2.1. Dvandva from a Semasiological Perspective
In
Bauer (
2008) and
Arcodia et al. (
2010), dvandvas are defined as a type of coordinate compound in which the components are parts/hyponyms and the entire construction is the whole/hypernym. In
Bauer’s (
2008) classification, the first type of dvandva is called
additive and it refers to the collective set of members that are co-meronyms or converses. Converses are a subtype of opposites.
3 Most English dvandvas, including those in (3), belong to this type. Given below are Japanese additive dvandvas cited from
Yonekura et al. (
2023, chap. 2). Hyphen-connected examples consist of freely occurring morphs, while the unconventional use of the word-internal equal sign captures those connecting bound morphs.
(4) | a. | oya-ko | Additive |
| | parent-child |
| | ‘parent and child’ |
| b. | te-asi | |
| | hand-foot |
| | ‘hands and feet, the limbs’ |
| c. | me-hana |
| | eye-nose |
| | ‘eyes and nose’ |
| d. | dan=zyo |
| | man-woman |
| | ‘man and woman’ |
| e. | huu=hu |
| | husband-wife |
| | ‘husband and wife’ |
| f. | zi=moku |
| | ear-eye |
| | ‘eyes and ears; one’s attention or notice’ |
As subtypes of the semantic relationship between binary opposites,
Cruse (
2011) distinguishes complementary, antonymic, reverse, and converse relationships (see note 3). Complementary or antonymic opposites produce the second type of dvandva called
exocentric. For example, the Japanese dvandvas in (5a, c) combine pairs of complementary opposites, while those in (5b, d) consist of antonymic opposites:
(5) | a. | zen-aku | Exocentric |
| | virtue-vice |
| | ‘virtue and vice, good and evil’ |
| b. | yosi-asi | |
| | good-bad |
| | ‘good and/or bad, right and/or wrong, merits and demerits’ |
| c. | ze=hi |
| | right-wrong |
| | ‘right and/or wrong, pluses and minuses, pros and cons’ |
| d. | sin=kyuu |
| | new-old |
| | ‘the old and the new, old or new’ |
By the very semantic nature of the connected lexemes, exocentric dvandvas may express disjunctive coordination in addition to conjunctive coordination. Moreover, this type has a third reading that refers to the underlying scale itself, as observed by
Scalise et al. (
2009) and
Shimada (
2013).
What
Bauer (
2008) calls the exocentric type is not only semantically but also categorially exocentric. It is a bit odd to have an “exocentric” type within a class whose defining characteristic is exocentricity, but the concept of “exocentric exocentrics” is bound to emerge due to the widely recognized fact that the exocentricity in compounding is not monolithic but determined by several parameters (
Namiki 2001;
Scalise et al. 2009;
Bauer 2010b,
2022;
Nóbrega and Panagiotidis 2020). In
Section 1, the exocentricity of dvandvas as a class was determined based on the “
Z is a type of X/Y” headedness test, that is, semantic exocentricity. In (5), on the other hand, the parameter at work is syntactic category. According to this parameter, the head of a compound is defined as a category determinant. Thus, in (5b), the free-morph components are adjectives, while the entire constructions are nouns; this shows categorial exocentricity. Interestingly,
Scalise et al. (
2009, p. 65) suggest that while the notion of categorial head may be tied with typological properties of the language, “the semantic requisites of a compound are not parameterized in any language”. These observations indicate that dvandvas lack a semantic head and may also be categorially headless.
