Language change in a constructional network: the emergence of Mandarin [bi N hai N] comparative constructions

2.1 Previous research

Previous research has mainly centered on three issues. One is the constructional meaning of [bi N_i hai N_i]. In this regard, Yang (2011) notes that [bi N_i hai N_i], as a non-canonical comparative construction, does not so much compare two entities as it profiles an attribute of the comparee (e.g., sunniness, attributed to ‘my heart’ in (2)). This view is also shared by Ji (2012) and Ma (2014), who hold that [bi N_i hai N_i] serves to profile the characteristics of the comparee. Cheng (2013: 62), then, argues that the constructional meaning of [bi N_i hai N_i] can be characterized as “super or extremely + the stereotypical attributes of the noun”, which is a non-compositional meaning that has a coloring of exaggeration.

A second focus lies with the semantic features of the nominal constituent N sanctioned by the construction. It has been suggested that only nouns of strong descriptive character occur in [bi N_i hai N_i]. The descriptor ‘strong descriptive character’ was used, for instance, in Luo and Lu (2016), Ma (2014), and Yang (2011), to refer to nouns with salient attributes. For instance, in 他比葛朗台还葛朗台 Ta bi Grandet hai Grandet ‘He is meaner than Grandet’, the noun 葛朗台Grandet (referring to the character in the famous novel Eugénie Grandet) easily fits into the construction because it is strongly associated with the quality of being mean (see also Cheng 2013; Li 2016). Luo and Lu (2016) note that the nouns sanctioned into the construction have even become bywords for the salient attributes and that the referent of the noun is usually familiar to both the speaker/writer and the interlocutor/reader.

Thirdly, previous research has also touched on the mechanism underlying the development of [bi N_i hai N_i]. In this respect, Yang (2011) notes that the development of [bi N_i hai N_i] is analogically modeled on [bi N hai A]. Cheng (2013) as well contends that [bi N hai A] is the source of [bi_i N hai N_i], and he further proposes that the development of the latter construction is pragmatically motivated, in that the use of a N(oun) rather than an A(djective) avoids direct expression of the construction’s dimension of comparison. That said, Yang’s and Cheng’s studies provide no in-depth exploration of the development of [bi N_i hai N_i], as it is not their main focus. In contrast, the present study presents a nuanced account of the development of [bi N_i hai N_i] from [bi N hai A], and specifically what motivates the extension from A(djectives) to N(ouns).

2.2 A brief introduction to [Adverb A], [Adverb N], and [bi N hai A]

It will be argued (see Section 3.1) that among the source constructions that are of primary importance in the emergence and development of [bi N_i hai N_i] and [bi N_i hai N_j] are [Adverb N] and [bi N hai A]. The [Adverb N] construction, which dates back to the Qing dynasty (1644–1912), developed from a previously existing high-frequency construction, [Adverb A], a canonical construction in Mandarin – an example is (4). Like the adverbs in [Adverb A], the adverbs in [Adverb N] are on the whole degree adverbs. There are at least 17 degree adverbs that participate in [Adverb N] (cf. Li 2016: 81), such as 很 hen ‘very’, 最 zui ‘most’, 太 tai ‘too’,非常 feichang ‘very’, exemplified in (5)–(8).

The other source construction is [bi N hai A], a frequently-used, canonical comparative construction in Mandarin Chinese, as in (1), first attested in the late Yuan dynasty (14th century) in the CCL corpus. As 还 hai is also an adverb in Mandarin, this construction is closely related to [Adverb A], in that the component part “hai A” of [bi N hai A] instantiates the more schematic [Adverb A]. A variant of hai, namely, 更 geng can collocate with either an adjective or a noun constituting constructions [geng A] or [geng N] with the capacity to function as predicates, as exemplified by instances such as 她更美 ta geng mei ‘she is more beautiful’ and 他更绅士 ta geng shenshi ‘he is more gentleman-like’. At the same time, [geng N] can also work as a component part of a comparative construction, viz., [bi N geng N],^[7] as in 他比老板更老板 ta bi laoban geng laoban ‘he is bossier than a boss’, which is a variant of [bi N hai N].

