2.1 Phonological primes

The approach described in Nasukawa (2014; 2015ab) (cf. Nasukawa 2011; 2012) denies the existence of precedence relations between units of phonological representation, eliminating not only units such as CV units, skeletal positions and Root nodes (which have been assumed to carry properties relating to precedence) but also traditional prosodic units such as onsets, nuclei and codas (although these may still be informally referred to for the sake of ease of understanding). Instead, features are regarded as the units that play a central role in building phonological structure. This contrasts well with orthodox phonological models where features are merely inherent attributes of a segmental position and segments (more precisely, CV units or skeletal positions) are treated as basic building blocks for constructing phonological structure. In this model features take the place of prosodic constituents like onset and nucleus, since features (which are, in phonological terms, the smallest units) themselves function as the basic building blocks of phonological structure. At the same time, a feature may also function as the head of a ‘nuclear’ expression, and by adding another feature to this head feature a complex expression is formed in which the additional feature takes the role of a dependent/complement. The feature model which most clearly illustrates this approach is the version of Element-based feature theory developed by Nasukawa (2012; 2014), in which each feature or elements is monovalent and fully interpretable on its own – to be phonetically realised it does not require the support of other elements. It follows that there is neither any universally fixed matrix of features nor any template-like feature organization. In accordance with certain principles, features can combine freely with one another.

In element-based feature theory, melodic structure is represented using the six monovalent elements |A I U ʔ H N|. These are to be understood as mental objects which are active in all languages. Conceived of within the perception-based view of melodic structure employed in the work of Jakobson (Jakobson et al. 1952; Jakobson & Halle 1956), elements map onto the phonetic exponent. The six elements are described in Table 1, along with their typical acoustic signature (Harris & Lindsey 2000; Harris 2005).

In principle, the elements may be employed in both consonant and vowel expressions. Table 2 shows the different phonetic categories associated with each element according to whether it appears in a consonant or a vowel (Nasukawa & Backley 2008; Nasukawa 2014: 3).

The first three elements |A I U| form a natural group of ‘resonance’ elements which typically describe vowel quality, prosodic phenomena such as pitch and intonation patterns, and also place of articulation (POA) in consonants. The other three elements |ʔ H N| are associated with non-resonance properties such as occlusion, aperiodicity and laryngeal-source effects.

2.2 Representing vowels in Japanese

Nasukawa (2014) claims that what is traditionally assumed to be a nucleus is replaced by one of the three resonance elements |A|, |I| or |U|, this language-specific choice determining the phonetic quality of a melodically empty nucleus in the given language. (Traditionally, it is assumed that an empty nucleus is pronounced as one of the central vowels ə, i(ɨ) or ɯ, according to parametric choice.) English selects |A|, which is realised as ə in its acoustically weak form, while Yoruba chooses |I| (realised as i) and Japanese selects |U| (realised as ɯ in the east part of Japan, as u in the west) Figure 1.

Figure 1

Weak vowels.

Thus languages divide into three types according to their baseline resonance: |A|-type (ə), |I|-type (i) and |U|-type (ɯ) (Nasukawa 2014: 13).

Given that the weak vocalic forms ə, i and ɯ are each represented by a single element |A|, |I| and |U| respectively, the question arises as to how the near-universal corner vowels a, i and u are represented structurally. In the case of |A|-type languages such as English, the baseline (which functions as a nucleus/V) takes another element as its dependent. If the baseline and |I| are concatenated, the whole expression is phonetically realised as i, and if the baseline and |U| form a set, the expression manifests itself as u. Furthermore, the set which consists of the baseline and |A| is phonetically interpreted as a. These structures may be represented as follows.

