- Email: [email protected]
Explaining phonotactics using NAD
Language Sciences 46 (2014) 6–17
Contents lists available at ScienceDirect
Language Sciences journal homepage: www.elsevier.com/locate/langsci
Explaining phonotactics using NAD Katarzyna Dziubalska-Kołaczyk* , Poland University of Poznan
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 27 July 2014
This paper presents a model of phonotactic grammar in which wellformedness of consonant clusters is measured by NAD. NAD stands for a Net Auditory Distance obtaining between segments in a cluster. The auditory distance is a net reflection of the differences between segments in terms of manner (MOA) and place of articulation (POA). It is calculated according to the Principle which states that a cluster is preferred if it satisfies a pattern of distances specified by the universal phonotactic preference relevant for its position in a word. Every position of a cluster in a word, i.e. initial, medial and final, is defined by a respective well-formedness (“goodness of cluster”) preference. The NAD Principle makes finer predictions than the sonority sequencing generalization (SSG). For example, it predicts that initial pr- is “better” (more preferred) that tr-, and they are both better than ps- or rt-, while the latter two are of comparable value. However, phonology alone does not fully account for clusters. Inflection, word-formation and compounding contribute to the creation of consonant clusters to an extent relative to a morphological type of a language. Therefore, a phonotactic grammar operates on basic, non-derived, lexical forms, while morphonotactics takes care of the remaining, morphologically complex, forms. Interaction between phonotactics and morphonotactics provides a richer insight into the understanding of cluster complexity. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Phonotactics Net Auditory Distance Morphonotactics Beats and Binding Phonology Polish
1. Introduction The model of phonotactic grammar presented here is embedded in the theory of Beats and Binding Phonology (B&B henceforth) which is in turn derived from the principles and assumptions of Natural Phonology (NP) and employs the epistemology of Natural Linguistics (NL). B&B phonotactics is based on universal preferences which are grounded in phonetics and expressed by means of the Net Auditory Distance (NAD) Principle. NAD functions as a measure of wellformedness of lexical (i.e. phonological) clusters. In clusters arising in morphologically complex words NAD Principle is often overridden by morphonotactics. Having characterised the theoretical background, I will formulate the hypotheses with reference to the Polish data obtained in the project1 and discuss the results of the analysis. In conclusion, NAD parameters and the NAD Principle as such will be placed under scrutiny and directions for future research will be suggested.
Abbreviations: NAD, Net Auditory Distance; MOA, manner of articulation; POA, place of articulation; SSG, sonority sequencing generalisation; NP, Natural Phonology; NL, Natural Linguistics; B&B, Beats and Binding; OT, Optimality Theory. , Al. Niepodleglosci 4, Poznan , Poland. Tel.: þ48 519 340 568. * Faculty of English, Adam Mickiewicz University in Poznan E-mail address: [email protected]. 1 The new data from Polish (and English) obtained within a project Phonotactics and morphonotactics of Polish and English: description, tools and appli , Paulina Zydorowicz, Paula Orzechowska and Dawid cations (N N104382540). My collaborators in the project have been Michał Jankowski, Piotr Wierzchon Pietrala (cf. e.g., Dziubalska-Kołaczyk et al., 2011, 2012). http://dx.doi.org/10.1016/j.langsci.2014.06.003 0388-0001/Ó 2014 Elsevier Ltd. All rights reserved.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
7
2. Epistemology 2.1. Beats and Binding Phonology Beats and Binding Phonology (Dziubalska-Kołaczyk, 2002) is a syllable-less theory of phonology embedded in Natural Phonology. The structure usually referred to as “the syllable” in other models is epiphenomenal here or emergent due to principled phonotactic forces. The latter are responsible for different degrees of intersegmental cohesion (Bertinetto et al., 2006) which, in turn, determines the behaviour of segments and creates the impression of syllable structure. B&B Phonology was proposed out of a need for better, more comprehensive and holistic explanations of phonological phenomena than the ones prompted by the syllable.2 Basic units, beats (mostly vowels) and non-beats (always consonants, including glides), are connected into an alternating sequence by means of bindings according to the principle of perceptual salience. Such alternation is both acoustically and physically grounded: “To construct a useful signalling system out of sound, there must be some differentiation between different parts of the signal in time. It appears that a basic organization of this differentiation of sound in all (spoken) languages consists of an alternation between louder and quieter levels of sound, with a period not too far from 150-200 ms” (Maddieson,1999: 2525) This amounts to “[a] fairly regular wave-like alternation of amplitude peaks and valleys. The occurrence and timing of this pattern have been suggested to be related to a natural frequency of the jaw, which can be approximately equated with a comfortable mastication rate” (Maddieson, 1999: 2525) The alternating sequence is cut into pieces called words and morphemes on its way from the prelexical to lexical level. Clusters of non-beats are “accidents” due to some distortions of the alternating sequence (e.g., dropping or breaking of some beats). The occurrence and subsequent behaviour of consonant clusters is controlled by phonotactic preferences (and morphonotactics). Bindings are used in B&B Phonology to account for both segmental and prosodic structure. In this paper, however, the focus is on phonotactics. 2.2. Natural Phonology and Natural Linguistics The explanation pattern in B&B Phonology is epistemologically grounded in Natural Phonology (Stampe, 1979; Donegan and Stampe, 1979) and Natural Linguistics (Dressler, 1996 and other publications). Basic thesis of NP has been that phonological systems are phonetically motivated. The way to understand it is that processes and sounds have phonetic grounding in all speakers. It does not mean that the same processes will apply in all languages or that they will derive identical outputs across a variety of contexts. Sounds, apart from being contrastive, also have sub-categorical features. NL has an expanded epistemology in order to be able to account for all the components of language. It is a functionalist framework with semiotic underpinning. It is a preference theory which heavily relies on general non-linguistic principles as well as on sources of the so called “external” evidence. Language user is central to the explanation, since “preference” implies a human agent who behaves functionally in his/her linguistic performance. However, the goals of a given performance may be contradictory and the circumstances highly complex. Therefore, NL attempts a holistic explanation, expressed in terms of hierarchies of preferences according to complex sets of relevant criteria. The explanatory model of NL may be summarised as in the table below (Table 1). The starting point is a higher order nonlinguistic principle from which a linguistic preference is derived and measured by respective parameters, which in consequence provides an explanation for a given language-specific structure. 2.3. B&B phonotactics The model of phonotactic grammar with NAD is part of B&B Phonology. While beats and non-beats alternate on the basis of perceptual contrast (as discussed above in Section 2.1), actual auditory distances between sounds in a sequence become relevant when the melody (phonetic content) is filled in and we deal with vowels and consonants. Clusters of consonants tend to be avoided, subject to the universal CV preference. Typologically, the CV is a universal constituent type which occurs in all languages. As shown by Maddieson (2009) on the basis of a sample of 486 languages: 12.5% of the languages allow only CV’s 56.6% of the languages have the moderately complex structure CCVC, with limitations, however, on which consonants may appear in the CC cluster, the second consonant typically being a liquid or a glide 30.9% of the languages have complex structures (C)(C)(C)V(C)(C)(C)(C).
2 The syllable itself has always been problematic to define, while the various syllable-based explanations have often been supported by such circular notions as ‘extrasyllabicity’ and ‘ambisyllabicity’. For a comprehensive discussion of the syllable-related issues, including the history of approaches to the syllable, cf. Dziubalska-Kołaczyk (2002).