2.3 Theoretical framework: (diachronic) construction grammar

The present study is couched within (Diachronic) CxG, a framework that exploits the “fit between the mechanisms of syntactic change and the basic principles of Construction Grammar” (Barðdal and Gildea 2015: 9). Or, as Perek (2020: 142) puts it, “Diachronic Construction Grammar aims to describe and explain language change by drawing on the idea that the grammar of a language consists of an inventory of form-meaning pairs, called constructions.” An important tenet of CxG is its network design, whereby all constructions form a structured inventory or a network of constructions, called a “construct-i-con” by Goldberg (2003: 219). This network comprises vertical (taxonomic and meronymic) as well as horizontal links^[8] between constructions. As Smirnova and Sommerer (2020: 3) point out, the incorporation of linguistic changes into CxG entails their reconceptualization as ‘network changes’, which may involve, among others, node creation (giving rise to constructionalization) and constructional network reconfiguration.^[9]

Like CxG, DCxG is usage-based in that linguistic change is shaped by language use (Barlow and Kemmer 2000; Bybee 2010; Diessel 2019; Hopper 1987; Langacker 2000; Schmid 2020). Major components of this usage-based approach which will be seen to play a role in the present study are frequency, entrenchment, and general cognitive abilities such as analogical reasoning and schematization (see also Smirnova and Sommerer 2020: 2).

Analogy, apart from being a mechanism of linguistic change, can be viewed as a domain-general cognitive processing mechanism: it is at the basis of speakers’ ability to create novel utterances (Bybee 2010: 75). The prerequisite of analogical processing is language users’ recognition – conscious or subliminal – of similarities between two structural or functional contexts.

Another central tenet in usage-based linguistics, and a major factor in grammatical change, is frequency of use. Frequent instances of a construction impact its cognitive representation, in that each instance contributes to the representation of the construction as a unit in the minds of language users, thus increasing its entrenchment and automatization (e.g., Diessel 2019: 1; Schmid 2020: 216). Frequency plays an important role in analogical extension: as Sommerer (2015: 11) points out, a speaker can extend a construction analogically only if they become aware of it in the first place and “[s]uch awareness will be favored or triggered by the construction’s high frequency”.

2.4 Nouns versus adjectives

Just like [Adverb N] has been said to have developed from [Adverb A] (see Section 2.2), the [bi N_i hai N_i] construction will be seen to develop from [bi N hai A] (Section 3.1). In either case, a noun is recruited into a slot where it is modified by an adverb; at the same time, adverbial modification is a typical characteristic of adjectives/verbs. Does this mean that the nouns recruited into this slot have become adjectives? To answer that question, we draw on Zhu’s (1997: 207, 213) insight that there is no correspondence between word class and syntactic function in Chinese: when an adjective or verb fills an argument position (e.g., subject or object) in a sentence, it remains a verb or adjective and does not change into a noun. In the same vein, when a noun fills the adjectival slot in [Adverb A] or [bi N hai A] (resulting in [Adverb N] and [bi N_i hai N_i]), it remains a noun, because it has properties not shared by adjectives. Zhu (1997: 204) notes that in Mandarin, the distinction between nouns and adjectives/verbs is prominent in that adjectives and verbs pertain to the category of 谓词 weici ‘predicate words’; this means that both verbs and adjectives usually serve as predicates in a sentence, and when they are predicates, they can be followed by aspect markers such as 了le or 着 zhe, while nouns are seldom used as predicates.

In general, nouns “characteristically represent entities, real or imagined, in the concrete or virtual world” (Shi 2016: 199) such as 孔子 Kongzi ‘Confucius’, 律师 lvshi ‘lawyer’, 会议 huiyi ‘conference’, while adjectives denote properties of entities (Huang et al. 2016: 276) such as 漂亮 piaoliang ‘beautiful’, 白 bai ‘white’, 安静 anjing ‘quiet’. Among the (more specific) distinguishing properties of adjectives are their amenability to affixation and reduplication (see Huang et al. 2016: 276–296; Shi 2016: 199–255). Nouns recruited into the [Adverb A] or [bi N hai A] structures, then, will not lose their nominal properties, nor will they assume these adjectival properties.^[10]

3.1 The multiple sources of [bi N_i hai N_i]

(i) [bi N hai A] as a source. It has been amply argued that language users can create novel expressions on the basis of analogical extension from (i.e., perceived similarity in form and/or meaning with) previously existing conventionalized phrases (Boas 2003; Bybee 2010: 60, 63; Gentner and Markman 1997; Krott et al. 2006). This means that many utterances are actually only partly novel. In this respect, Schmid (2020: 100) defines innovations as “partly licensed utterances”, and Traugott and Trousdale (2013: 58) note that “no construction is entirely new (except those that are borrowings and some coinings)”.