The leftmost structure in Figure 2 shows the representation of the English baseline, a sole |A|, which determines the quality of unstressed vowels and of the default epenthetic vowel, both of which are phonetically manifested as ə.³ On the other hand, the baseline resonance may also have the acoustic pattern of an additional (dependent) element superimposed on to it: for example, in the structures for a, i and u respectively, the dependents |A|, |I| and |U| have acoustic patterns with greater prominence than those of their baseline. These phonetic values a, i and u are the exaggerated forms of ə, i and ɯ respectively (where ə, i and ɯ are to be understood as the phonetic interpretation of |A|, |I| and |U| as bare heads) (Nasukawa 2014: 11–12). Following notational conventions, the head occupies the position at the top of the tree diagram and labels the entire structure. The corresponding relations between structural head-dependency and phonetic prominence are described and developed in Nasukawa & Backley (2015): heads are important and unmarked for structure-building but they lack phonetic prominence, whereas dependents are unimportant for structure-building but are phonetically more prominent. The same situation is found in other modules of the grammar, including syntax, where the default stress pattern in the verb phrase [kissed Mary] of [John [kissed Mary]_VP] shows that the dependent (complement) of the phrase [Mary] is phonetically more prominent than the head [kissed]. For a detailed discussion, the reader is referred to Nasukawa & Backley (2015).

Figure 2

Vowels in the |A|-type language.

The same configuration also applies to |I|-type and |U|-type languages. One example is the |U|-type language Japanese, which will be discussed in the latter half of this paper. In the case of Japanese, |U| is the baseline (head), which is phonetically realised as the unrounded vowel ɯ when there is no dependent element.

When the baseline takes |A|, |I| or |U| as a dependent,⁴ the acoustic pattern (phonetic exponence) of this dependent element overrides that of the baseline. As a result, the structures are phonetically realised as a, i, and ɯ respectively,⁵ as shown in Figure 3. The remaining two vowels of Japanese, e and o, are represented as follows.

Figure 3

Vowels in the |U|-type language.

In the domain marked out with a dotted line, the set where |I| (solely phonetically interpreted as i) has |A| (interpreted as a) as a dependent is phonetically realised as the mid front vowel e: in acoustic terms, the additional (dependent) ‘mass’ pattern is added to the (structurally headed) ‘dip’ pattern. In this configuration, the dependent ‘mass’ pattern is more prominent than the head ‘dip’ pattern since |A| is the most embedded dependent, making it phonetically more prominent than the head. The same is true in the structure for o in Figure 4: in the |U|-headed set of |U| and |A|, the dependent |A| is acoustically more prominent than the head |U|.

Structures which are the reverse of those in Figure 4 are also employed in Japanese, as given in Figure 5: the |A|-headed set consisting of |A| and |I| is phonetically interpreted as the light diphthong ja (ĭa) rather than as a monophthong, while the |A|-headed set consisting of |A| and |U| phonetically manifest itself as the light diphthong ɰa (ɯ̆a).

Figure 4

Mid vowels e and o.

Figure 5

Light diphthongs jɑ (ĭɑ) and ɰ̆ɑ (ɯ̆ɑ).

The remaining light diphthongs permitted in Japanese are represented as follows.

Along the same lines, Figure 6 shows how the |U|-headed set consisting of |U| and |I| is phonetically interpreted as the light diphthong ju ((ĭu), while the whole structure is realised as the light diphthong jo ((ĭo) in which the |A|-headed set comprising |A| and |I| (phonetically interpreted as ja) is embedded in the dependent |A| part of the |U|-headed set consisting of |U| and |A| (phonetically interpreted as o).

Figure 6

Light diphthongs ju (ĭu) and jo (ĭo).

As discussed in Nasukawa (2015a), the above structures find support in the observation that jV of CjV (rather than Cj) behave as constituents in phonological phenomena, as demonstrated below (where phonetic symbols in the brackets are phonetically realised forms).

The pattern emerging from (2) is that a front vowel cannot follow a Cj sequence in Japanese. This is often taken to be a co-occurrence restriction which bans a sequence comprising the palatal glide j and a front (palatal) vowel (i/e). Yet in fact, not only CjV sequences but also jV sequences are subject to the same distributional restriction.

