2.1. The Jingdian shiwen

Our main source on the morphological alternations in Archaic Chinese is the 經典釋文 Jingdian shiwen, a collection of sound glosses on pre-Han classical texts by the Tang dynasty scholar Lu Deming 陸德明 (556–630).Footnote ³

This work records the pronunciation of rare characters, or of characters with several alternative readings, using either the fanqie 反切 method (glossing pronunciation with two characters, the first indicating the initial consonant, and the second the rhyme)Footnote ⁴ or directly glossing pronunciation with a homophonous character.

Two main types of morphological alternations recorded in this book have been systematically studied by a considerable number of scholars (Downer Reference Downer1959, Zhou Reference Zhou1962, Mei Reference Mei1980, Schuessler Reference Schuessler, Thurgood, Matisoff and Bradley1985, Sun Reference Sun2007, Jīn Reference Jīn2006, Bi Reference Bi2014, Wang Reference Wang2014, Baxter and Sagart Reference Baxter and Sagart2014), illustrated by a few examples in Table 1.

Table 1 Examples of morphological alternations inherited from Middle Chinese and preserved in modern Sinitic languages

The most frequent type of alternation involves a change from level, rising and entering tones to the departing tone.Footnote ⁵ This alternation has seven or eight distinct functions, including nominalization (for instance 數 ʂjuX “count” → ʂjuH “number”) and intransitivization (轉 ʈjwenX “(make) turn” (transitive) → 轉 ʈjwenH “turn”).Footnote ⁶

Second, we find examples of a voicing alternation involving initial obstruents: the form with an unvoiced onset is transitive, and its voiced counterpart is intransitive, as in 折 tɕet “break” (transitive) vs. 折 dʑet “break” (intransitive) or 敗 pæjH “defeat” vs. 敗 bæjH “be defeated”. In addition, the velar k- alternates with the voiced fricative ɣ- in Middle Chinese (繫 kejH “attach (transitive)” vs. 繫 ɣejH “be attached”).

There are in addition a number of less common, and poorly understood, alternations involving vowels or aspiration (see Schuessler Reference Schuessler2007: 51–78 for possible examples), but these will not be treated in the present paper.

2.2. Morphological alternations and analogy

It is undeniable that alternations originating from voicing or tonal alternations in Middle Chinese are present in standard Mandarin and most Sinitic languages for which appropriate documentation exists, as illustrated in Table 1.

Since the verbs in Table 1 belong to colloquial vocabulary, the idea that these alternations are purely scholarly inventions (§2.3.2) cannot be accepted: it is unrealistic to suppose that alternative readings invented by scholars could have penetrated the colloquial layer when literacy was not universal.

Alternative readings shared by both Mandarin and Middle Chinese are, however, rare. Most of the alternations in the Jingdian shiwen have left no trace in Mandarin and other Sinitic languages. However, this in itself is not surprising in view of the fact that even the pairs in Table 1 are not maintained by all Mandarin speakers. Some people generalize the transitive 折 zhé “break” even to describe spontaneous breaking, or use 繫 xì for transitive uses, saying for instance 繫安全带 xì ānquándài “fasten your seat belt” instead of jì ānquándài.

This variation is due to the generalization of one of the two members of the pair (which can be either the transitive or the intransitive verb), an elementary case of analogical levelling. These fossilized alternations are comparable to irregular verbs, and their loss can be viewed as a type of regularization, by analogy with non-alternating labile verbs, which constitute the majority.

However, not all alternations found in modern Sinitic languages necessarily go back to Middle Chinese or earlier. A particularly telling example is that of 背, which has two Middle Chinese readings pwojH (in the meaning “back”) and bwojH (“leave in one's rear,Footnote ⁷ betray”). In Mandarin, the character 背 has two readings, bèi (corresponding to the two previous nominal and verbal meanings), and also bēi “carry on the back”, a pronunciation without support from Middle Chinese, and absent from Cantonese and other Sinitic languages. The expected Middle Chinese corresponding to bēi would be †pwoj, but this reading is absent from all ancient sources, and bēi is not listed in dictionaries before the 1930s (Shi Reference Shi1999).Footnote ⁸

Table 2 provides an account of how the verb bēi “carry on the back” was created within Mandarin in the early twentieth century: 背 bèi “back” was reanalysed as an instrumental nominalization like 鑽 zuàn “drill (n)”, and a verb bēi “carry on the back” with level tone was invented by backformation, following the pattern of the alternation between 鑽 zuān “drill (v)” and 鑽 zuàn.