Against this background, we argue that it is through analogical extension that the construction [bi N_i hai N_i] developed from the prior, conventionalized [bi N hai A], as in (9). The source construction [bi N hai A] is attested from the 14th century^[11] (i.e., the late Yuan Dynasty) in the corpus CCL, while the earliest attested instance of [bi N_i hai N_i] (see example (10)) can only be dated back to 1969. In this process of analogical modeling, frequency has been argued to play an important role (Mańczak 1980; Bybee 2010: 25, 63). As Sommerer (2015: 118) states: “[T]he speaker can extend a specific schema analogically only if s/he becomes aware of it in the first place. Such awareness will be favored … by the high frequency of syntactic patterns which are compatible with such an underlying schematic representation”. The attested normalized frequency of occurrences instantiating [bi N hai A] is 7.26 per million characters in the multi-genre^[12] sub-corpus of BCC. Note that the normalized frequency of [bi N_i hai N_i] is 0.18 per million characters in the same corpus. Its raw frequency, number of types and type/token ratio are presented in Table 1, and Table 2 shows the raw frequencies of the top 10 types of this construction.

What does the analogical extension consist in? [bi N hai A] is a conventionalized construction conveying that a comparee takes a position on a standard or scale (expressed by A – the dimension of comparison) that is even higher (hai) than the (high) position taken by N (the standard of comparison). By way of example, consider examples (1) and (9). In instances such as these, [bi N hai A] provides an analogical base for [bi N_i hai N_i]. Specifically, in employing [bi N_i hai N_i], speakers retain the meaning of the previously existing conventionalized expression [bi N hai A]. They also retain a good part of its form, but they do modify A into N. In the process, non-scalar N is adjusted or reinterpreted in terms of N’s typical, scalar attributes; for instance, 阳光 yangguang ‘sunlight’ in (2) is interpreted as ‘sunniness’, i.e., in terms of the scalar attributes of the sun; in (3), 天才tiancai ‘genius’ is reinterpreted as (scalar) ‘genius-like’ characteristics. We suggest that the interpretation of N as A, and hence the extension of [bi N_i hai N_i] from [bi N hai A], can be attributed to coercion^[13] (Lauwers and Willems 2011; Michaelis 2005; Traugott and Trousdale 2013: 204–206). As Perek (2020: 147) points out, coercion captures the fact that “speakers occasionally push the boundaries of conventional usage”. In this case, while the noun in the construction does not have a scalar meaning of itself and would therefore make [bi N_i hai N_i] ill-formed, it can adopt the scalar meaning under influence of [bi N hai A]’s constructional/schematic meaning ‘higher (than expected) position on a scale than’. What facilitates reinterpreting N as A is likely the metonymic relationship between N and its stereotypical (scalar or N-like) attributes.^[14] Perek (2020: 147–148) further highlights that “if similar instances of coercion recur, the adjusted schema can itself become a conventional unit”. A possible consequence is that a higher-order schema [bi N hai X] may emerge, covering both the original [bi N hai A] construction and the adjusted/analogized construction [bi N_i hai N_i], as Figure 1 shows.^[15]

Figure 1:

Network of the [bi N hai X] comparative construction family.

In this case, then, we can observe that the criterion for sanctioning an item entering the schematic slot of a construction is not semantic similarity to the central member(s) of the slot (as, for instance, insane and nuts are sanctioned in [drive someone X] via semantic similarity to the prototypical fillers mad or crazy (see Bybee 2010: 36–37)), but semantic contiguity, i.e., a metonymic relationship.

Within the framework of DCxG, these linguistic changes involve node creation ([bi N_i hai N_i]) and possibly also constructional network reconfiguration (if a higher-order schematic construction [bi N hai X] emerges). The analogical extension from [bi N hai A] to [bi N_i hai N_i] is to be situated at the same level of schematicity; it thus constitutes a horizontal link (see, e.g., Smirnova and Sommerer 2020: 25–28) symbolizing partial similarity and non-inheritance.^[16]