8
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17 Table 1 The explanatory model of Natural Linguistics. higher principles (e.g., the principle of the least effort, of cognitive economy) preferences (e.g., a preference for simple phonotactics, for a CV structure) preference parameters (pronunceability, perceptibility) consequences of preferences (absence of clusters in a language)
non-linguistic (cognitive, phonetic, psychological, sociological etc) linguistic functional and semiotic linguistic
The higher level principle which underlies the perceptual contrast in the CV is that of figure and ground (cf. the Rubin’s vase/ face illusion, discovered by Edgar Rubin, cf. Rubin, 1915). There is also a clear phonetic grounding for the CV: larger modulations have more survival value than lesser ones and therefore will persist in languages (Ohala, 1990a) “it is generally the case that the most salient acoustic modulations in a syllable occur near the CV interface”, and “auditory cues present in CV’s are more robust than those in VC’s” (Ohala and Kawasaki, 1984: 117, 118) “since there is a richer, more reliable set of place cues in the CV transition than the VC transition, listeners weight the former more heavily than the latter in deciding what they’ve heard” (Ohala, 1990b: 265)3 on the speaker’s side, according to Ohala and Kawasaki (1984: 119), “the speaker actively tries to create temporally more well defined, more precise, articulations near the CV as opposed to VC interface”. Since CV is a preferred phonological structure and clusters of consonants tend to be avoided across languages and in performance, there must be a phonological means to let them function in the lexicon relatively naturally. This is achieved by auditory contrast and its proper distribution across the word. Importantly, auditory impression is the overall product of articulation, mediated by acoustics. It is believed, therefore, that auditory (perceptual) distance can be expressed by respective combinations of articulatory features which eventually bring about the auditory effect. The underlying motivation behind a phonotactic preference is to counteract the CV preference, shown to be universal typologically and in performance (although clusters arise in performance, too). Any cluster in a structure which is more complex than CV is susceptible to change leading to CV, e.g., via cluster reduction (consonant deletion) CCV / CV or vowel epenthesis CCV / CVCV or at least vowel prothesis CCV / VCCV. A way to counteract this tendency is to introduce another salience into the structure, i.e. increase the perceptual distance between the consonants (CC of the CCV) to counterbalance the distance between the C and the V (CV of the CCV). In this way preferences which define “the survival” of complex structures in all positions of the word will be formulated. Besides the above markedness criterion, cluster size remains an obvious measure of cluster complexity: longer clusters are unanimously more complex than the shorter ones. Using the OT terminology of ranking, cluster size ranks higher as a constraint than the markedness criterion expressed by the auditory distance (see below for NAD). 3. NAD Principle and phonotactic preferences 3.1. NAD The Net Auditory Distance Principle (Dziubalska-Kołaczyk, 2009) defines cluster preferability in relation to the position in the word (initial, medial and final). It reads: A cluster is preferred if it satisfies a pattern of distances specified by the universal phonotactic preference relevant for its position in the word. NAD is a measure of distance between two neighbouring elements of a cluster in terms of differences in MOA (manner of articulation) and POA (place of articulation). The difference in voicing (Lx) has been considered, too, however, laryngeal features are non-redundant within subclasses of sounds only (e.g., they are non-redundant within obstruents and largely redundant within sonorants) and as such will have to be included in more refined, class-specific calculations4 (see Kehrein and Golston, 2004 and the criterion of spread glottis in Basbøll, 2005). In fact, an indefinite number of articulatory features as well as detailed acoustic cues would have to be investigated in terms of the degree of their contribution to the overall auditory effect obtained in a cluster (cf. e.g., Johnson, 2003).
3 4
A reviewer notices that there are exceptions, e.g., retroflexion is said to be more audible in VC. In fact, I’d like to introduce the Obstruent/Sonorant – O/S distinction instead.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
9
Table 2 Distances in MOA and POA.
Manner and place of articulation, however, constitute a rather reliable starting point since they form a backbone of any articulation, carrying responsibility both for perceptual clarity (manner, an articulatory correlate of sonority, of a dissimilatory nature) and production ease (place, of an assimilatory nature). The distances in terms of MOA and POA have originally been calculated on the basis of Table 2. The manners and places assumed in the table are selected according to their potential relevance: 6 manners (stop, affricate, fricative, sonorant stop, approximant, semivowel), where affricates and semivowels are, tentatively, attributed half a distance due to their ambiguous nature; and 5 places (labial, coronal, dorsal, radical and laryngeal or glottal). Manners refer to the most generally acknowledged version of the so-called sonority scale, while places are taken from Ladefoged (2006: 258). Both lists are extendible and modifiable (e.g., Ladefoged’s list consists of 5 nodes which branch into 12 more detailed features), depending on the amount of detail we want to include in the definition of distance. Importantly, vowels need to be differentiated according to their colour, too, which will allow to reflect the intersegmental distances with a higher degree of precision. Within the project quoted here (cf. footnote 2 above), separate tables for Polish and English consonants5 have been used, reflecting the differences between the two systems as well as including more detailed MOA and POA scales (see Tables 3 and 4). It remains to be verified in further research whether the proposed scaling actually accounts for the data more accurately (see also the discussion in Section 7). 3.2. Preferences The NAD Principle evaluates cluster markedness with reference to universal phonotactic preferences. The preferences, describing two- and three-consonant clusters in word-initial, medial and final position, are listed below. An example of a calculation is provided for the first preference. 3.2.1. C1C2V NAD (C1,C2) NAD (C2,V) Calculation: NAD CC ¼ j(MOA1 MOA2)j þ j(POA1 POA2)j
5
Polish consonants based on Jassem (IPA, 2003), English consonants based on Roach (BrE, IPA, 2001).
10
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17 Table 3 Distances in MOA and POA: Polish.
Table 4 Distances in MOA and POA: English.
NAD CV ¼ jMOA1 MOA2j e.g., prV in Polish pr: j(MOA1 MOA2)j þ j(POA1 POA2)j j5 2j þ j1 2.3j ¼ j3j þ j1.3j ¼ 4.3 so, NAD CC ¼ 4.3 rV: jMOA1 MOA2j ¼ j2 0j ¼ 2 NAD CV ¼ 2 so, NAD CC NAD CV ¼ 2.3 so, the preference NAD (C1,C2) NAD (C2,V) is observed since 4.3. > 2.3 3.2.2. VC1C2 NAD (V,C1) NAD (C1,C2) 3.2.3. V1C1C2V2 NAD (V1,C1) NAD (C1,C2) < NAD (C2,V2) *NAD (C1,C2) > 0
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
11
Fig. 1. The phonotactic calculator.
3.2.4. C1C2C3V NAD (C1,C2) < NAD (C2,C3) NAD (C3,V) 3.2.5. VC1C2C3 NAD (V,C1) NAD (C1,C2) > NAD (C2,C3) 3.2.6. V1C1C2C3V2 NAD (V,C1) NAD (C1,C2) & NAD (C2,C3) < (C3,V2) 3.3. The phonotactic calculator6 In order to automatically calculate NAD, a phonotactic calculator has been developed (Dziubalska-Kołaczyk and Krynicki, 2007; Dziubalska-Kołaczyk and Pietrala, Phonotactics and morphonotactics project, Pietrala, 2014). It allows for statistical analysis of phonetic dictionaries and phonetically annotated corpora from various languages. It works on various lengths of clusters in all word positions and estimates them with respect to the universal phonotactic preferences (Dziubalska-Kołaczyk, 2002; cf. above Section 3.2). It provides fast feedback on the predictability value of the hypotheses expressed through the preferences. The calculating platform opens as shown below in Fig. 1. Fig. 2 shows an example of NAD calculation, for the most frequent 14 two-consonant initial clusters of Polish. Fig. 3 illustrates the graphic representation of the results of the above calculation generated by the calculator. As can be observed, 10 out of 14 most frequent Polish initial clusters of two consonants have a positive value of NAD, i.e. they are unmarked according to the NAD Principle. 4. Morphonotactics Phonology alone does not fully account for clusters. Inflection, word-formation and compounding contribute to the creation of consonant clusters to an extent relative to a morphological type of a language. A phonotactic grammar operates on basic, non-derived, lexical forms. Morphotactic forms need a wider account, however. As we suggested in Dressler and Dziubalska-Kołaczyk (2006), morphonotactics is the area of interaction between morphotactics and phonotactics and provides a richer insight into the understanding of cluster complexity. Language specific morphonotactics provides thus an additional parameter constraining the actual outcome of universal phonotactic preferences.
6 The phonotactic calculator will be made available online for research purposes. It will be open to modifications, including the criteria considered as well as the values assigned to them.
12
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
Fig. 2. The phonotactic calculator at work.
Fig. 3. The phonotactic calculator: a NAD graph.
