Is discourse made up of sentences? Focusing on dependent grafted speech in modern standard Japanese

3.1 Lexical determinacy

Languages such as English and Russian are considered to constitute a group in which the lexical accent is not determined by matters such as the phoneme structure of words and grammatical qualities (Lyons 1968), and modern standard Japanese can also be included in this group. The lexical accent of modern standard Japanese is not predetermined but varies among individual words (Beckman and Pierrehumbert 1986: 256), as seen in the examples of saru (‘monkey’), which has a word-initial accent pattern (a high accent on the first syllable); inu (‘dog’), which has a word-final accent pattern; and kizi (‘pheasant’), which has a flat accentless pattern.^[5]

3.2 Tendency toward inconspicuousness

These observations concern the lexical accents of simplex words, while the lexical accents of complex words in modern standard Japanese are known to exhibit a considerable degree of regularity (e.g., McCawley 1968). Furthermore, various tendencies can be observed even among complex words that deviate from this regularity. Here we will propose, as one such tendency, the tendency toward inconspicuousness described under (5) below.

This tendency demonstrates that one reason why lexical accents of complex words deviate from general regularity is because it is difficult to recognize words that are syntactically, semantically, and phonologically inconspicuous as words in their own right. See Examples (6)–(10) (the capitals in the examples are used for moras with high pitch except for term abbreviations.^[6] The same applies hereafter).

The words in Example (6) are complex words in which the latter elements (in the first example (a), genshō, ‘phenomenon’) consist of two feet of Chinese-derived words (that is, two characters). Incidentally, the latter elements of the complex words in Examples (6)–(10) are words spoken with an accentless pattern. These were chosen as examples that would make it easy to judge whether or not the accents were compounded.

Generally, the accent on a complex word with a two-foot latter element is, as in Example (6), one in which the accent nucleus (the ending point of the high accent) is positioned in the first mora of the latter element. For example, the word kaigenshō (‘strange phenomenon’) in (6a) has the accent nucleus positioned in the first mora ge of the latter element genshō (ka: low; i: high; ge: high; n: low; sho: low; o: low, LHHLLL hereafter).^[7]

However, when the former element of a complex word is the honorific o or go—that is, in the case of a prefix, with a light semantic content (which probably could be described as merely one of politeness), and only one syllable in length, this general regularity does not necessarily apply. While some complex words have this general regularity (such as o-den’wa LHLL ‘telephone’ and go-sidō LHLL ‘guidance’ in Examples (7) and (9)), others deviate from it and exhibit a tendency toward inconspicuousness (such as o-benkyō LHHHH ‘study’ and go-byōki LHHH ‘disease’ in Examples (8) and (10)). The latter complex words are accentless from the accent pattern of their latter elements. That is, in Examples (8) and (10), the prefixes o and go are ignored with regard to accent, and the lexical accent information of the latter elements (i.e., no accent nucleus) is succeeded by the entire complex word without change. This is what is referred to by the tendency toward inconspicuousness (5): “Inconspicuous words … will tend to be inconspicuous in terms of … lexical accent as well, … conforming to their surrounding environments.”

As with prefixes, a tendency toward inconspicuousness can also be observed in suffixes. Examples (11)–(13) illustrate complex words in which the latter element consists of one foot of a Chinese-derived word (that is, one character).^[8]

Of these, Example (11) lists complex words that conform to general regularity, while Example (13) lists complex words that exhibit the tendency toward inconspicuousness as they include the latter element -kun.

Let us first consider Example (11). While in a complex word with a one-foot latter element the accent nucleus is placed on the final foot of the former element (e.g., ki in hyōga=ki LHHL ‘ice age’),^[9] for certain latter elements, the accent follows the accentless pattern (e.g., go ‘word,’ gairaigo LHHHH; ‘foreign-derived word,’ Nihongo LHHH ‘Japanese language,’ and Tyūgokugo LHHHH ‘Chinese language’), as in Example (12).^[10]

However, this does not apply in the case of complex words with -kun, in which the latter element is a suffix, lighter in semantic content (it could probably be described as a way of referring to somebody), and only one syllable in length, as seen in Example (13). Specifically, this includes (a) complex words, in which -kun is attached to a former element that originally has a word-initial accent (e.g., Satō HLL), take a word-initial accent (HLLLL); (b) complex words, in which -kun is attached to a former element that originally has a word-internal or word-final accent (e.g., Okutō LHLL), take a word-internal of word-final accent (LHLLLL); and (c) complex words, in which -kun is attached to a former element that originally has no accent nucleus (e.g., Naitō LHHH), take an accentless pattern (LHHHHH). In these ways, whether there is an accent nucleus and its position are unchanged by attaching -kun. The same can also be observed in the cases of the other personal suffixes -san and -chan.