(ii) [Adverb N] as a source. A second construction motivating the emergence of [bi N_i hai N_i] is the schematic construction [Adverb N],^[17] dating back to the Qing dynasty (1644–1912) in the CCL corpus.^[18] Unlike the source construction [bi N hai A], which is linked horizontally with [bi N_i hai N_i] (i.e., situated at the same level of schematicity) by analogical extension, [Adverb N] is taxonomically linked up with the constructional constituent “hai N” of [bi N_i hai N_i] through inheritance. In this respect, Goldberg (1995: 98) notes that constituents of constructions can inherit from other, higher-order constructions as well. Hence, the constituent “hai N” of [bi N_i hai N_i] can inherit from [Adverb N]. This notion of inheritance “captures a relation between more abstract constructions, which are situated towards the top of the constructional network, and more specific constructions, which are found in lower levels of the constructional hierarchy” (Hilpert 2014: 57). The type of inheritance link at stake here is an instance link, which is the basic type of inheritance link (see Goldberg 1995: 79–80; Hilpert 2014: 60; Traugott and Trousdale 2013: 50): 还 hai is an adverb and when it is used as a component part of a comparative construction, it can denote a degree difference in quality, namely, expressing ‘being more (than expected)’ (Lü 1999). Thus, 还 hai in [bi N hai X] is an instantiation of the general category ‘(Degree) Adverb’. Summing up so far, we contend that the emergence of [bi N_i hai N_i] is not only the result of analogical extension (see above), but it has also developed through inheritance, that is, by its instantiating the highly frequent schema [Adverb N].^[19]

A caveat is in order here. Although [hai N] (as in (11)) is an independent construction in Mandarin, it is impossible for it to be the up-level schema of [bi N_i hai N_i] as the meaning of independent [hai N] is quite different from that of the component “hai N” in [bi N_i hai N_i]. As example (11) shows, in [hai N ne/na], the noun is referential while in instances such as (2), the noun following hai in [bi N_i hai N_i] is not referential, i.e., it does not denote an entity, but it metonymically refers to the entity’s stereotypical attributes. Similarly, the noun in [Adverb N], as in (5)–(8), is not referential either. According to Lü (1999), 还 hai is an adverb expressing tones (e.g., of sarcasm, surprise, praise) in Mandarin and only when it is used in comparative constructions can it denote a degree difference in quality, namely, expressing ‘being more (than expected)’. Hence, in [hai N ne/na], as in (11), 还 hai is an adverb denoting tones while in the comparative construction such as [bi N hai A] and [bi N hai N], it expresses a degree difference in quality. In addition, there is a distinction in grammatical stress placement between [hai N ne/na] and [bi N hai N]. In [hai N ne/na], the stress falls on the N, whereas in [bi N hai N], the stress falls on 还 hai.

As Zong (1995) notes, in actual usage events, [hai N ne/na] denotes that the subject should have behaved the way Ns usually behave and thus conveying a tone of blame or sarcasm, as expressed by [hai xuesheng ne] in (11). Specifically, what the speaker intends to express in (11) is that stealing others’ work is not what a student should do, but the student did it anyway. Zong (1995) also compares the semantics of the nouns in [hai N ne] with those in [bi N_i hai N_i] and finds that the semantic categories of nouns in these two constructions are quite different: the nouns in [hai N ne] are predominantly nouns denoting people’s occupations, titles/ranks or status, while those in [bi N_i hai N_i] cover a much wider range of categories.

(iii) The related constructions [Adverb A] and [Adverb N]. Importantly, [Adverb N] can itself be said to have developed from [Adverb A] by analogical extension, whereby – in a way similar to the shift from [bi N hai A] to [bi N_i hai N_i] – the meaning ‘A to a still high/higher degree’ is retained, but the form gets adjusted (A > N). Moreover, given that the schematic constructions [Adverb A] and [Adverb N] are both present for speakers of Mandarin at a time when the [bi N_i hai N_i] construction had not developed yet,^[20] it is likely that these speakers viewed them as related (that is, as horizontally linked nodes in a constructional network). We suggest these linked schematic nodes may have facilitated the shift of the dimension-of-comparison slot from A in [bi N hai A] to N in [bi N_i hai N_i], as illustrated in Figure 2.

Figure 2:

Network of [bi N hai A/N] and [Adverb A/N].

3.2 Finding quantitative evidence for [Adverb N] as a source

If our assumption is correct that [bi N_i hai N_i] emerged via the extension of the A slot in [bi N hai A] to nouns via inheriting the nominal feature from [Adverb N], it would not be unreasonable to expect that the semantic categories of nouns that are significantly attracted to [bi N_i hai N_i] are correspondent to those attracted to [Adverb N]. This type of analogy does not involve identifying a common pattern (as in [bi N_i hai A] > [bi N_i hai N_i] or in [Adverb N] > “hai N” as a component of [bi N_i hai N_i]), but rather consists in similarity of conceptual content (labeled “concrete property matches” in Gentner and Smith 2012: 130, or “physical similarities” in Behrens 2017: 222). To investigate this hypothesis, we conducted a collostructional analysis.