A need to signal a morphological boundary may override a phonologically driven phonotactic preference and, consequently, lead to the creation of a marked cluster. Therefore, one expects relatively marked clusters across morpheme boundaries and relatively unmarked ones within morphemes. In view of the above discussion, two types of clusters have been distinguished: phonotactic (lexical) clusters, as in band, past and morphonotactic clusters, as in ban(n)þed, passþed. The following sources of morphonotactic clusters have been considered: concatenative, e.g., Polish, /ft-/ in w þ toczyc ‘to roll into’, z þ robic ‘do’, s þ pasc ‘fall’ (perfective vs. imperfective), English /-fs, -vz/ in laughs, loves, wife’s, wives non-concatenative, e.g., Polish len vs. lnu ‘linen’ (nom.sg. vs. gen.sg.), przeste˛ pstwo w przeste˛ pstw [-mpstf] ‘crime’ (nom.sg. vs. gen.pl.), tratwa w tratw [-tf] ‘raft’ (nom.sg. vs. gen. pl.) miejsce vs. miejsc ‘place’ (nom.sg. vs. gen.pl.), punkt vs. punkcie ‘point’ (nom.sg. vs. loc.).
5. Hypotheses and data The phonotactic and morphonotactic assumptions lead to the following hypotheses concerning clusters.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
13
Table 5 The most complex clusters from the corpus. Cluster length LEX MORPH Examples 6
–
100%
5
–
100%
4
5%
95%
przeste˛ pstwie, wewna˛ trzwspólnotowy, wewna˛ trzzwia˛ zkowy nstwie, bezwzgle˛ dnie, rozbrzmiewa, kontrprzykład, postzwia˛ zkowy, tysia˛ cstronicowy dzieci /rstf/ warstwa (lex 3%), ministerstwo (-stwo morph 97%), krwia˛ (lex), krna˛ brny (lex), zbło (lex), roztkliwiac (morph), zd odstrzał (morph), je˛ drny (lex)
Fig. 4. Word-initial clusters from the dictionary: cluster types and words.
Hypothesis 1: Cluster size corresponds to morphological complexity. The longer a cluster, the more likely it is to be morphonotactic. Hypothesis 2: The degree of phonological preferability is inversely proportional to morphological complexity. Morphonotactic clusters are expected to have a lower degree of preferability than phonotactic ones. Hypothesis 3: The degree of cluster preferability is directly proportional to frequency. Preferred clusters are expected to be more frequent than dispreferred clusters. The hypotheses have been verified against the Polish data of three types: a dictionary of 8.000 morphologically parsed lemmas, a paradigm list of 190.000 inflected forms with morphological parsing of 5 most frequent initial and final clusters and a corpus of 500.000 inflected forms as a reference for the frequency of use. 6. Data analysis and discussion Unsurprisingly, Hypothesis 1 has been fully confirmed. The most complex clusters from the corpus, those consisting of 5 or 6 consonants, are all exclusively morphonotactic, those consisting of 4 consonants, in 95 percent of the cases (see Table 5). The dictionary data had been analysed with respect to the clusters of 2 and 3 consonants for final clusters, and 2, 3 and 4 consonants for the initial ones. Fig. 4 shows the percentage of lexical (phonotactic) and morphological (morphonotactic) clusters in word initial position with respect to cluster types and the number of words. It is clear that while the number of lexical clusters decreases with cluster size, the opposite is true for morphonotactic ones: 70 percent of 2-consonant clusters are phonotactic while 70 percent of the 4-consonant long clusters are morphologically complex. The same is true for the word final clusters: 90 percent of 2-consonant clusters are phonotactic while –CCC clusters tend to be triggered by morphology especially in word occurrences (see Fig. 5). Hypothesis 2 has been corroborated with respect to morphonotactic clusters which have a strong tendency to be marked (dispreferred according to NAD), as shown in Figs. 6 and 7 (P stands for ‘preferred’, D for ‘dispreferred’). However, one would expect phonotactic clusters to have a higher degree of preferability in comparison to the morphonotactic ones. This remains to be accounted for. Hypothesis 3 concerned the relationship between cluster markedness and frequency. Paradigm and corpus results have been compared to those from the dictionary. The hypothesis has been only partially corroborated. In the paradigm and the corpus, 5 most frequent initial CC clusters are pʂ7, pr, st, mj, sp (of which pr and mj are unmarked) and 5 most frequent final CC clusters are st, ɕṭɕ, nt, nʦ, ŋk (of which nt and ŋk are preferred). The relatively marked clusters are partially accounted for by morphology, especially the initial sp and most of the finals, as shown in Figs. 8 and 9.
7 The symbol ʃ in the data for Polish should be interpreted as a phonetic ʂ. The same symbol (ʃ) was used in our Polish and English data for the sake of simplicity.
14
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
Fig. 5. Word-final clusters from the dictionary: cluster types and words.
Fig. 6. Word initial CC- clusters from the dictionary: types and words.
Fig. 7. Word final -CC clusters from the dictionary: types and words.
Additionally, pʂ is the initial cluster of three highly productive prefixes, przed, przy and prze. Five most frequent clusters from the dictionary diverge from the paradigm/corpus ones: initial mj and final nʦ, ŋk are not found in the five. Five most frequent initial CC clusters are pʂ, pr, st, sp, pj and 5 most frequent final CC clusters are ɲṭɕ, ɕṭɕ, nt, zm, st. The unmarked clusters are clearly lexical, while the dispreferred ones are mixed (see Fig. 10). In order to get a wider perspective on Hypothesis 3, 14 most frequent initial clusters of two consonants were extracted from the dictionary, which included altogether 238 initial cluster types (Fig. 11). Out of the fourteen, only four clusters are marked: pʂ and the three s þ stop clusters (see Fig. 12). While the frequent occurrence of pʂ receives a morphological account, s þ stop clusters need a separate justification (cf. Section 7 for the discussion). Otherwise, NAD successfully accounts for the frequency. 7. NAD parameters and assumptions: discussion and conclusions The study has firmly established that longer clusters tend to be morphologically complex and morphonotactic clusters tend to be phonologically dispreferred (marked). This means that morphonotactics indeed helps to explain
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
15
Fig. 8. Frequent initial clusters in the paradigm (p) and in the corpus (c).
Fig. 9. Frequent final clusters in the paradigm (p) and in the corpus (c).
Fig. 10. Frequent initial and final clusters in the dictionary.
cluster complexity.8 Phonological explanations based on the NAD Principle, however, apparently do not cover all cases and therefore require modification or extension. Below I will discuss the issues to be resolved in further research. Firstly, there are dispreferred (marked according to NAD) lexical clusters. In the Polish data these were, e.g., pʂ: this is the type of a cluster in the case of which an extra-phonological explanation is called for, since it occurs within highly productive affixes, e.g., in przed-, przy-, prze-. Thus, morphological productivity is one source of explanation for the presence of phonologically marked clusters in the lexicon.
8 Morphology has a semiotic priority over phonology and motivates meaning. A salient, phonologically marked cluster performs a morphological function better than a phonologically preferred, natural one. Thus, morphonotactics contributes independent predictions about the shape of clusters.
16
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
Fig. 11. Total distribution of 238 cluster types in the dictionary.
Fig. 12. Fourteen most frequent initial clusters from the dictionary.