Given that the co-occurrence restriction works within a domain/constituent, as demonstrated by consonant clusters and diphthongs cross-linguistically, it follows that a CjV sequence must be syllabified as C-jV, where j is part of the nucleus rather than part of the onset (Cj-V). This is motivated by the fact that any consonant in the Japanese consonant inventory (except for j, w and the placeless nasal ɴ) may appear before a permitted jC sequence, i.e. the same distributional freedom as a single consonant that precedes any of the five monophthong vowels a, i, u, e, o. To capture this distributional restriction involving jV sequences, Nasukawa (2015a) claims that jV as a whole forms a nucleus rather than a CV sequence. That is, jV is a light diphthong (ĭa) of the kind which is also found in languages such as Korean and Chinese.

This view is also supported by the way these sounds are written in the Japanese syllabary, where kja (きゃ) is represented as a combination of き(ki) and a subscriptゃ(ja): i.e. ki is modified by the addition of ja.

2.3 Representing consonants in Japanese

Before proceeding to the analysis of Japanese palatalisation in §3, let us clarify how consonants are represented in the precedence-free model. It is assumed that consonants are structurally dependent on vowels, since vowels are generally taken to be obligatory in constituents such as ‘syllable’ and ‘word’ whereas consonants are optional. From this it follows that the vocalic part of the constituent forms its head (and is therefore unmarked and essential for structure-building), while the consonantal part takes the role of a dependent (and is therefore unimportant for structure-building).

On this basis – and in light of the above discussion on the relation between structural head-dependency and phonetic prominence – it may be claimed that consonants are more prominent than vowels since consonantal properties tend to function as phonetic cues to prosodic information (e.g. English aspiration as a marker of foot-initial position) while vowels have no comparable function (e.g. despite being more sonorous than consonants, vocalic properties are unmarked and do not show any acoustically-defined abrupt changes). This is consistent with the point made in §2.2 that heads lack phonetic prominence while dependents are phonetically more prominent.

Let us return to the argument that the part of a constituent which is phonetically more prominent and/or contrastively richer should occupy a more deeply embedded position. We may represent the consonantal part using the structure in Figure 7, where elements under a vertical line are heads and those under a slanting line are dependents.

Figure 7

The structure of ta ‘rice field’.

As illustrated above, the consonantal part is dominated by the vocalic part: in the left-hand structure in Figure 7, the consonantal |H|-headed set of three elements (which phonetically manifests itself as t) is dependent on the baseline |U| that is the ultimate head of the expression. And the |U|-headed set of |H|˝ and |U| (= |U|ˊ) takes |A| as its dependent at the next level down. As discussed in §2.2, the part consisting of |U| and |A| is phonetically interpreted as a since the head |U| is a resonance baseline, the acoustic quality of which is masked by that of its dependent element. As a whole, the structure on the right-hand side is realised phonetically as ta ‘rice field’.

As mentioned earlier, and as discussed in Nasukawa (2011; 2014; 2015a; b) and Nasukawa & Backley (2008), representations of this kind make no reference to precedence relations between the units within phonological representations. There is therefore no difference between the two structures in Figure 7: both exhibit the same dependency relations between the units in their respective structures. In this model, as argued in Nasukawa (2011) who discusses in detail two types of dependency (endocentric dependency and exocentric dependency), linear precedence is to be regarded as the natural result of performance systems interpreting the hierarchical structure present in phonological representations.

Referring to the configurations in Figure 7, the element structures permitted to appear in the consonantal part⁶ are given below.

Since the noise element |H| is present in all obstruents, it serves to define the class of obstruents. (Conversely, the absence of |H| indicates a sonorant expression.) |H| is deemed the head of the consonantal expression in which it appears, while the nature of the hierarchical relation whereby |I|, |U| or |A| is dominated by |H| determines any acoustic effects relating to place of articulation. In addition, a whole expression is identified as a stop or an affricate if the edge element |ʔ| is present, as in Figure 8, while the same expression without |ʔ| is interpreted as a fricative, as shown in Figure 9.

Figure 8

Stops and affricates in Japanese.

Figure 9

Fricatives in Japanese.

Let us limit the present discussion to palatality (in representational terms, the property associated with the |I| element), since this will be the focus of §3. The element |I| is found in ʨ/ʥ (Figure 8), in ɕ (Figure 9) and in ç (also Figure 9); in all of these, |I| is in the most deeply embedded part of the structure, and for this reason, is interpreted as palatality.