The third and fourth types of dvandvas are called
co-hyponymic and
co-synonymic, respectively. The following Japanese examples suggest that dvandva coordination can function to neutralize the boundary between subtypes named by co-hyponyms or co-synonyms and to synthesize them into a single new type:
(6) | a. | kusa-ki | Co-hyponymic |
| | grass-tree |
| | ‘plants, vegetation’ |
| b. | gyo=kai | |
| | fish-shellfish |
| | ‘fish and shellfish’ |
| c. | tyoo=zyuu |
| | bird-beast |
| | ‘birds and animals, wildlife’ |
(7) | a. | sugata-katati | Co-synonymic |
| | figure-shape |
| | ‘outward appearance’ |
| b. | kai=ga | |
| | picture-picture |
| | ‘a picture; pictorial arts’ |
In (6a),
kusa-ki can literally refer to grasses and trees but is more commonly used to refer indiscriminately to everyday plants. In (7a),
sugata and
katati are synonyms expressing ‘outer form’, with the former being used for the outer form of someone and the latter for the external form of something. However, this boundary is neutralized in the dvandva, so
sugata-katati does not discriminate the animacy of the possessor.
Independently of the semasiological classification, there are two observations to be made with regard to the morphophonological aspect. First, Japanese NN dvandvas retain the accent nucleus of the left component, as in (4)
o’yako,
te’asi, (6)
kusa’ki, (7)
su’gatakatati. This is said to deviate from a more dominant accent pattern of Japanese compounding where the right component acts as the determiner (
Tsujimura 2014, pp. 86–96). A good example of the latter would be:
a’kusento ‘accent’ +
ki’soku ‘rule’ →
akusentoki’soku ‘accent rule’.
Second, throughout the data in (4–7), dvandva components appear in either the [free morph + free morph] or [bound morph + bound morph] pattern; mixed realizations are not observed. This observation, which was made in
Shimada (
2013), fits naturally with what we saw in
Section 1: the formal parallelism between coordinates. At the same time, Shimada’s observation raises the question of which pattern is more fundamental. The answer seems to be the bound + bound pattern, since many contemporary free + free examples have corresponding bound + bound realizations used in earlier times. For example, the example in (4b) transcribes the so-called
kun-reading ‘Japanese reading’ of the word written手足, but this word used to be read as
syu=soku in the
on-reading ‘Chinese reading’. Significantly, the
kun and
on-readings of a kanji character are bound and free morphic realizations of the lexeme represented by that kanji (
Nagano and Shimada 2014). Presumably, the morphophonological alternation of kanji characters contributed to the gradual emergence of the free + free pattern from the bound + bound pattern. The proposed diachronic relationship between the two realization patterns is also consistent with the emergence of minor exceptions to the rule of the same-morph type realization.
4To summarize this section, classic dvandvas are semantically exocentric compounds that asyndetically coordinate co-meronyms, converse/complementary/antonymic opposites, co-hyponyms, or co-synonyms. Furthermore, the formal parallelism between the coordinates, or a strong propensity toward it, can be observed in a number of languages.
2.2. Dvandva from an Onomasiological Perspective
Dvandva compounds typically name a higher-level concept or superset that includes the referents of its components as smaller subsets. Thus, from an onomasiological perspective, classic dvandvas belong to the catalogue of constructions that have been gathered under the umbrella of
lexical plurals (
Acquaviva 2008;
Lauwers and Lammert 2016;
Gardelle 2019).
For example, additive dvandvas are similar to group nouns such as
committee,
family,
herd,
nation, etc., and bipartite nouns such as
a pair of {shoes/socks/earrings/gloves…} (cf.
a shoe) and
a pair of {glasses/scissors/trousers…} (cf. *
a glass) (
Huddleston and Pullum 2002, pp. 340–42). They all name spatially bounded units that are composed of separable similar internal units. In the LSF (Lexical Semantic Framework), where lexical semantics is described as bundles of features (
Chomsky 1965), such nouns share the
quantity-related semantic features [+B (bounded), +CI (composed of individuals)] defined as follows (
Lieber 2004, p. 136):
[B]: This feature stands for “Bounded.” It signals the relevance of intrinsic spatial or temporal boundaries in a situation or substance/thing/essence. If the feature [B] is absent, the item may be ontologically bounded or not, but its boundaries are conceptually and/or linguistically irrelevant. If the item bears the feature [+B], it is limited spatially or temporally. If it is [−B], it is without intrinsic limits in time or space.