Secondly, the notorius s þ stop clusters, which are both lexical and morphonotactic, are dispreferred by NAD. These have been approached in a variety of ways (see Olender, 2013a, b for an extensive overview) none of which seems to have given a satisfactory general account of their omnipresence across the languages of the world. The closest might have been Basbøll’s SSM model (Sonority Syllable Model, cf. Basbøll, 2005) which refers to a universal logic of segment types (sonorants vs. obstruents, all sonorants are voiced, not all voiced sounds are sonorants, etc.). This leads him to use the feature [spread glottis] to account for apparent violations of the sonority hierarchy. In particular, the SSM predicts that marginal segments in absolute initial and final position have [spread glottis], e.g., st-, ts-; -st, -ts (which agrees with final devoicing e.g., in Fr. –fl). However, forcing a general allembracing account for all cluster types may not be the most effective approach. The alternative would be detailed phonetic or phonological accounts for sub-types. Olender (2013a, b), for instance, proposes that leftmost s þ stop clusters should be treated as highly cohesive sibilant þ stop class clusters bound by a “unique perceptual bond”. This bond is highly sensitive to breaking: even the minimal vocalic intrusion (epenthesis) immediately results in the loss of a cluster in perception. Such well-informed phonetics may very well explain the preference for s þ stop clusters without reference to any phonological theory. A phonological theory can also make allowances for sub-types, or, alternatively, provide general as well as languagespecific accounts. This brings us to the third point in the discussion of NAD: the scope of NAD. On the one hand, we may provide descriptions of phonotactics of particular languages based on language-specific parameter values; on the other hand, the same (universal) parameter values guarantee cross-language comparability. Another related question concerns the number and sensitivity of the NAD parameters as well as their relative weight within a cluster. In the present version of NAD I have moved away from including the feature [voice] in the calculation of distances since the laryngeal gesture on its own seems not to distinguish consonants in a cluster. Besides, it is not functional within the class of sonorants. Basbøll’s proposal of [spread glottis] may be considered instead. The parameters of MOA and POA are modifiable as well, both with respect to the set up and values of features, rendering scales of different sensitivity, and with respect to their mutual weight in their overall NAD. The set up can be influenced by actual auditory measurements in clusters. As to the question how much MOA matters (weighs) in relation to POA for the overall perceptual effect, the fact may be considered that within one manner sounds do not contrast well just by place. Compare, for example, stop clusters pt, pk or fricative clusters fq, sʃ vs. mixed manner clusters like stop þ fricative or stop þ approximant.9 This entails that MOA weighs more in a cluster. To decide about the relative degree of
9
Additionally, as a reviewer notices, stop clusters appear to be more frequent than frivatives clusters and certainly more frequent than sonorant clusters.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
17
influence of manner and place, a multifactorial statistical analysis for the weights might be performed. An additional issue of importance for the measurement of NAD is a necessity to include the place values corresponding to vowels. This will allow to derive more precise predictions from the universal phonotactic preferences. Finally, the assumptions made by the NAD Principle shall be discussed. Here, at least two issues arise. One concerns the juxtaposition of perceptual contrast and ease of articulation.10 There is an apparent conflict between the two. However, the auditory distance sustaining clusters is the result of ideal, phonemic, intentional articulation. Clusters reduce and change in casual speech in which assimilation happens much more often than dissimilation. The general tendency is to reduce clusters, however, marked clusters arise, too (cf. Dressler et al., 2001). Again, NAD Principle applies to motivate the outcome of those processes. There is no conflict since articulation always contributes to perception and a speaker is always simultaneously a listener. The second, essential issue concerns the prediction made by the NAD Principle. It predicts that the least marked (most preferred) clusters are those which maximally satisfy the phonotactic preferences. The question is whether these will be also the most frequent ones. Pietrala (2014) found that the most frequent clusters happen to be the ones which are neither strongly preferred or dispreferred. He also observed, however, that once the clusters of obstruent þ glide (w/j) and clusters of s þ stop are removed from the calculations, the frequent clusters are actually the preferred ones and the dispreferred clusters are very rare. These results call for the modifications of the measurement cues used in NAD and do not undermine the major prediction made by the NAD Principle. Nevertheless, in the present study, a majority of frequent clusters have turned out to be preferred according to NAD. This study has also shown that access to corpus data is indispensible in order to verify any phonotactic and morphonotactic hypothesis. References Basbøll, H., 2005. The Phonology of Danish. In: Phonology of the World’s Languages. Oxford Linguistics. Oxford University Press, Oxford. Bertinetto, P.M., Scheuer, S., Dziubalska-Kołaczyk, K., Agonigi, M., 2006. Intersegmental Cohesion and Syllable Division in Polish. Extended Version. Quaderni del Laboratorio di Linguistica 6. Scuola Normale Superiore, Pisa. http://linguistica.sns.it/QLL06.htm. Donegan, P.J., Stampe, D., 1979. The study of Natural Phonology. In: Dinnsen, D.A. (Ed.), Current Approaches to Phonological Theory. Indiana University Press, Bloomington, pp. 126–173. Dressler, W.U., 1996. Principles of naturalness in phonology and across components. In: Hurch, B., Rhodes, R.A. (Eds.), Natural Phonology: the State of the Art. Mouton de Gruyter, Berlin, pp. 41–52. Dressler, W.U., Dziubalska-Kolaczyk, K., Spina, R., 2001. Sources of markedness in language structures. Folia Linguist. Hist. 22 (1–2), 103–135. Dressler, W.U., Dziubalska-Kołaczyk, K., 2006. Proposing morphonotactics. Italian Journal of Linguistics 18 (2), 249–266. Dziubalska-Kołaczyk, K., 2002. Beats-and-Binding Phonology. Peter Lang, Frankfurt/Main. Dziubalska-Kołaczyk, K., 2009. NP extension: B&B phonotactics. PSiCL 45 (1), 55–71. Dziubalska-Kołaczyk, K., Krynicki, G., 2007. Phonotactic Preferences in Polish, English and German: Quantitative Perspective. Paper presented at the 38th Linguistic Meeting (PLM2007), 13–16 Sept 2007, Gniezno, Poland. Poznan , P., 2011. Classification of the Lexicon of modern polish according to the structure of consonant clusters. Dziubalska-Kołaczyk, K., Jankowski, M., Wierzchon In: Lee, W.-S., Zee, E. (Eds.), Proceedings of the 17th International Congress of Phonetic Sciences. 2011, Hong Kong, pp. 619–622. , P., Orzechowska, P., Zydorowicz, P., Jankowski, M., Pietrala, D., 2012. Explaining Morphonotactics: Phonological and Dziubalska-Kołaczyk, K., Wierzchon Linguistic Meeting, 2013. Morphological Constraints on Clusters. Poster presented at the 43rd Poznan Jassem, W., 2003. Polish. JIPA 33 (1), 103–108. Johnson, K., 2003. Acoustic and Auditory Phonetics. Blackwell, Oxford. Kehrein, W., Golston, C., 2004. A prosodic theory of laryngeal contrasts. Phonology 21 (3), 325–357. Ladefoged, P., 2006. A Course in Phonetics, fifth ed. Heinle & Heinle, Boston. Maddieson, I., 1999. In search of universals. In: Ohala, J.J., Hasegawa, Y., Ohala, M., Granville, D., Bailey, A.C. (Eds.), Proceedings of the 14th International Congress of Phonetic Sciences 1999, San Francisco, pp. 2521–2528. Maddieson, I., 2009. Calculating phonological complexity. In: Pellegrino, F., Marsico, E., Chitoran, I., Coupé, C. (Eds.), Approaches to Phonological Complexity. Mouton de Gruyter, Berlin, pp. 85–109. Ohala, J.J., 1990a. Alternatives to the sonority hierarchy for explaining segmental sequential constraints. The parasession on the syllable in phonetics and phonology. CLS 26 (2), 319–338. Ohala, J.J., 1990b. The phonetics and phonology of aspects of assimilation. In: Kingston, J., Beckman, M. (Eds.), Papers in Laboratory Phonology I. Cambridge University Press, Cambridge, pp. 258–275. Ohala, J.J., Kawasaki, H., 1984. Prosodic phonology and phonetics. Phonol. Yearb. 11, 113–129. Olender, A., 2013a. A Synchronic Investigation of Leftmost /s/ þ stops (M.A. thesis). Olender, A., 2013b. Acoustic Evidence for Word-initial /s/ þ stop Sequences as Onset Clusters: “Perceptual bond” as a Cross-Linguistic Predictor of Prothesis. Linguistic Meeting 2013. Poster presented at the 44th Poznan Pietrala, D., 2014. The Application of Linguistic Principles in Programming: Phonemic Transcription and Phonotactic Constraints (Unpublished Ph.D. dissertation). Adam Mickiewicz University in Poznan. Roach, P., 2001. English Phonetics and Phonology, third ed. Cambridge University Press, Cambridge. Rubin, E., 1915. Synsoplevede Figurer. Nordisk Forlag, Gyldendalske Boghande, Copenhagen and Christiania. Stampe, D., 1979. A Dissertation on Natural Phonology. IULC, Bloomington.
10
We do not know, in fact, whether sustaining a gesture or changing the gesture (modulation) is easier for the speaker (cf. wu, ji, etc.).