Table 2 Backformation of the verb 背 bēi “carry on the back” in Mandarin

Thus, analogical levelling can have taken place between Middle Chinese and Mandarin, and similar phenomena may have existed between Archaic and Middle Chinese. Nevertheless, for analogical extensions of alternations and backformation such as that illustrated above to take place, a kernel of inherited alternations must have existed in the first place, and the alternations found in Middle Chinese cannot have been entirely made up out of nothing.

2.3. Arguments against the reliability of the Jingdian shiwen

The neglect of the Jingdian Shiwen in studies on Archaic Chinese syntax can be partly explained by a perceived lack of reliability of the readings contained in this source (Zádrapa Reference Zádrapa2011: 71–6). In this section, I present the main arguments that have appeared in print against the reliability of the Jingdian shiwen, and provide counter-arguments.

2.3.1. Absence in modern Sinitic languages

One line of argument against the value of Jingdian shiwen comes from the fact that few of the variant readings it contains are attested in Mandarin, and that even among those that are recorded, they are mere dictionary pronunciations, and do not exist in the spoken language. As an example of this phenomenon, Zádrapa (Reference Zádrapa2011: 74) cites the alternation between the noun 衣 yī “clothes” (from Middle Chinese ʔjɨj) and the transitive verb yì “wear” (from ʔjɨjH): the latter is obsolete and rarely distinguished even by scholars when reading Classical Chinese.

However, the obsolescence of this reading is expected in view of the fact that the verb 衣 ʔjɨjH “wear” has been replaced in the spoken language by 著 ʈjak (since the Han dynasty, still in use in Cantonese) and by 穿 tɕʰwen (since the Tang dynasty). While the reading ʔjɨjH is not living in any mainstream Sinitic language, it has been preserved by Bai (Gong Reference Gong2015: 2). For instance, in Jianchuan Bai we have ji̠²¹ “wear” vs ji⁵⁵ “clothes” with the regular tonal correspondences.

This remarkable archaism in the Bai language, regardless of the controversy regarding the phylogenetic status of this language,Footnote ⁹ shows that the verb 衣 ʔjɨjH cannot be a scholarly invention. The verb 衣 ʔjɨjH “wear” is attested already in the Shijing, and occurs in a figura etymologica with its cognate noun, as in (1).

The same construction still exists in Bai, where the collocation ji̠²¹ ji⁵⁵ “wear clothes” is attested.

This example illustrates the fact that Mandarin is not particularly conservative when it comes to preserving traces of morphological alternations, and that a broader survey of alternations in other Sinitic languages (including Bai), or in borrowings from Chinese into Hmong-Mien, Kra-Dai, or Vietic languages, would be needed to evaluate the proportion of readings in the Jingdian shiwen which are indeed not confirmed by data from modern languages.

2.3.2. Doubts about the reality of some of the alternations in the writings of early philologists

It is a well-known fact that Qing dynasty philologists such as Gu Yanwu and Qian Daxin believed that the alternations recorded in the Jingdian shiwen came into existence after the Han dynasty (see for instance Shen Reference Shen2007: 104). However, such a hypersceptical view does not explain how these alternations could have come into being, and it is not compatible with the existence of traces of such alternations in Bai (§2.3.1).

Moreover, there are cases when tonal and voicing alternations are reflected by distinct characters, such as 受 dʑuwX “receive” and its causative counterpart 授 dʑuwH “offer”, a distinction already attested in the Oracle bone inscriptions (Takashima Reference Takashima, Peng and Shi2013, pace Djamouri Reference Djamouri2013).