Data for this study were retrieved from the corpus BCC (Xun et al. 2016). BCC has a total size of about 9.5 billion characters and it comprises five genre-based sub-corpora: news reportage (2 billion), literary works (3 billion), dialogues (600 million), multi-genre (1.9 billion), and Literary Chinese (2 billion). The corpus was tagged for part of speech.

Since the [bi N_i hai N_i] construction is a relatively recent development – its earliest attested instance can only be dated back to 1969 and the earliest attested occurrence of [bi N_i hai N_j] dates from 1980,^[21] the data were retrieved from the multi-genre sub-corpus of the BCC Contemporary Chinese corpus. This sub-corpus is balanced by genre, including texts from newspapers, fictions, micro-blogs, and academic journal articles of science and technology.

We searched the multi-genre sub-corpus of the BCC corpus with the query “bi n hai n”. In all, 559 concordances of [bi N_i hai N_i] and [bi N_i hai N_j] were retrieved. After manual removal of noise, 314 instances of [bi N_i hai N_i] and 149 instances of [bi N_i hai N_j] were retained. To compare the semantic features of nouns in [bi N_i hai N_i] with those in [Adverb N], we searched the multi-genre sub-corpus of the BCC corpus with the query “hen n” and obtained 64,648 concordances, from which we randomly sampled 1000 concordances. After discarding noise, we retained 264 instances of [hen N].^[22]

To examine the similarities between the lexical preferences expressed by N in [bi N_i hai N_i] and [Adverb N], we conducted a simple collexeme analysis^[23] (e.g., Stefanowitsch 2006, 2013]; Stefanowitsch and Gries 2003), which can determine which nouns are statistically significantly associated with each of the two constructions when compared against the language as a whole. Albeit subject to some criticism (e.g., Divjak 2008; Schmid 2010; Schmid and Küchenhoff 2013), collostructional analysis is still regarded as a useful method to calculate the association strength between a construction and its slot-fillers.

For every noun in [bi N_i hai N_i] and in [hen N], we calculate its association strength with the construction via computing a 2-by-2 contingency table which contains four different frequencies of occurrence: the frequency of the target noun in the target construction (A), the frequency of the target noun in the corpus in general (B), the frequency of the construction filled with other nouns than the target noun (C), and the frequency of all other constructions with lexemes other than the target noun (D).

On the basis of these frequencies, the expected frequencies and the association measure can be calculated. Based on the contingency table, we also derived the log odds ratio as well as the 95 % confidence interval for every noun (Schmid and Küchenhoff 2013). A significantly positive log odds ratio suggests that the noun is attracted to the construction whereas a significantly negative log odds ratio indicates that the noun is repelled. Figures 3 and 4 visualize the most important part of the results: the top 30 significantly attracted nouns (purple bars) and the top 4 repelled nouns (green bars).

Figure 3:

Collostructional strengths of [bi N_i hai N_i] constructions with logOR >3.2 and logOR <−1.8. Strength is measured by log(OR) (horizontal axis). Error bars denote 95 % confidence intervals. Purple denotes attracting nouns (‘attractor’; log(OR) significantly positive); green denotes repelling nouns (‘repellor’; log(OR) significantly negative).

Figure 4:

Collostructional strengths of [hen N] constructions with logOR >3.1 and logOR <−1.95. Strength is measured by log(OR) (horizontal axis). Error bars denote 95 % confidence intervals. Purple denotes attracting nouns (‘attractor’; log(OR) significantly positive); green denotes repelling nouns (‘repellor’; log(OR) significantly negative); light gray denotes constructions which are neither repellors nor attractor.