Conformity to this tendency toward inconspicuousness can also be observed in many cases among postpositional particles and copulas used to form phrases connected to nouns. Example (14) concerns postpositional particles and Example (15) copulas. Here, the noun saru (‘monkey’) is used as representative of the word-initial accent pattern noun inu (‘dog’) with word-internal/final and kizi (‘pheasant’) with an accentless pattern (See the end of Section 3.1.).

Here, the word-initial accent of the host word (saru ‘monkey,’ inu ‘dog,’ and kizi ‘pheasant’) is maintained even in a phrase that includes an enclitic (e.g., (14a) saru=ga HLL, inu=ga LHL, kizi=ga LHH). When the enclitic has a length of one mora, for the most part, the only exception is the genitive enclitic no. When no is attached to a word-final-accented word (e.g., inu), the word-final accent is not maintained and is removed (inu=no LHH).^[12]

Even when the enclitic has a length of two mora or longer, it can be observed that the first mora of the enclitic within the phrase often conforms to the “tendency toward inconspicuousness” (with the second and later morae taking a low accent). Example (16) illustrates this in the case of a postpositional particle and Example (17) in the case of a copula.^[13]

For example, while de in the postposition demo (“to”) of Example (16a) and de in the copula desu of Example (17a) have low accents when they immediately follow the word-initial-accented word saru or the word-final-accented word inu and high accents when they immediately follow the accentless word kizi, the mo in demo and the su in desu consistently take low accents.

3.3 Lexical accent as a boundary tone

Next, we will observe the lexical accent required of words positioned on the boundaries of utterances—that is, lexical accent as a boundary tone—by using these “inconspicuous” words that conform to the “tendency toward inconspicuousness” as a litmus test paper to assess the surrounding environment.

As seen in Section 3.2, the lexical accent on a one-mora long case particle that is ignored and conforms to its surrounding environment should be high when it immediately follows an accentless word. However, exceptional cases exist, such as when some kind of boundary is recognized between the accentless word and the case particle (Sadanobu 2017, 2019). Examples are presented under (18). Hereinafter, a boundary is indicated by a slash [/].

In the environment of directly following an accentless word (such as ataru ‘hit’ in Example (18a)), a high accent would be expected (as in the way ta and te in atatta ‘hit (past tense)’ and atatte ‘hit and’ take a high accent). Nevertheless, the accusative particle that immediately follows o has a low accent. This is because a boundary is recognized in terms of parts of speech between the accentless word and the particle. Such a boundary in terms of parts of speech is explained below.

One might say ryokō=ga suki=desu (‘I love travel’) or ryokōsuru=no=ga suki=desu (“I love to travel”) but would not say *ryokō=suru=ga suki=desu. In general, a verb will not be connected directly to a case particle in that order. These examples demonstrate that even when connecting such words together directly on an exceptional basis, a prominent boundary remains at the point of connection as this connection itself is not very strong. The examples illustrated under (18) are truly exceptional cases. They include two types of exceptions: idiomatic phrases (specifically, ataru=o saiwai [‘By chance …’] in (a), … suru=ga yoi [‘You had better …’] in (b), … suru=ni kosita koto=wa nai [‘It’s a good idea to …’] in (c), and iu=ni kotokaite … [‘be so stupid to (say …)’] in (d)), and cases in which verbs are quoted ((e)). In both of these cases, a verb is connected directly to a particle in that order.^[14]

From a cross-linguistic perspective, the phenomenon of spoken utterances taking a special form before and after boundaries is not unusual, both in terms of rhythm (e.g., Hyman and Leben 2000: 592) and phonation/voice quality (e.g., Chafe 2001: 674; Gordon and Ladefoged 2001: 391–392). Recognizing the tendency toward boundary tones (19) in modern standard Japanese, we can consider this to be the lexical accent version of that tendency (20).

In this way, with regard to lexical accent, we can understand the tendency of a particle, which tends to be ignored and conform to its surrounding environment, to take a low accent in the environment immediately following a verb even if it is accentless due to conformity with the tendency described under (20).

The tendency “to be lower before and after a boundary” under (20) includes cases in which an inconspicuous word is positioned before a boundary. Example (21) illustrates examples of inconspicuous words traditionally positioned immediately before clause boundaries, as conjunctive particles, and most of these can be understood in accordance with the tendency in (20).