Among the reasons for the scepticism of Qing scholars, we find an oft-quoted passage by the Northern Qi scholar Yan Zhitui:

Scholars from the Jiangnan area, when they read the Zuozhuan, have a made-up rule that was orally transmitted, whereby every time one's army is defeated, the verb “defeat” 敗 read bæjH (蒲 bu+邁 mæjH), and whenever one defeats an enemy army, it is read pæjH (補 puX+敗 bæjH). However, in all transmitted documents, I have never seen the reading pæjH. In Xu Xianmin (Xu Miao)'s reading of the Zuozhuan, there is only one example of this reading, and it has nothing to do with the difference between “be defeated” and “defeat someone”, (showing that the reading rule of Jiangnan scholars) is far-fetched and implausible.Footnote ¹⁰

In this passage however, Yan Zhitui does not express doubts on the system of alternations as a whole, but more specifically on the reading of the character 敗 in the tradition of the Jiangnan area. His stated reason for refusing to accept their tradition is the absence of such alternation in Xu Miao's work, and presumably in his own speech. However, rather than a testimony of the unreliability of the Jingdian shiwen, this passage shows that the Jiangnan dialect had preserved a distinction lost in Yan Zhitui and Xu Miao's dialects (both scholars were from present-day Shandong, though the latter lived nearly centuries before the former), a phenomenon no more surprising than the obsolescence of the reading 衣 yì “wear” in Mandarin mentioned above (§2.3.1).

A second line of argument by Gu Yanwu against the reality of the alternations concerns the example of 惡 ʔak “evil” and 惡 ʔuH “hate”. Gu Yanwu points out that the character 惡, used as a noun “evil, wrongdoings” in the Chuci (2) unexpectedly rhymes with 固 kuH and 寤 ŋuH. It is clear that in this passage 惡 is a noun, since it serves as the object of 稱 tɕʰiŋ “proclaim” and is opposed to the noun 美 mijX “beauty, good quality”.Footnote ¹¹

For the rhyme to work, a Middle Chinese reading ʔuH would be expected, running counter to the function of this reading in the Shiwen. Gu Yanwu then concludes that “From this we know that the difference between the qusheng and rusheng readings (of the character 惡) is just a matter of weak and strong pronunciation, and there is no difference between them”.Footnote ¹² This statement can be interpreted as the hypothesis that these two readings were in free variation in Archaic Chinese. This example (and other rhymes from the Han dynasty adduced by Gu Yanwu), however, does not prove that the alternation between ʔak “evil” and ʔuH “hate” in the Shiwen is spurious.

The reading of 惡 implied by the rhyme in (2) is not outlandish when one carefully investigates the glosses of the Shiwen. In example (3) from the Liji, for example, 惡 means “bad quality, defect” and is clearly opposed to 美 mijX “beauty, good quality” as in (2) above. However, the sound gloss in the Shiwen is 烏路反又如字, which contains the reading ʔuH (based on the fanqie spelling ʔu+luH) and the additional note 又如字 “also (read by some people as) the default reading”, in this case ʔak.

Of the two alternative readings proposed by the Shiwen, the first, 惡 ʔuH, has the qusheng reading, and is clearly in nominal function, exactly like the rhyming word in (2).

Thus, the Chuci rhyme in (2) supports, rather than discredits, the value of the Shiwen as preserving traces of morphological alternations. What these examples demonstrate is that the function of the qusheng alternation in the case of 惡 ʔuH is not limited to verbalizing denominalization.

Rather, (3) indicates that there were actually two abstract nouns meaning “defect, ugliness”, derived from the adjective “be bad, be ugly”, one derived by zero-derivation, and the other by qusheng alternation (originating from a *-s nominalization suffix through a *-h stage, according to Haudricourt Reference Haudricourt1954), as represented in Figure 1. The noun ʔuH “ugliness, defect”, however, disappeared to the advantage of the former.

Figure 1. Derivations of the readings of 惡 *ʔˁak

2.4. Non-specificity

Another argument to minimize the value of morphological alternations is the perceived non-specificity of the qusheng derivation, since it can derive nouns from verbs, verbs from nouns, intransitive verbs from transitive ones, and transitive verbs from intransitive ones. As Downer (Reference Downer1959: 262) put it:

A major drawback in Downer's approach, however, is to treat Archaic Chinese morphology as an isolated problem, without any reference to non-Chinese Sino-Tibetan languages.

