear ‘parameter impact’ values, representing the impact of the phonological similarity parameter on the phon-sem correlation. Two conditions, syntax and no syntax are shown for the cvcv and cvccv word groups. White bars= consonant-related parameters;

grey bars=vowel-related parameters; black bars=stress-related parameters; striped bar=structure-related parameter. Unless otherwise stated, p<0.01.

cvcvcv, syntax

Figure 6.4: Same as Figure 6.3 above, but for cvcvcv words. Unless otherwise stated, p<0.01.

The parameter impact values show a high level of coherence across the three word groups. Comparable parameters across groups are highly correlated, as shown in Table 6.3.

Table 6.3. Consistency across word-groups in the ‘syntax’ and ‘no syntax’ conditions: R² of counterpart parameter impact values indicates covariance of the parameters with respect to lexicon systematicity in three independent word groups - cvcv, cvccv and cvcvcv. The number of parameters in each condition is shown in brackets. All p<0.01.

These high across-group correlations provide support for the methodology employed, indicating that the phonological parameters have the same impact on phon-sem systematicity in three independent subsets of the lexicon. In

other words, this consistency supports the existence of systematicity between the phonology and the semantics of the lexicon - if there was no phon-sem systematicity in the lexicon, the phon-sem correlation would not have generated the same phonological parameter values in three independent word groups.

One similarity between the three word groups is that sharing all consonants or all vowels (tc, 3c, tv, 3v) tends to have greater impact on systematicity (higher parameter values) than sharing single consonants or vowels (c1, c2, c3, v1, v2, v3). The only exception is sharing the final vowel (v3) in the 'syntax' condition in cvcvcv words, with a higher impact than sharing any combination of vowels. The morphosyntactic information encoded by the final vowel may explain its positive impact in the ‘syntax’ condition.

Figures 6.3 and 6.4 show that most consonant parameters (in white) have positive impact on systematicity, while vowel parameters (in grey) have a negative impact. The only exceptions are negative c2 (the syllable-final consonant) and c3 (the second-syllable initial consonant) in cvccv words.

Other exceptions are sharing all vowels in cvccv and cvcvcv words, and, as mentioned earlier, the last vowel in cvcvcv words. Note that impact value of the final vowel is much lower in the 'no syntax' condition than in the 'syntax' condition in all three word groups. This may be explained by the fact that the final vowel carries in many instances morphosyntactic information:

when correlated with syntax-laden semantic representations, the phonological representations are more influenced by the weight of the last vowel.

Another common feature of the 'syntax' condition across word groups is the high impact value of stress parameters in the last and one-but-last syllable.

In the three word groups, sharing the stress on the same syllable brings two words close together in the phonological similarity space. Because the parameter impact value is so high, we know that words sharing the stress on the same syllable must be close together in the semantic similarity space too.

Sharing the same stressed vowel has very different effects depending on the syllable. The same stressed vowel on the final syllable makes words very phonologically similar. The stressed final vowel, as explained in § 3.2.2.5, encodes important verb morphosyntactic information. The fact that the present methodology so clearly picks up the importance of parameter sv2 in the phonological similarity space when correlated with a syntax-laden semantic space both endorses the methodology and confirms the syntax-phonology space correlation proposed by the phonological typicality literature (Durieux & Gillis, 2000; Kelly, 1992, 1996; Monaghan, Chater &

Christiansen, 2003). Enhancing final stressed vowel distinction at the phonological level greatly improves the phon-sem correlation, so this parameter must be driven by verb endings, with their highly systematic relationships between phonological and cooccurrence-based representations.

Sharing the stressed vowel on the penultimate syllable has a very negative impact on systematicity. Over 80% of Spanish bi- and trisyllablic words are stressed on the penultimate syllable, so the negative impact value indicates that sharing the same stressed vowel in the penultimate syllable (phonologically similar words) makes words semantically dissimilar. This is going against the systematicity pressure, but may help an opposed pressure:

the pressure for words to be easily distinguished from each other, particularly words that occur in similar contexts.

This suggests that while the identity of the final stressed vowel organises the systematic lexicon on a morphosyntactic dimension, the identity of the vowel in the stressed penultimate syllable may be crucial for word differentiation and recognition.

The linear impact values have given us an idea of the role each phonological similarity parameter plays on the phon-sem correlation. The next section explores which regression models best predict the behaviour of the parameters.