Contents lists available at ScienceDirect
Language Sciences journal homepage: www.elsevier.com/locate/langsci
Explaining phonotactics using NAD Katarzyna Dziubalska-Kołaczyk* , Poland University of Poznan
a r t i c l e i n f o
a b s t r a c t
Article history: Available online 27 July 2014
This paper presents a model of phonotactic grammar in which wellformedness of consonant clusters is measured by NAD. NAD stands for a Net Auditory Distance obtaining between segments in a cluster. The auditory distance is a net reflection of the differences between segments in terms of manner (MOA) and place of articulation (POA). It is calculated according to the Principle which states that a cluster is preferred if it satisfies a pattern of distances specified by the universal phonotactic preference relevant for its position in a word. Every position of a cluster in a word, i.e. initial, medial and final, is defined by a respective well-formedness (“goodness of cluster”) preference. The NAD Principle makes finer predictions than the sonority sequencing generalization (SSG). For example, it predicts that initial pr- is “better” (more preferred) that tr-, and they are both better than ps- or rt-, while the latter two are of comparable value. However, phonology alone does not fully account for clusters. Inflection, word-formation and compounding contribute to the creation of consonant clusters to an extent relative to a morphological type of a language. Therefore, a phonotactic grammar operates on basic, non-derived, lexical forms, while morphonotactics takes care of the remaining, morphologically complex, forms. Interaction between phonotactics and morphonotactics provides a richer insight into the understanding of cluster complexity. Ó 2014 Elsevier Ltd. All rights reserved.
Keywords: Phonotactics Net Auditory Distance Morphonotactics Beats and Binding Phonology Polish
1. Introduction The model of phonotactic grammar presented here is embedded in the theory of Beats and Binding Phonology (B&B henceforth) which is in turn derived from the principles and assumptions of Natural Phonology (NP) and employs the epistemology of Natural Linguistics (NL). B&B phonotactics is based on universal preferences which are grounded in phonetics and expressed by means of the Net Auditory Distance (NAD) Principle. NAD functions as a measure of wellformedness of lexical (i.e. phonological) clusters. In clusters arising in morphologically complex words NAD Principle is often overridden by morphonotactics. Having characterised the theoretical background, I will formulate the hypotheses with reference to the Polish data obtained in the project1 and discuss the results of the analysis. In conclusion, NAD parameters and the NAD Principle as such will be placed under scrutiny and directions for future research will be suggested.
Abbreviations: NAD, Net Auditory Distance; MOA, manner of articulation; POA, place of articulation; SSG, sonority sequencing generalisation; NP, Natural Phonology; NL, Natural Linguistics; B&B, Beats and Binding; OT, Optimality Theory. , Al. Niepodleglosci 4, Poznan , Poland. Tel.: þ48 519 340 568. * Faculty of English, Adam Mickiewicz University in Poznan E-mail address: [email protected]. 1 The new data from Polish (and English) obtained within a project Phonotactics and morphonotactics of Polish and English: description, tools and appli , Paulina Zydorowicz, Paula Orzechowska and Dawid cations (N N104382540). My collaborators in the project have been Michał Jankowski, Piotr Wierzchon Pietrala (cf. e.g., Dziubalska-Kołaczyk et al., 2011, 2012). http://dx.doi.org/10.1016/j.langsci.2014.06.003 0388-0001/Ó 2014 Elsevier Ltd. All rights reserved.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
7
2. Epistemology 2.1. Beats and Binding Phonology Beats and Binding Phonology (Dziubalska-Kołaczyk, 2002) is a syllable-less theory of phonology embedded in Natural Phonology. The structure usually referred to as “the syllable” in other models is epiphenomenal here or emergent due to principled phonotactic forces. The latter are responsible for different degrees of intersegmental cohesion (Bertinetto et al., 2006) which, in turn, determines the behaviour of segments and creates the impression of syllable structure. B&B Phonology was proposed out of a need for better, more comprehensive and holistic explanations of phonological phenomena than the ones prompted by the syllable.2 Basic units, beats (mostly vowels) and non-beats (always consonants, including glides), are connected into an alternating sequence by means of bindings according to the principle of perceptual salience. Such alternation is both acoustically and physically grounded: “To construct a useful signalling system out of sound, there must be some differentiation between different parts of the signal in time. It appears that a basic organization of this differentiation of sound in all (spoken) languages consists of an alternation between louder and quieter levels of sound, with a period not too far from 150-200 ms” (Maddieson,1999: 2525) This amounts to “[a] fairly regular wave-like alternation of amplitude peaks and valleys. The occurrence and timing of this pattern have been suggested to be related to a natural frequency of the jaw, which can be approximately equated with a comfortable mastication rate” (Maddieson, 1999: 2525) The alternating sequence is cut into pieces called words and morphemes on its way from the prelexical to lexical level. Clusters of non-beats are “accidents” due to some distortions of the alternating sequence (e.g., dropping or breaking of some beats). The occurrence and subsequent behaviour of consonant clusters is controlled by phonotactic preferences (and morphonotactics). Bindings are used in B&B Phonology to account for both segmental and prosodic structure. In this paper, however, the focus is on phonotactics. 2.2. Natural Phonology and Natural Linguistics The explanation pattern in B&B Phonology is epistemologically grounded in Natural Phonology (Stampe, 1979; Donegan and Stampe, 1979) and Natural Linguistics (Dressler, 1996 and other publications). Basic thesis of NP has been that phonological systems are phonetically motivated. The way to understand it is that processes and sounds have phonetic grounding in all speakers. It does not mean that the same processes will apply in all languages or that they will derive identical outputs across a variety of contexts. Sounds, apart from being contrastive, also have sub-categorical features. NL has an expanded epistemology in order to be able to account for all the components of language. It is a functionalist framework with semiotic underpinning. It is a preference theory which heavily relies on general non-linguistic principles as well as on sources of the so called “external” evidence. Language user is central to the explanation, since “preference” implies a human agent who behaves functionally in his/her linguistic performance. However, the goals of a given performance may be contradictory and the circumstances highly complex. Therefore, NL attempts a holistic explanation, expressed in terms of hierarchies of preferences according to complex sets of relevant criteria. The explanatory model of NL may be summarised as in the table below (Table 1). The starting point is a higher order nonlinguistic principle from which a linguistic preference is derived and measured by respective parameters, which in consequence provides an explanation for a given language-specific structure. 2.3. B&B phonotactics The model of phonotactic grammar with NAD is part of B&B Phonology. While beats and non-beats alternate on the basis of perceptual contrast (as discussed above in Section 2.1), actual auditory distances between sounds in a sequence become relevant when the melody (phonetic content) is filled in and we deal with vowels and consonants. Clusters of consonants tend to be avoided, subject to the universal CV preference. Typologically, the CV is a universal constituent type which occurs in all languages. As shown by Maddieson (2009) on the basis of a sample of 486 languages: 12.5% of the languages allow only CV’s 56.6% of the languages have the moderately complex structure CCVC, with limitations, however, on which consonants may appear in the CC cluster, the second consonant typically being a liquid or a glide 30.9% of the languages have complex structures (C)(C)(C)V(C)(C)(C)(C).
2 The syllable itself has always been problematic to define, while the various syllable-based explanations have often been supported by such circular notions as ‘extrasyllabicity’ and ‘ambisyllabicity’. For a comprehensive discussion of the syllable-related issues, including the history of approaches to the syllable, cf. Dziubalska-Kołaczyk (2002).