Based on the results of the collostructional analysis, we classified the top 30 attracted nouns in [bi N_i hai N_i] and those in [hen N] into five categories: human, animal, place, object, and abstract nouns (this categorization was taken from Shao et al. 2019). ‘Human’ includes nouns such as 男人 nanren ‘men’, 女人 nüren ‘women’, 老板 laoban ‘boss’, 绅士 shenshi ‘gentlemen’, and 警察 jingcha “policemen”; “animal” includes 熊猫 xiongmao ‘panda’ and 骡子 luozi ‘mule’; ‘place’ involves 宾馆 binguan ‘hotel’ and 地狱 diyu ‘hell’;^[24] ‘object’ encompasses 教科书 jiaokeshu ‘textbooks’ and 闹钟 naozhong ‘alarm clocks’. The label ‘abstract noun’ here refers to nouns that are “non-referrable to concrete objects or entities in the natural or human world” (Po-Ching and Rimmington 2004: 10), and are products of human epistemology, such as 真理 zhenli ‘truth’ and 梦想 mengxiang ‘dreams’, which is in contrast to the other four categories that express concrete concepts. The semantic distribution of the top 30 nouns attracted to [bi N_i hai N_i] and [hen N] is presented in Table 3.

To examine whether the (types of) nouns attracted to the two constructions are similar, we tested the association between the semantic categories of the most attracted nouns in [bi N_i hai N_i] and in [Adverb N], making use of a Fischer’s exact test. This test revealed that there is no statistically significant difference (p = 0.1164) between the frequencies of the categories of most attracted nouns in the two constructions, indicating that there is a fair degree of semantic similarity between the nouns attracted to the two constructions.

In summary, we have seen that the emergence of [bi N_i hai N_i] can be attributed to multiple sources: there exist both formal and semantic similarities between the analogical model construction [bi N hai A] and the target [bi N_i hai N_i], such that the latter emerged by analogy with the former; next, [bi N_i hai N_i] also inherits from [Adverb N]; furthermore, the development [Adverb A] > [Adverb N] might have facilitated the shift from [bi N_i hai A] to [bi N_i hai N_i] (Trousdale 2013; Van de Velde et al. 2013).

4.1 The emergence of [bi N_i hai N_j]

We have seen that in [bi N_i hai N_i], the noun following hai – expressing the dimension of comparison – does not refer to a specific entity but metonymically refers to the entity’s stereotypical attributes and is used to specify the attributes of the comparee. What [bi N_i hai N_j], as exemplified in (12), shares with [bi N_i hai N_i] is that N_j does not refer to a specific entity either. What differentiates the two constructions is that in [bi N_i hai N_j], the noun expressing the dimension of comparison (N_j) differs from the noun expressing the standard of comparison (N_i). Accordingly, in (12), N_j 剧情 juqing ‘plot’ metonymically refers to the quality of being plotty, and the comparee (the case analysis) has more of the stereotypical attributes of a plot (N_j) than a novel (N_i). Since N_j is different from N_i, this construction enjoys relatively more freedom to denote comparison than [bi N_i hai N_i].

We retrieved (after manual cleanup) 149 tokens of [bi N_i hai N_j] from the multi-genre sub-corpus of the BCC corpus, with 133 types and 101 hapax legomena, as shown in Table 4. The type/token ratio is very high, which suggests that the construction is very productive (Bybee 2010; Dąbrowska 2008, among others).

We suggest that [bi N_i hai N_i] has served as an analogical base for [bi N_i hai N_j], whereby the latter construction has retained the structural pattern as well as meaning of the former. Structurally, the standard and the dimension of comparison are each expressed by a noun; semantically, the two constructions each convey that the comparee has even more of the stereotypical attributes of the dimension of comparison (N_j/N_i) than the standard of comparison (N_i/N_i). The change that can be observed is that the nouns expressing the standard and dimension of comparison no longer need to be identical. That is, [bi N_i hai N_i] only partially sanctions [bi N_i hai N_j], just as [bi N hai A] only partially sanctions [bi N hai N]. Therefore, in Figure 5, the link between the dimension slot of [bi N_i hai N_i] and that of [bi N_i hai N_j] is indicated by a dotted arrow.

Figure 5:

The relationship among [bi N hai A], [bi N_i hai N_i] and [bi N_i hai N_j].