Using the melodic structures just outlined, the next section describes two seemingly opposing phenomena involving palatality: palatal dissimilation and palatal assimilation.

3.1 Palatal dissimilation

The process of palatal dissimilation in Japanese (see §2.2) imposes a ban on sequences of a palatal glide j followed by a front (palatal) vowel (i/e).

The prohibited sequences *ji and *je are instead produced as i and e respectively (e.g. idiɕɕu < jɪdɪʃ ‘Yiddish’ and eritsiɴ < jeltsin ‘Yeltsin (Boris)’). This process is typically seen as a co-occurrence restriction, which makes appeal to the OCP (Obligatory Contour Principle) or Identity Avoidance since it disallows sequences of j plus i/e (Nasukawa 2015b; cf. Yip 1988; 1998). In terms of element structure, Japanese *ji and *je are represented as follows.

Recall that the structural part containing the vocalic set has one of the three elements |I|, |U| or |A| as its head, and that in the case of Japanese it is |U| which dominates and provides the baseline for the entire structure. In this vocalic part, only a single |I| element can appear (*|I I|: Nasukawa 2015b). Thus, by suppressing an |I| element which is more deeply embedded, as shown on the left in Figure 10, we arrive at a structure identical to that in the second from the right in Figure 3; this resulting structure is phonetically interpreted as i. The same applies to the right-hand structure in Figure 10: suppressing the most deeply embedded |I| leaves an expression which is interpreted as e, the same as in the left-hand structure in Figure 4.

Figure 10

The prohibited sequences *ji (ĭi) and *je (ĭe).

In loanword phonology, on the other hand, *je is occasionally accommodated as ie rather than e as in ieti (*eti) < jeti ‘Yeti’ and iereɴ (*ereɴ) < jelən ‘Yellen (Janet)’. The unpacking of je to ie may be analysed as follows.

Rather than by suppressing the most deeply embedded element |I| as in the right-hand structure in Figure 10, the structure for ie is generated by breaking the input structure at the highest level and placing the most deeply embedded |I| (as in Figure 11, left-hand side) in the dependent position of |U|_Dep, which is the first dependent of |U|_Head.

Figure 11

The unpacking of je (ĭe) to ie.

In addition to the alternations *je > i and *je > ie, je (unlike *ji) is occasionally allowed in recent loanwords (e.g. jesu < jes ‘yes’ and jeroo < jeləʊ ‘yellow’). On the other hand, *ji is disallowed in all word types including loanwords (ijaa < jɪə ‘year’ and idiɕɕu < jɪdɪʃ ‘Yiddish’). Under the proposed representations in Figure 10, the difference between *ji and *je is attributed to the presence/absence of |A|: the structure which consists of only |I|s in the domain in question (the case of ji in Figure 10, left-hand side) is strictly prohibited by the requirement of Identity Avoidance *|I I|. On the other hand, the structure which contains |A| in addition to the two |I|s in the domain in question (the case of je in Figure 10, right-hand side and Figure 11 may be interpreted depending on various factors such as donor language and word frequency: in some recent loanwords, the existence of |A| flanked hierarchically by the |I|s (as in Figure 10, right-hand side) protects the otherwise ill-formed *|I I| structure; while in others, the existence of |A| is transparent and renders the entire structure ungrammatical.

As we will see in the following section, |I| can also appear in the non-vocalic domain too; specifically, |I| is allowed to occupy a position within the domain where the non-resonance element |H| is the head (i.e. |H||ʔ||L|). It is possible for the same element to appear twice in an expression if the two tokens of that element reside in different (vocalic and consonantal) parts of the structure (the reader is referred to the discussion preceding and following Figure 13).

3.2 Palatal assimilation as SEARCH |H| and COPY |I|

The palatal dissimilation process just described for Japanese is observed in the vocalic set, while the opposite process of palatal assimilation involves palatality in both the vocalic and the dependent consonantal sets. The process itself targets only coronal obstruents (s, z, t, d) and the glottal fricative (h) when they precede the front high vowel i or the light diphthong jV. This is illustrated in (5). (Note that Japanese word-initial z is realised as [ʣ], which requires further explanation that is beyond the scope of this paper. In (5b) and (6b), therefore, the issue is avoided by only showing examples in which z appears word-internally.)