8
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17 Table 1 The explanatory model of Natural Linguistics. higher principles (e.g., the principle of the least effort, of cognitive economy) preferences (e.g., a preference for simple phonotactics, for a CV structure) preference parameters (pronunceability, perceptibility) consequences of preferences (absence of clusters in a language)
non-linguistic (cognitive, phonetic, psychological, sociological etc) linguistic functional and semiotic linguistic
The higher level principle which underlies the perceptual contrast in the CV is that of figure and ground (cf. the Rubin’s vase/ face illusion, discovered by Edgar Rubin, cf. Rubin, 1915). There is also a clear phonetic grounding for the CV: larger modulations have more survival value than lesser ones and therefore will persist in languages (Ohala, 1990a) “it is generally the case that the most salient acoustic modulations in a syllable occur near the CV interface”, and “auditory cues present in CV’s are more robust than those in VC’s” (Ohala and Kawasaki, 1984: 117, 118) “since there is a richer, more reliable set of place cues in the CV transition than the VC transition, listeners weight the former more heavily than the latter in deciding what they’ve heard” (Ohala, 1990b: 265)3 on the speaker’s side, according to Ohala and Kawasaki (1984: 119), “the speaker actively tries to create temporally more well defined, more precise, articulations near the CV as opposed to VC interface”. Since CV is a preferred phonological structure and clusters of consonants tend to be avoided across languages and in performance, there must be a phonological means to let them function in the lexicon relatively naturally. This is achieved by auditory contrast and its proper distribution across the word. Importantly, auditory impression is the overall product of articulation, mediated by acoustics. It is believed, therefore, that auditory (perceptual) distance can be expressed by respective combinations of articulatory features which eventually bring about the auditory effect. The underlying motivation behind a phonotactic preference is to counteract the CV preference, shown to be universal typologically and in performance (although clusters arise in performance, too). Any cluster in a structure which is more complex than CV is susceptible to change leading to CV, e.g., via cluster reduction (consonant deletion) CCV / CV or vowel epenthesis CCV / CVCV or at least vowel prothesis CCV / VCCV. A way to counteract this tendency is to introduce another salience into the structure, i.e. increase the perceptual distance between the consonants (CC of the CCV) to counterbalance the distance between the C and the V (CV of the CCV). In this way preferences which define “the survival” of complex structures in all positions of the word will be formulated. Besides the above markedness criterion, cluster size remains an obvious measure of cluster complexity: longer clusters are unanimously more complex than the shorter ones. Using the OT terminology of ranking, cluster size ranks higher as a constraint than the markedness criterion expressed by the auditory distance (see below for NAD). 3. NAD Principle and phonotactic preferences 3.1. NAD The Net Auditory Distance Principle (Dziubalska-Kołaczyk, 2009) defines cluster preferability in relation to the position in the word (initial, medial and final). It reads: A cluster is preferred if it satisfies a pattern of distances specified by the universal phonotactic preference relevant for its position in the word. NAD is a measure of distance between two neighbouring elements of a cluster in terms of differences in MOA (manner of articulation) and POA (place of articulation). The difference in voicing (Lx) has been considered, too, however, laryngeal features are non-redundant within subclasses of sounds only (e.g., they are non-redundant within obstruents and largely redundant within sonorants) and as such will have to be included in more refined, class-specific calculations4 (see Kehrein and Golston, 2004 and the criterion of spread glottis in Basbøll, 2005). In fact, an indefinite number of articulatory features as well as detailed acoustic cues would have to be investigated in terms of the degree of their contribution to the overall auditory effect obtained in a cluster (cf. e.g., Johnson, 2003).
3 4
A reviewer notices that there are exceptions, e.g., retroflexion is said to be more audible in VC. In fact, I’d like to introduce the Obstruent/Sonorant – O/S distinction instead.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
9
Table 2 Distances in MOA and POA.
Manner and place of articulation, however, constitute a rather reliable starting point since they form a backbone of any articulation, carrying responsibility both for perceptual clarity (manner, an articulatory correlate of sonority, of a dissimilatory nature) and production ease (place, of an assimilatory nature). The distances in terms of MOA and POA have originally been calculated on the basis of Table 2. The manners and places assumed in the table are selected according to their potential relevance: 6 manners (stop, affricate, fricative, sonorant stop, approximant, semivowel), where affricates and semivowels are, tentatively, attributed half a distance due to their ambiguous nature; and 5 places (labial, coronal, dorsal, radical and laryngeal or glottal). Manners refer to the most generally acknowledged version of the so-called sonority scale, while places are taken from Ladefoged (2006: 258). Both lists are extendible and modifiable (e.g., Ladefoged’s list consists of 5 nodes which branch into 12 more detailed features), depending on the amount of detail we want to include in the definition of distance. Importantly, vowels need to be differentiated according to their colour, too, which will allow to reflect the intersegmental distances with a higher degree of precision. Within the project quoted here (cf. footnote 2 above), separate tables for Polish and English consonants5 have been used, reflecting the differences between the two systems as well as including more detailed MOA and POA scales (see Tables 3 and 4). It remains to be verified in further research whether the proposed scaling actually accounts for the data more accurately (see also the discussion in Section 7). 3.2. Preferences The NAD Principle evaluates cluster markedness with reference to universal phonotactic preferences. The preferences, describing two- and three-consonant clusters in word-initial, medial and final position, are listed below. An example of a calculation is provided for the first preference. 3.2.1. C1C2V NAD (C1,C2) NAD (C2,V) Calculation: NAD CC ¼ j(MOA1 MOA2)j þ j(POA1 POA2)j
5
Polish consonants based on Jassem (IPA, 2003), English consonants based on Roach (BrE, IPA, 2001).
10
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17 Table 3 Distances in MOA and POA: Polish.
Table 4 Distances in MOA and POA: English.
NAD CV ¼ jMOA1 MOA2j e.g., prV in Polish pr: j(MOA1 MOA2)j þ j(POA1 POA2)j j5 2j þ j1 2.3j ¼ j3j þ j1.3j ¼ 4.3 so, NAD CC ¼ 4.3 rV: jMOA1 MOA2j ¼ j2 0j ¼ 2 NAD CV ¼ 2 so, NAD CC NAD CV ¼ 2.3 so, the preference NAD (C1,C2) NAD (C2,V) is observed since 4.3. > 2.3 3.2.2. VC1C2 NAD (V,C1) NAD (C1,C2) 3.2.3. V1C1C2V2 NAD (V1,C1) NAD (C1,C2) < NAD (C2,V2) *NAD (C1,C2) > 0
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
11
Fig. 1. The phonotactic calculator.
3.2.4. C1C2C3V NAD (C1,C2) < NAD (C2,C3) NAD (C3,V) 3.2.5. VC1C2C3 NAD (V,C1) NAD (C1,C2) > NAD (C2,C3) 3.2.6. V1C1C2C3V2 NAD (V,C1) NAD (C1,C2) & NAD (C2,C3) < (C3,V2) 3.3. The phonotactic calculator6 In order to automatically calculate NAD, a phonotactic calculator has been developed (Dziubalska-Kołaczyk and Krynicki, 2007; Dziubalska-Kołaczyk and Pietrala, Phonotactics and morphonotactics project, Pietrala, 2014). It allows for statistical analysis of phonetic dictionaries and phonetically annotated corpora from various languages. It works on various lengths of clusters in all word positions and estimates them with respect to the universal phonotactic preferences (Dziubalska-Kołaczyk, 2002; cf. above Section 3.2). It provides fast feedback on the predictability value of the hypotheses expressed through the preferences. The calculating platform opens as shown below in Fig. 1. Fig. 2 shows an example of NAD calculation, for the most frequent 14 two-consonant initial clusters of Polish. Fig. 3 illustrates the graphic representation of the results of the above calculation generated by the calculator. As can be observed, 10 out of 14 most frequent Polish initial clusters of two consonants have a positive value of NAD, i.e. they are unmarked according to the NAD Principle. 4. Morphonotactics Phonology alone does not fully account for clusters. Inflection, word-formation and compounding contribute to the creation of consonant clusters to an extent relative to a morphological type of a language. A phonotactic grammar operates on basic, non-derived, lexical forms. Morphotactic forms need a wider account, however. As we suggested in Dressler and Dziubalska-Kołaczyk (2006), morphonotactics is the area of interaction between morphotactics and phonotactics and provides a richer insight into the understanding of cluster complexity. Language specific morphonotactics provides thus an additional parameter constraining the actual outcome of universal phonotactic preferences.
6 The phonotactic calculator will be made available online for research purposes. It will be open to modifications, including the criteria considered as well as the values assigned to them.
12
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
Fig. 2. The phonotactic calculator at work.
Fig. 3. The phonotactic calculator: a NAD graph.