4.2 Interim summary

The change from [bi N hai A] to [bi N_i hai N_i] and then to [bi N_i hai N_j] shows an expansion of the dimension-of-comparison slot. In the first stage of the development, from [bi N hai A] to [bi N_i hai N_i], the dimension-of-comparison slot expands from an adjective to a noun which is the same in form as the noun denoting the standard of comparison. We have seen that there are several motivating factors, or sources, of this change: (i) analogy-based extension from [bi N hai A], whereby its scalar meaning is transferred to a formally similar structure, with, in its dimension-of-comparison slot, a noun that is identical to the noun in the standard-of-comparison slot; the noun in the dimension-of-comparison slot metonymically refers to the typical scalar attribute(s) of its referent, thus aligning its semantics with the scalar semantics of adjectives; (ii) [Adverb N], which is taxonomically linked up with the constructional component “hai N” of [bi N_i hai N_i] through inheritance: the category Adverb is instantiated by hai and the dimension-of-comparison slot of [bi N hai A] inherits the nominal feature of [Adverb N]; (iii) the shift [Adverb A] > [Adverb N] facilitates the shift to [bi N_i hai N_i]; (iv) conceptual similarity between the type of nouns attracted to [Adverb N] and [bi N_i hai N_i]. The second stage, from [bi N_i hai N_i] to [bi N_i hai N_j], is directly analogous to [bi N_i hai N_i], with [bi N hai A] as a latent analogical supporting pattern. As it did in the first stage, metonymy plays a facilitating role in this stage as well.

4.3 Quantitative description

We have seen that in the [bi N_i hai N_i] construction, the nouns expressing the standard and dimension of comparison are formally identical; in the [bi N_i hai N_j] construction, which developed from [bi N_i hai N_i], this is no longer the case. It therefore seems worthwhile investigating how these two constructions compare; specifically, we will examine the semantic features of the nouns in the dimension-of-comparison slot of the two constructions.^[25] To that end, the semantic annotation scheme used to categorize the nouns in [bi N_i hai N_i] and [Adverb N] (in particular [hen N]) (see Section 3.2) is employed here as well; that is, nouns are classified into five categories: ‘abstract concept’, ‘place’, ‘object’, ‘animal’, and ‘human’.

Figure 6 illustrates the change in the distribution of semantic features of the nouns in the dimension-of-comparison slot of the two constructions, viz., the second N_i in [bi N_i hai N_i] and N_j in [bi N_i hai N_j]. In [bi N_i hai N_i], 53 percent of collocates (i.e., 165 instances) consist of nouns referring to humans. Nouns denoting animals rank second, 18 percent (56 instances). Instances boasting the third largest number are those involving nouns encoding objects (40 instances), which is followed by instances with nouns denoting abstract concepts (38 instances). At 5 % (15 instances), nouns expressing places take up the smallest share. In [bi N_i hai N_j], a different picture is observed. The majority of nouns in the N_j slot now denote abstract concepts, accounting for 57 percent (85 instances) of all nouns, or four times more than their share in [bi N_i hai N_i]. Further, the proportion of ‘animals’ rises by almost 18 percent to 26 percent (39 instances); in contrast, nouns designating people drop to about 12 percent (17 instances) – a big decline from their share in [bi N_i hai N_i]. Finally, there are no nouns denoting places appearing in the N_j slot of [bi N_i hai N_j]. It can be seen, then, that the majority of N_j slots is filled by abstract nouns denoting an abstract quality of the comparee, as in 女人比男人还野心 nüren bi nanren hai yexin ‘this woman is even more ambitious than men’ (literally, this woman BI men HAI ambition). This suggests that the [bi N_i hai N_j] construction can be aligned with (or, is similar to) [bi N hai A], in that abstract concepts are very much in line with the general semantic profile of adjectives, which prototypically denote a(n) (abstract) quality. This semantic alignment of N_j with adjectives, denoting abstract qualities, we argue, is in keeping with Gentner and Smith’s (2012: 130) “concrete property matches” (see also Section 3.2; Behrens 2017: 222).

Figure 6:

The semantic property of the noun in [bi N_i hai N_i] and [bi N_i hai N_j].

The development from [bi N_i hai N_i] to [bi N_i hai N_j] is in accord with Perek’s (2016) observation that diachronically, new types of a construction tend to be semantically close to already existing types. Also related to this fact is the attraction effect proposed by De Smet et al. (2018), who note that as a result of analogy, “competing forms often show attraction, becoming functionally more (instead of less) alike”. This may also be the case for the development from [bi N_i hai N_i] to [bi N_i hai N_j]. That is, the canonical comparative construction [bi N hai A] attracts the noun comparative construction [bi N_i hai N_i] and makes it remove the constraint on the second noun (i.e., the second noun should be formally identical to the first noun), thus giving rise to the emergence of [bi N_i hai N_j]. In this process, the second noun can be extended to more abstract nouns, leading the whole construction to behave more similarly to [bi N hai A].