All the target segments of palatalisation are obstruents, which suggests that the noise element |H| (for obstruency) is crucial to the process. Note that this quite unlike the palatal dissimilation discussed above, which takes place in the vocalic domain where |H| is absent. Also, from the observation that palatal assimilation targets only coronal obstruents (s, z, t, d) and the glottal fricative (h) when they precede i or jV we can expect there to be something which is common to the internal structures of the target segments.

As seen in the three rightmost structures in Figure 12, the target segments (t, s/z, h) all have the noise element |H|, and in addition, all lack the rump element |U|, whereas non-target segments such as p/b and k/ɡ do contain |U|. In other words, palatalisation affects a consonantal set which has |H| but no |U| and which is dominated by a vocalic set containing |I| (the source of palatality). Consider t-palatalisation as an example.

Figure 12

The target segments of palatalisation.

As illustrated above, the process which palatalizes a consonantal structure may be analysed as palatality-spreading/copying on condition that this structure is recognized as obstruent. In element terms, it can be said that the existence of |H| (which defines obstruency) forces the most deeply embedded |I| in the vocalic set to COPY itself to the most deeply embedded part of the consonantal set. This may be formally expressed as in (7).

In the case of the sequence ti, the dependent |I| in the vocalic set (i.e. the only – and therefore, the most deeply embedded – token of |I|) copies itself onto the most deeply embedded part of the consonantal set containing |H|.

At this point let us address the following questions arising from this analysis.

Regarding (8a), ET shows a clear connection between |H| (obstruency) and |I| (palatality) in that both are united as members of the group of ‘light’ elements. As discussed in detail in Backley & Nasukawa (2009) and Backley (2011), the ‘light’ elements comprise the set |I H ʔ| while the remaining elements |U A N| are ‘dark’. Here it is claimed that palatalisation is driven by a mechanism in which the light element |I| seeks out another light element |H|, the former being copied onto a position where the latter is already present Table 3.

The reason why |I| and |H| behave as a set in the process in question is that they freely appear in both consonantal and vocalic domains whereas |ʔ| is typically limited to consonantal domains. Because the process in question (palatalisation) is an interaction between consonantal and vocalic domains, only elements which can function in this way naturally form a group. The same is true in the ‘dark’ group in some systems such as native (Yamato) Japanese, where |U| (labiality) and |L| (nasality/voicing) behave as a set: |U| can be employed in a single consonantal segment when it is accompanied by |L| as in m (|U L ʔ|) and b (|U L ʔ H|) while |U| with no |L| can only appear in a geminate consonant as in -pp- (|U ʔ H|).

As for (8b), the property that is copied occupies the most deeply embedded position (terminal dependent) in the whole structure, being subject to three levels of embedding. As such, it is able to maximize the effects of |I| percolating through the entire domain to ensure the most effective agreement of the active property.

The operations SEARCH |H| and COPY |I| in (7) also work in fricatives, as these also contain |H| in their structures. The same palatalisation process is observed in the case of fricatives, as illustrated below.

Unlike the stop t in Figure 13, the fricative s in Figure 14 has no |ʔ|; yet the V-dependent |I| is still copied to the most deeply embedded part of the |H| domain. Additionally, the glottal fricative h, consisting of a sole |H| element, is also a target for |I|-copying to its dependent position when the SEARCH and COPY operations apply, as shown in Figure 15.

Figure 13

Obstruent palatalisation: coronal stop.

Figure 14

Obstruent palatalisation: coronal fricative.

Figure 15

Obstruent palatalisation: glottal fricative.

By contrast, the |H|-headed set that has |U| is immune to palatalisation, as illustrated Figure 16.

Figure 16

No palatalisation: bilabial stop.

Even though the conditions for SEARCH and COPY are met, the presence of the rump element |U| prevents |I| from being copied to the |H|-headed domain. This is attributed to the following co-occurrence restriction which is operative in Japanese.