These are examples of main conjunctive particles connected directly to the accentless verb iku (‘go’). Of these, ba, ga, kara, kedo, nara, node, noni, and si take low accents ((a)–(h)), and tara, tari, and temo, which include ta and te, conform to the surroundings in the first mora ta and te, while their second morae ra, ri, and mo, the last morae directly before the boundaries, take a low accent ((j)–(n)). Only the te and to take high accents ((i), (m)).^[15] Each of these conjunctive particles is a clitic that is light in semantic content (which probably serve only as connectors), short in length with only one or two morae, and inconspicuous, and each conforms to the “tendency toward inconspicuousness” in terms of accent, by conforming to its surrounding environment. However, the surroundings to which it conforms is not only the immediately preceding accentless word but also the immediately following clause boundary; whether it takes a high accent under the influence of the accentless word or a low accent in accordance with the tendency of boundary tone (20) varies by word.

The same can be said of the particles wa and mo, which should not be connected directly to elements other than nouns, as could be said of ordinary particles. See Example (22).

These are examples of exceptional combinations of accentless verbs ((a)(b)(d)(e)(f)) or negative ending ((c)) plus particles wa and mo. Such combinations are possible in idiomatic phrases including proverbs ((a)–(e)) and quotation ((f)). In these various environments, wa and mo consistently take low accents.^[16]

An observation similar to that above can be made in the case of exceptionally connecting an element other than a noun to another copula. See Example (23) below.

While, in general, a copula follows a noun, in exceptional cases, a copula may follow a word other than a noun. For example, it may follow a verb in the case of utterances by speakers of rural, small children, or milady-like person types (“kyara,” cf. Sadanobu 2020a), as in Example (a–c). Example (d) classically would be analyzed to involve the omission of iku=no (‘go’ + Genitive) between syatyoo=to (“with president”) and da (copula), leading to the sense of some kind of gap, or boundary, between the two. The same is true of Examples (e) and (f), which classically would be analyzed to involve the omission of =no koto (Genitive + ‘event’) immediately before da. Additionally, Examples (g–j) are cases of da appearing immediately following quotations or utterances resembling quotations (of sugu iku ‘go right away’ in (g), akkanbeee and iii ‘nyah nyah nyah’) in (h), ideophone gunyooo in (i), and iku ‘go’ in (j), and here too one senses a boundary between these utterances and the copula da. These utterances include not only accentless words (iku) but also ones for which the accent pattern is unclear to begin with (akkanbeee, iii, gunyooo); they all are followed by da with a low accent, and thus, they also can be considered to be exceptional cases due to the boundaries.

The same applies to disfluent utterances. See Examples (24), (25), and (26) below.

Here, a phrase consisting of an accentless noun plus a copula (ippai=da in (24), guuzen=zya in (25), and zettai=desu in (26)) appears (a) at the end of the sentence and (b) in the middle of the sentence spoken disfluently. As generally there is no problem connecting a noun to a copula, the copula following an accentless noun should take a high accent. However, this is the case only in the (a) examples above, with the copula in the (b) example taking a low accent. This is because, in the (b) examples, the utterances (hito=ga ippai haitte … in (24), uchi=ni guuzen ite … in (25), and husei=wa zettai yurusanai=to=iu sisei=ga … in (26)) are not spoken in a single breath but instead in a disfluent manner with small breaks. This way of speaking has a structure like that of a quotation, in which what the speaker wants to say is cited as “hito=ga,” “ippai,” and “haitte,” and as such, the copulas take low accents because of the boundaries between the phrases and the copula.

3.4 Cases in which no boundary tone is apparent

However, conjunctions positioned on sentence boundaries, which are the largest of all boundaries, exhibit surprising behaviors. See Example (27) below.

These examples list 60 words that can be considered typical conjunctions. Of these, those listed under (a) have lexical accents that start low (for example, in the first of these conjunctions, aruiwa, the accent on the first mora a is low), while for those under (b), the lexical accent starts high. There is no major difference in the numbers of words in these two groups, with 30 words listed under (a) and 30 under (b). However, when we look solely at conjunctions that begin with copulas, we see that all of these are listed under (b). For these conjunctions, the lexical accent starts high. The five words with dashed underlines are ones for which the naturalness is in doubt (danoni “but”)^[17] or for which it is unclear whether or not they can be said to start with a copula (nanode ‘so,’ nanoni ‘nevertheless,’ nara ‘then,’ naraba ‘then’), and thus perhaps they should not be included here. Still, that leaves 15 words with solid underlines. On the other hand, not one word fits in category (a) of those for which the lexical accent starts low.