A need to signal a morphological boundary may override a phonologically driven phonotactic preference and, consequently, lead to the creation of a marked cluster. Therefore, one expects relatively marked clusters across morpheme boundaries and relatively unmarked ones within morphemes. In view of the above discussion, two types of clusters have been distinguished: phonotactic (lexical) clusters, as in band, past and morphonotactic clusters, as in ban(n)þed, passþed. The following sources of morphonotactic clusters have been considered: concatenative, e.g., Polish, /ft-/ in w þ toczyc ‘to roll into’, z þ robic ‘do’, s þ pasc ‘fall’ (perfective vs. imperfective), English /-fs, -vz/ in laughs, loves, wife’s, wives non-concatenative, e.g., Polish len vs. lnu ‘linen’ (nom.sg. vs. gen.sg.), przeste˛ pstwo w przeste˛ pstw [-mpstf] ‘crime’ (nom.sg. vs. gen.pl.), tratwa w tratw [-tf] ‘raft’ (nom.sg. vs. gen. pl.) miejsce vs. miejsc ‘place’ (nom.sg. vs. gen.pl.), punkt vs. punkcie ‘point’ (nom.sg. vs. loc.).
5. Hypotheses and data The phonotactic and morphonotactic assumptions lead to the following hypotheses concerning clusters.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
13
Table 5 The most complex clusters from the corpus. Cluster length LEX MORPH Examples 6
–
100%
5
–
100%
4
5%
95%
przeste˛ pstwie, wewna˛ trzwspólnotowy, wewna˛ trzzwia˛ zkowy nstwie, bezwzgle˛ dnie, rozbrzmiewa, kontrprzykład, postzwia˛ zkowy, tysia˛ cstronicowy dzieci /rstf/ warstwa (lex 3%), ministerstwo (-stwo morph 97%), krwia˛ (lex), krna˛ brny (lex), zbło (lex), roztkliwiac (morph), zd odstrzał (morph), je˛ drny (lex)
Fig. 4. Word-initial clusters from the dictionary: cluster types and words.
Hypothesis 1: Cluster size corresponds to morphological complexity. The longer a cluster, the more likely it is to be morphonotactic. Hypothesis 2: The degree of phonological preferability is inversely proportional to morphological complexity. Morphonotactic clusters are expected to have a lower degree of preferability than phonotactic ones. Hypothesis 3: The degree of cluster preferability is directly proportional to frequency. Preferred clusters are expected to be more frequent than dispreferred clusters. The hypotheses have been verified against the Polish data of three types: a dictionary of 8.000 morphologically parsed lemmas, a paradigm list of 190.000 inflected forms with morphological parsing of 5 most frequent initial and final clusters and a corpus of 500.000 inflected forms as a reference for the frequency of use. 6. Data analysis and discussion Unsurprisingly, Hypothesis 1 has been fully confirmed. The most complex clusters from the corpus, those consisting of 5 or 6 consonants, are all exclusively morphonotactic, those consisting of 4 consonants, in 95 percent of the cases (see Table 5). The dictionary data had been analysed with respect to the clusters of 2 and 3 consonants for final clusters, and 2, 3 and 4 consonants for the initial ones. Fig. 4 shows the percentage of lexical (phonotactic) and morphological (morphonotactic) clusters in word initial position with respect to cluster types and the number of words. It is clear that while the number of lexical clusters decreases with cluster size, the opposite is true for morphonotactic ones: 70 percent of 2-consonant clusters are phonotactic while 70 percent of the 4-consonant long clusters are morphologically complex. The same is true for the word final clusters: 90 percent of 2-consonant clusters are phonotactic while –CCC clusters tend to be triggered by morphology especially in word occurrences (see Fig. 5). Hypothesis 2 has been corroborated with respect to morphonotactic clusters which have a strong tendency to be marked (dispreferred according to NAD), as shown in Figs. 6 and 7 (P stands for ‘preferred’, D for ‘dispreferred’). However, one would expect phonotactic clusters to have a higher degree of preferability in comparison to the morphonotactic ones. This remains to be accounted for. Hypothesis 3 concerned the relationship between cluster markedness and frequency. Paradigm and corpus results have been compared to those from the dictionary. The hypothesis has been only partially corroborated. In the paradigm and the corpus, 5 most frequent initial CC clusters are pʂ7, pr, st, mj, sp (of which pr and mj are unmarked) and 5 most frequent final CC clusters are st, ɕṭɕ, nt, nʦ, ŋk (of which nt and ŋk are preferred). The relatively marked clusters are partially accounted for by morphology, especially the initial sp and most of the finals, as shown in Figs. 8 and 9.
7 The symbol ʃ in the data for Polish should be interpreted as a phonetic ʂ. The same symbol (ʃ) was used in our Polish and English data for the sake of simplicity.
14
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
Fig. 5. Word-final clusters from the dictionary: cluster types and words.
Fig. 6. Word initial CC- clusters from the dictionary: types and words.
Fig. 7. Word final -CC clusters from the dictionary: types and words.
Additionally, pʂ is the initial cluster of three highly productive prefixes, przed, przy and prze. Five most frequent clusters from the dictionary diverge from the paradigm/corpus ones: initial mj and final nʦ, ŋk are not found in the five. Five most frequent initial CC clusters are pʂ, pr, st, sp, pj and 5 most frequent final CC clusters are ɲṭɕ, ɕṭɕ, nt, zm, st. The unmarked clusters are clearly lexical, while the dispreferred ones are mixed (see Fig. 10). In order to get a wider perspective on Hypothesis 3, 14 most frequent initial clusters of two consonants were extracted from the dictionary, which included altogether 238 initial cluster types (Fig. 11). Out of the fourteen, only four clusters are marked: pʂ and the three s þ stop clusters (see Fig. 12). While the frequent occurrence of pʂ receives a morphological account, s þ stop clusters need a separate justification (cf. Section 7 for the discussion). Otherwise, NAD successfully accounts for the frequency. 7. NAD parameters and assumptions: discussion and conclusions The study has firmly established that longer clusters tend to be morphologically complex and morphonotactic clusters tend to be phonologically dispreferred (marked). This means that morphonotactics indeed helps to explain
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
15
Fig. 8. Frequent initial clusters in the paradigm (p) and in the corpus (c).
Fig. 9. Frequent final clusters in the paradigm (p) and in the corpus (c).
Fig. 10. Frequent initial and final clusters in the dictionary.
cluster complexity.8 Phonological explanations based on the NAD Principle, however, apparently do not cover all cases and therefore require modification or extension. Below I will discuss the issues to be resolved in further research. Firstly, there are dispreferred (marked according to NAD) lexical clusters. In the Polish data these were, e.g., pʂ: this is the type of a cluster in the case of which an extra-phonological explanation is called for, since it occurs within highly productive affixes, e.g., in przed-, przy-, prze-. Thus, morphological productivity is one source of explanation for the presence of phonologically marked clusters in the lexicon.
8 Morphology has a semiotic priority over phonology and motivates meaning. A salient, phonologically marked cluster performs a morphological function better than a phonologically preferred, natural one. Thus, morphonotactics contributes independent predictions about the shape of clusters.
16
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
Fig. 11. Total distribution of 238 cluster types in the dictionary.
Fig. 12. Fourteen most frequent initial clusters from the dictionary.