Notably, (9) applies not only to the consonantal (|H|-headed) domain but also to the vocalic (|U|-headed) domain. As discussed in §2.2, Japanese disallows the combination of |I| and |U| in a vocalic domain (note that the baseline element |U|, which is the ultimate head of the domain, does not count as a dependent |U|). And as shown below, what applies to p also applies in the case of k, since k also contains |U|).

The co-occurrence restriction in (9) prevents the velar stop k from being copied to the most deeply embedded part of the |H|-headed domain. (Note that k is phonetically palatalized, but the degree of palatalisation is perceptibly different from t, s, z, h.)

In Standard Japanese, thus, the arguments of COPY and the co-occurrence restriction are |I| and *|I U| respectively. Arguments for COPY are parametrically selected: |U| for rounding assimilation (e.g. round harmony in Turkish and Finnish), |A| for height assimilation (e.g. height harmony in Chichewa and Basque), |L| for nasal/voicing assimilation (e.g. postnasal voicing in Zoque and Japanese), |H| for voiceless assimilation (e.g. English and Swedish) and |ʔ| for stop gemination (e.g. Italian and Danish) (Harris 1994; Harris & Lindsey 1995; Nasukawa 2005; Backley 2011). As for the co-occurrence restriction, not only *|I U| (both of which are ‘colour’ elements) but also *|H L| (both of which are ‘source’ elements) are observed cross-linguistically (Harris 1994; Nasukawa & Backley 2005; Backley 2011). In principle, any combination of elements has the potential to act as a co-occurrence restriction although in practice there are clear tendencies: *|ʔ H| is marked although it does function when no other elements are present.

Returning to the copying of |I| (palatalisation), unlike Figure 17, some dialects of Japanese exhibit the palatalisation of velar stops: e.g. cɨŋko ‘safe’ in the Shiroishi dialect (kiŋko in Standard Japanese) and ɨɟɨ ‘railway station’ in the Morioka dialect (eki in Standard Japanese). This is illustrated as follows.

Figure 17

No palatalisation: velar stop.

In this dialect, unlike Figure 17, COPY(|I|) requires |I| in the vocalic domain to be copied at the highest dependent part (rather than the most deeply embedded part), which forces |U| from its position. As a result, as shown on the right in Figure 18, the structure of the consonantal domain is phonetically interpreted as c.

Figure 18

Palatalisation: velar stop.

The same process can be found in the case of ci < ti (e.g. cɨkara ‘power, force’ in the Morioka dialect (ʨikara in Standard Japanese)) where (according to the requirement of *|I U|) |A| in the highest dependent position of the consonantal domain is forced out and instead |I| in the vocalic domain is copied in the position Figure 19.

Figure 19

Palatalisation: coronal stop.

Like Standard Japanese, however, no bilabial stop palatalisation is observed in this dialect since two |U|s in the consonantal domain blocks COPY (|I|), as illustrated Figure 20.

Figure 20

No palatalisation: bilabial stop.

Another question to be addressed is why the vowel e, which also contains |I|, does not trigger palatalisation Figure 21.

Figure 21

No palatalisation before e.

As illustrated in Figure 4 in §2.2, e has |I| as a head and |A| as a dependent, so the most deeply embedded element is |A| rather than |I|. Since the operation COPY in (7) targets the most deeply embedded |I| in the V domain, the head |I| in e cannot be a source for the copying operation. As such, e fails to palatalize a preceding coronal obstruent or glottal fricative. However, in some dialects of Japanese (e.g. in Kyushu) the sequence se manifests itself as ɕe, suggesting that in those dialects it does not matter if the source |I| is in the most deeply embedded part of the structure or not: parametrically any |I| is copied to the consonantal set if it is present in the dominant vocalic set.

A final point to note is that the string ɕe is permitted in loanwords, in which case ɕ is not the result of palatalisation triggered by the following e: rather, it is simply a sequence consisting of ɕ plus any vowel (a, i, u, e, o), which is possible in Japanese loanwords.