In Hebrew, one of the arguments in favour of the genuineness of the Masoretic vocalization comes precisely from Semitic comparative phonology and morphology (Suchard Reference Suchard2019: 23). In a similar way, a satisfactory assessment of Archaic Chinese morphology cannot be undertaken in isolation from other languages of the Sino-Tibetan family.

Downer wrote these words before Forrest (Reference Forrest1960) showed, based on Haudricourt's (Reference Haudricourt1954) theory of tonogenesis, that the nominalization function of the qusheng could be compared to the -s nominalizing suffix in Tibetan (on which, see Zhang Reference Zhang2009, Hill Reference Hill, Lieber and Štekauer2014).

In addition, nominalization is not the only function of the qusheng derivation which has possible comparanda in other Sino-Tibetan languages. As shown in Jacques (Reference Jacques2016), sibilant or dental stop suffixes with functions similar or identical to that of the qusheng derivation are found in morphologically richer languages of the Sino-Tibetan family, especially Rgyalrongic, Kiranti, Nungish, and West-Himalayish (Table 3).

Table 3 Possible Sino-Tibetan comparisons for the functions of the qusheng

The idea that the qusheng could originate in part from dental stops, first suggested by Schuessler (Reference Schuessler2007: 42), is based on two observations. First, qusheng in Middle Chinese is much too frequent to originate exclusively from *-s: in Archaic Chinese reconstruction systems which adopt this hypothesis, the quantity of *-s far exceeds anything found in the rest of the Trans-Himalayan family, including Rgyalrongic and Tibetan. Second, from a typological perspective, the dental stop coda alternates with -s when following consonants in many languages of the family, including Tibetan (due to a sound change *-d → -s / {m,b,ŋ,g}_#, Coblin Reference Coblin1976), Kiranti (in particular Khaling and Dumi, see the internal reconstruction in Jacques et al. Reference Jacques, Lahaussois, Michailovsky and Rai2012), and West Himalayish (Martinez Reference Martinez2021).

Rather than a vague system of derivation insensitive to parts of speech, what these comparisons imply is that the remnants of morphology preserved in Middle Chinese glosses is the result of the convergence of unrelated suffixes due to the drastic phonetic attrition that has occurred between Archaic Chinese and Middle Chinese.

While Chinese is not phylogenetically close to either Gyalrongic or Kiranti, sharing with them only a limited number of cognates (Zhang et al. Reference Zhang, Jacques and Lai2019, Sagart et al. Reference Sagart, Jacques, Lai, Ryder, Thouzeau, Greenhill and List2019), the exuberant verbal morphology of these languages offers a framework against which hypotheses on the interpretation of morphological alternations can be tested (Gong Reference Gong, Park and Cheng2017, Zhang Reference Zhang2022), and makes it possible to go beyond the circularity of Chinese-internal analysis.

3.1. Nominalization

Among the functions of the qusheng derivation described by Downer (Reference Downer1959), one of the most prominent is nominalization, as mentioned above in §2.1. For instance, the verb 守 ɕuwX “defend, guard” (4) has the reading ɕuwHFootnote ¹³ when nominalized, either as agent nominalization “guard” (5), or as patient nominalization “area that is being guarded by the country” (6).

However, most verbs are nominalized without any tonal alternation. This is in particular the case with verbs already in qusheng tone, which have nothing to alternate with. For instance, the transitive verb 患 ɣwænH “be worried about, be troubled by” (7) has the nominal forms “trouble, disaster” and “person in distress” as in (8), which are also pronounced ɣwænH.

Zero conversion of verbs into nouns in Archaic Chinese (动词活用为名 词, Chen Reference Chen1922) is one line of argument in favour of the notion of Archaic Chinese as a “pre-categorial” language (Bisang Reference Bisang2008, Zádrapa Reference Zádrapa2011, Sun Reference Sun2020).

However, we find in Archaic Chinese not only verbs turned into nouns, but also entire phrases. For instance, in (9), 有禮 hjuwX lejX does not mean “X has ritual property”, but rather “person having ritual property”.