This tendency of conjunctions, which should appear at the boundaries of sentences, to start with a high accent in all cases, instead of starting with a low accent when starting with a copula, and which should be sensitive to its environment, is most likely to be explained by the fact that the speaker is (hyperbolically speaking) “intentionally” connecting them to the preceding utterances through conjunctions.

The same can be said of utterances that begin with particles, as in the underlined passages of Examples (28) and (29). As these particles are bound, these passages can be said to be grafted speech.

As already demonstrated in (14i, m) in Section 3.2, the locative marker ni and the topic marker wa are inconspicuous enclitics (that is to say, their sensitivity to the environment is made known, like a litmus test, by the prosody produced by the speaker). Additionally, nii at the top of (28A2) and waa at the top of the staff’s speech in (29) both have high accents. This is because the speaker, in starting with these dependent parts of grafted speech, is attempting to connect them to the preceding utterances.

Turning at last to the observation of the dependent grafted speech, the central subject of consideration in this paper, we see that the same applies here. In most cases of dependent grafted speech, as seen in Example (30), the elements at the start of the utterance have intrinsic lexical accents (e.g., the word-initial accent in the case of kasira “Is that so?”), and these lexical accents are used when uttering them.

On the other hand, in principle, those that start with the copulas da and desu, which would tend to conform to their surrounding environment, begin with a high accent. See Example (31).

While for those in (31b), the accent starts low, for those in (31a), it starts high.^[19] Unlike the cases of conjunctions and other examples seen previously, instead of all falling into the category of those for which the accent starts high, as in (a), there are some cases that fall into group (b). However, these cases in (b) could be understood as following another kind of regularity, where a dependent word (that is, an inconspicuous word) in the mora immediately preceding an interactional particle na(a) or ne(e) at the end of utterances takes a low accent. This regularity applies not only to (b) but to (a) as well. For example, in da=yo=na(a) (‘I agree with you’), the yo has a low accent. Furthermore, it also can be observed in ka=na(a) (‘I am not sure’ Example (4) in Section 2) and ka=ne(e) (“I am not sure”), which do not start with copulas. In these cases, ka has a low accent. Thus, it has a high degree of generality.^[20]

However, if the observations above are valid, then we probably would need to review our understanding of the boundaries of utterances. The view of boundaries up till now has held that boundaries are assigned by parts of speech, and the meanings of sentences and prosody (or, in this case, lexical accent) reflect these boundaries. This is illustrated in Figure 3 below.

Figure 3:

The traditional view of boundaries. The vertical line indicates a boundary, the boxes “Word 1” and “Word 2” indicates inconspicuous words, and horizontal lines indicate low tones as boundary tones.

In Figure 3, there is a boundary between “Word 1” and “Word 2,” and as these words are inconspicuous, a low tone occurs before and after these boundaries to reflect these, as a boundary tone (or a low accent as described in Section 3.3).

The view of boundaries in Figure 3 is valid for our observations through Section 3.3. However, it is unable to explain the fact that the lexical accent begins high in the cases of utterances of conjunctions starting with copulas ((30b)) and those starting with particles and binding particles ((31) (32)), or the fact that in principle the accent starts high in the case of grafted speech beginning with a copula ((31a)), as seen here in Section 3.4. See Figure 4 below.

Figure 4:

The view of boundaries proposed by this paper. The vertical line indicates a boundary, and the horizontal lines indicate boundary tones. The top arrow points to when Word 1 is not intentionally connected to Word 2; the bottom arrow points to when Word 1 is intentionally connected to Word 2. The horizontal axis indicates an abstract time.

Unless intentionally connecting a new utterance (Word 2 in the figure) to one already uttered (Word 1 in the figure), a boundary tone appears before and after the boundary, as seen in Figure 3. However, when intentionally connecting the new utterance to the one already uttered, the boundary is canceled ex post facto, and as a result, the boundary tone remains only before the boundary (the developments from left to bottom right in Figure 4).

In summing up, the initial high accent of dependent grafted speech listed in (31a) reveals that the speaker cancels the boundary when they utter them. This indicates that the speech that we call “dependent grafted speech” resembles a conjunction type of speech far from the full-fledged sentence type of speech. Its “dependency” is not just nominal or conceptual but real and substantial, with prosodic substantiation.