Secondly, the notorius s þ stop clusters, which are both lexical and morphonotactic, are dispreferred by NAD. These have been approached in a variety of ways (see Olender, 2013a, b for an extensive overview) none of which seems to have given a satisfactory general account of their omnipresence across the languages of the world. The closest might have been Basbøll’s SSM model (Sonority Syllable Model, cf. Basbøll, 2005) which refers to a universal logic of segment types (sonorants vs. obstruents, all sonorants are voiced, not all voiced sounds are sonorants, etc.). This leads him to use the feature [spread glottis] to account for apparent violations of the sonority hierarchy. In particular, the SSM predicts that marginal segments in absolute initial and final position have [spread glottis], e.g., st-, ts-; -st, -ts (which agrees with final devoicing e.g., in Fr. –fl). However, forcing a general allembracing account for all cluster types may not be the most effective approach. The alternative would be detailed phonetic or phonological accounts for sub-types. Olender (2013a, b), for instance, proposes that leftmost s þ stop clusters should be treated as highly cohesive sibilant þ stop class clusters bound by a “unique perceptual bond”. This bond is highly sensitive to breaking: even the minimal vocalic intrusion (epenthesis) immediately results in the loss of a cluster in perception. Such well-informed phonetics may very well explain the preference for s þ stop clusters without reference to any phonological theory. A phonological theory can also make allowances for sub-types, or, alternatively, provide general as well as languagespecific accounts. This brings us to the third point in the discussion of NAD: the scope of NAD. On the one hand, we may provide descriptions of phonotactics of particular languages based on language-specific parameter values; on the other hand, the same (universal) parameter values guarantee cross-language comparability. Another related question concerns the number and sensitivity of the NAD parameters as well as their relative weight within a cluster. In the present version of NAD I have moved away from including the feature [voice] in the calculation of distances since the laryngeal gesture on its own seems not to distinguish consonants in a cluster. Besides, it is not functional within the class of sonorants. Basbøll’s proposal of [spread glottis] may be considered instead. The parameters of MOA and POA are modifiable as well, both with respect to the set up and values of features, rendering scales of different sensitivity, and with respect to their mutual weight in their overall NAD. The set up can be influenced by actual auditory measurements in clusters. As to the question how much MOA matters (weighs) in relation to POA for the overall perceptual effect, the fact may be considered that within one manner sounds do not contrast well just by place. Compare, for example, stop clusters pt, pk or fricative clusters fq, sʃ vs. mixed manner clusters like stop þ fricative or stop þ approximant.9 This entails that MOA weighs more in a cluster. To decide about the relative degree of
9
Additionally, as a reviewer notices, stop clusters appear to be more frequent than frivatives clusters and certainly more frequent than sonorant clusters.
K. Dziubalska-Kołaczyk / Language Sciences 46 (2014) 6–17
17
influence of manner and place, a multifactorial statistical analysis for the weights might be performed. An additional issue of importance for the measurement of NAD is a necessity to include the place values corresponding to vowels. This will allow to derive more precise predictions from the universal phonotactic preferences. Finally, the assumptions made by the NAD Principle shall be discussed. Here, at least two issues arise. One concerns the juxtaposition of perceptual contrast and ease of articulation.10 There is an apparent conflict between the two. However, the auditory distance sustaining clusters is the result of ideal, phonemic, intentional articulation. Clusters reduce and change in casual speech in which assimilation happens much more often than dissimilation. The general tendency is to reduce clusters, however, marked clusters arise, too (cf. Dressler et al., 2001). Again, NAD Principle applies to motivate the outcome of those processes. There is no conflict since articulation always contributes to perception and a speaker is always simultaneously a listener. The second, essential issue concerns the prediction made by the NAD Principle. It predicts that the least marked (most preferred) clusters are those which maximally satisfy the phonotactic preferences. The question is whether these will be also the most frequent ones. Pietrala (2014) found that the most frequent clusters happen to be the ones which are neither strongly preferred or dispreferred. He also observed, however, that once the clusters of obstruent þ glide (w/j) and clusters of s þ stop are removed from the calculations, the frequent clusters are actually the preferred ones and the dispreferred clusters are very rare. These results call for the modifications of the measurement cues used in NAD and do not undermine the major prediction made by the NAD Principle. Nevertheless, in the present study, a majority of frequent clusters have turned out to be preferred according to NAD. This study has also shown that access to corpus data is indispensible in order to verify any phonotactic and morphonotactic hypothesis. References Basbøll, H., 2005. The Phonology of Danish. In: Phonology of the World’s Languages. Oxford Linguistics. Oxford University Press, Oxford. Bertinetto, P.M., Scheuer, S., Dziubalska-Kołaczyk, K., Agonigi, M., 2006. Intersegmental Cohesion and Syllable Division in Polish. Extended Version. Quaderni del Laboratorio di Linguistica 6. Scuola Normale Superiore, Pisa. http://linguistica.sns.it/QLL06.htm. Donegan, P.J., Stampe, D., 1979. The study of Natural Phonology. In: Dinnsen, D.A. (Ed.), Current Approaches to Phonological Theory. Indiana University Press, Bloomington, pp. 126–173. Dressler, W.U., 1996. Principles of naturalness in phonology and across components. In: Hurch, B., Rhodes, R.A. (Eds.), Natural Phonology: the State of the Art. Mouton de Gruyter, Berlin, pp. 41–52. Dressler, W.U., Dziubalska-Kolaczyk, K., Spina, R., 2001. Sources of markedness in language structures. Folia Linguist. Hist. 22 (1–2), 103–135. Dressler, W.U., Dziubalska-Kołaczyk, K., 2006. Proposing morphonotactics. Italian Journal of Linguistics 18 (2), 249–266. Dziubalska-Kołaczyk, K., 2002. Beats-and-Binding Phonology. Peter Lang, Frankfurt/Main. Dziubalska-Kołaczyk, K., 2009. NP extension: B&B phonotactics. PSiCL 45 (1), 55–71. Dziubalska-Kołaczyk, K., Krynicki, G., 2007. Phonotactic Preferences in Polish, English and German: Quantitative Perspective. Paper presented at the 38th Linguistic Meeting (PLM2007), 13–16 Sept 2007, Gniezno, Poland. Poznan , P., 2011. Classification of the Lexicon of modern polish according to the structure of consonant clusters. Dziubalska-Kołaczyk, K., Jankowski, M., Wierzchon In: Lee, W.-S., Zee, E. (Eds.), Proceedings of the 17th International Congress of Phonetic Sciences. 2011, Hong Kong, pp. 619–622. , P., Orzechowska, P., Zydorowicz, P., Jankowski, M., Pietrala, D., 2012. Explaining Morphonotactics: Phonological and Dziubalska-Kołaczyk, K., Wierzchon Linguistic Meeting, 2013. Morphological Constraints on Clusters. Poster presented at the 43rd Poznan Jassem, W., 2003. Polish. JIPA 33 (1), 103–108. Johnson, K., 2003. Acoustic and Auditory Phonetics. Blackwell, Oxford. Kehrein, W., Golston, C., 2004. A prosodic theory of laryngeal contrasts. Phonology 21 (3), 325–357. Ladefoged, P., 2006. A Course in Phonetics, fifth ed. Heinle & Heinle, Boston. Maddieson, I., 1999. In search of universals. In: Ohala, J.J., Hasegawa, Y., Ohala, M., Granville, D., Bailey, A.C. (Eds.), Proceedings of the 14th International Congress of Phonetic Sciences 1999, San Francisco, pp. 2521–2528. Maddieson, I., 2009. Calculating phonological complexity. In: Pellegrino, F., Marsico, E., Chitoran, I., Coupé, C. (Eds.), Approaches to Phonological Complexity. Mouton de Gruyter, Berlin, pp. 85–109. Ohala, J.J., 1990a. Alternatives to the sonority hierarchy for explaining segmental sequential constraints. The parasession on the syllable in phonetics and phonology. CLS 26 (2), 319–338. Ohala, J.J., 1990b. The phonetics and phonology of aspects of assimilation. In: Kingston, J., Beckman, M. (Eds.), Papers in Laboratory Phonology I. Cambridge University Press, Cambridge, pp. 258–275. Ohala, J.J., Kawasaki, H., 1984. Prosodic phonology and phonetics. Phonol. Yearb. 11, 113–129. Olender, A., 2013a. A Synchronic Investigation of Leftmost /s/ þ stops (M.A. thesis). Olender, A., 2013b. Acoustic Evidence for Word-initial /s/ þ stop Sequences as Onset Clusters: “Perceptual bond” as a Cross-Linguistic Predictor of Prothesis. Linguistic Meeting 2013. Poster presented at the 44th Poznan Pietrala, D., 2014. The Application of Linguistic Principles in Programming: Phonemic Transcription and Phonotactic Constraints (Unpublished Ph.D. dissertation). Adam Mickiewicz University in Poznan. Roach, P., 2001. English Phonetics and Phonology, third ed. Cambridge University Press, Cambridge. Rubin, E., 1915. Synsoplevede Figurer. Nordisk Forlag, Gyldendalske Boghande, Copenhagen and Christiania. Stampe, D., 1979. A Dissertation on Natural Phonology. IULC, Bloomington.
10
We do not know, in fact, whether sustaining a gesture or changing the gesture (modulation) is easier for the speaker (cf. wu, ji, etc.).