This noun phrase, which is frequent in the Zuozhuan, is a zero-marked relative clause, resulting from the elision of the nominalizer 者 tɕæX, a phenomenon pointed out by von der Gabelentz (Reference von der Gabelentz1881: §477), but relatively neglected in the literature. The full relative without relativizer elision happens also to be attested (10).

This type of unmarked relative can be detected in particular when a negation is present, as in the case of 不服 pjuw bjuwk “(those who) do not submit” in (11). The occurrence of this phrase in post-verbal position, as the object of 討 tʰawX, shows that 不服 pjuw bjuwk is a relative clause.

The problem of zero-nominalization can thus be turned around: there is no difficulty in analysing agent and abstract nominalizations as minimal relative clauses with elided 者 tɕæX nominalizer. Tonal alternation can thus serve as a criterion for true nominalization, as opposed to zero-relativization.

In other languages of the Sino-Tibetan family with productive affixal nominalization such as Tibetan or Japhug, zero nominalization (Hill Reference Hill2019) or finite relative clauses with no overt mark of nominalization (Jacques Reference Jacques2021b: §23.2.2, 1258) have been documented, without any need to appeal to a notion of pre-categoriality.

3.2. Passivization, relativization, and complementation

Another well-attested function of the qusheng, mentioned in §2.1 above, is turning a transitive verb into an intransitive one. Among these intransitive derivations, we find a few examples of antipassive (Jacques Reference Jacques, Janic and Makarevich2021a) and passive derivations.

For instance, the verb 使 ʂiX “send, despatch”, itself a denominal verb coming from the root reflected by the noun 吏 liH “official” has the passive form 使 ʂiH “be sent on a mission”.Footnote ¹⁴

This passive form is the one employed to derive the lexicalized relative clause 使者 ʂiH tɕæX “envoy”, as illustrated by (13).Footnote ¹⁵

The marker 者 tɕæX cannot be used to relativize objects on its own (Aldridge Reference Aldridge2013a). Thus passivization here is a strategy to make the patient accessible to subject relativization in 者 tɕæX (Keenan and Comrie Reference Keenan and Comrie1977).

Passivization also affects complement clauses, as illustrated by (14) and (15), which both have 請 tsʰjeŋX “ask” as complement-taking verb, and 使 ʂiX/H as complement verb.

In (14), there is co-reference between the transitive subject of 請 tsʰjeŋX and the intransitive subject of the passive 使 ʂiH “be sent”.

By contrast, in (15), there is no subject conference when the verb of the complement clause 使 ʂiX “send, let” is not passivized.

3.3. Consequences for the syntax of Archaic Chinese

These examples show that the morphology of Archaic Chinese is not simply a matter of word formation, but that it is directly relevant to syntactic analysis.

It is true that many verbs lack a distinct passive form expressed by tonal or voicing alternation. For instance, the verb 拘 “retain, seize”, has only one reading kju, even in the passive relative clause 拘者 kju tɕæX “those who had been retained”, as in example (16) (see Luo and Wu Reference Luo and Wu1983: 137).

However, examples of passive relatives in 者 tɕæX without overt passivization marking are in fact uncommon in Archaic Chinese: apart from (16), there is but one clear example of relative clause of the same type in the Zuozhuan.Footnote ¹⁶

Zero-passivization in Chinese is an instance of lability, a phenomenon that is also widely observed in languages with overt marking of transitivity. In the Sino-Tibetan family for instance, although Limbu has a very elaborate polypersonal indexation system, a certain number of frequent verbs, such as khutt- “steal”, can be conjugated either transitively or intransitively without any additional derivational morphology, as illustrated by (17) and (18) (from van Driem Reference van Driem1991: 527).

Archaic Chinese syntax differs from that of Limbu in degree rather than nature. The fact that a majority of verbs are labile does not imply that transitivity is an irrelevant parameter in Archaic Chinese syntax, but rather that the morphosyntactic contexts where transitivity is relevant are fewer than in Limbu. Focusing precisely on the syntax of verbs whose passive and antipassive forms are recorded in the Shiwen, and studying their syntax in a systematic way, could bring new light on the whole field of Archaic Chinese syntax, and restore the distorted image of this language brought about by the habit of reading it in standard Mandarin.