[CI]: This feature stands for “Composed of Individuals.” The feature [CI] signals the relevance of spatial or temporal units implied in the meaning of a lexical item. If an item is [+CI], it is conceived of as being composed of separable similar internal units. If an item is [−CI], then it denotes something which is spatially or temporally homogeneous or internally undifferentiated.
A brief introduction of the LSF would be appropriate. At the most basic level, lexemic concepts are categorized into either
situation (which encompasses
event/state) or
substance/thing/essence. These categories are then broken down into two-stratum bundles of features: body, which contains encyclopedic information, and skeleton, which is a hierarchically organized structure of functions (F) and their arguments. Skeletons are represented in a format suggested below, with 1 and 2 representative ones.
1 | [F1 ([argument])] |
2 | [F2 ([argument], [F1 ([argument])])] |
The F(unction) stands for a grammatically relevant semantic feature, such as [material], [dynamic], [Loc] (location), [IEPS] (inferable eventual position or state), [Scalar], [Animate], [B], and [CI]. For instance, the skeleton of the simplex noun
dog is of the first type, as illustrated below, where [+material] refers to its primary functional feature and the inner brackets represent its referential argument. The complex word
writer has a more intricate skeleton in which one set of features and arguments is embedded under another:
1 | dog | [+material ([ ])] |
2 | writer | [+material, dynamic ([i ], [+dynamic ([i ], [ ])])] |
The second structure reflects the process of word formation, where the higher set corresponds to the affix (
-er) skeleton and the embedded set corresponds to the base verb (
write) skeleton. The highest arguments of the two componential representations are coindexed through the working of the Principle of Coindexation (
Section 4.2.1, (27)).
As elaborated in
Lieber (
2004, chap. 5), the quantity-related semantic features [B] and [CI] are given in the skeletons of nouns and verbs and capture the widely recognized parallelism between nouns and verbs in their quantitative semantic properties, namely, number and lexical aspect. Thus, the parallelism between singular count nouns (e.g.,
person,
fact) and non-repetitive punctual verbs (e.g.,
explode,
name) is captured by their possession of the featural complex [+B, −CI]. Mass nouns (e.g.,
furniture,
water) and non-repetitive durative verbs (e.g.,
descend,
walk) are similar because they share the featural complex [−B, −CI]. The feature [+CI], our focus, underlines group nouns (e.g.,
committee,
herd), plural nouns (e.g.,
cattle,
sheep), and repetitive durative verbs (e.g.,
totter,
pummel,
wiggle). As summarized in
Table 1, group nouns have no verbal counterparts because “[…] for a verb to be intrinsically [+B, +CI] it would have to denote an event that is at the same time instantaneous/punctual and yet made up of replicable individual events, a combination which does not seem possible” (
Lieber 2004, p. 139):
These features differ from
Jackendoff’s (
1996) features [b] and [i] (
Lieber 2004, pp. 141–44). Because the latter features are aimed to account for the property of telicity, an aspectual property observed at the level of the verb phrase or whole sentence, they are underspecified at the lexical level. On the other hand, the target of the LSF theory is lexemes and morphology, and the features [B] and [CI] belong to nouns, verbs, and derivational affixes.
5The next important point is that LSF features can be manipulated by morphology. In English, for example, the semantic contribution of the progressive suffix
-ing is to add the feature [−B] to the base lexeme, while the plural suffix
-s contributes the feature complex [−B, +CI]. In derivational morphology, the suffixes
-ery and
-age produce collective nouns from singular count nouns, as in
jewelry from
jewel,
peasantry from
peasant,
mileage from
mile,
wreckage from
wreck. This observation is explained if the suffixes “add the features [+B, +CI] to their base, indicating that the derived noun is to be construed as a bounded aggregate or collectivity of individuals related to the base noun” (
Lieber 2004, p. 149). In other words, the two derivational suffixes possess the skeleton [+B, +CI ([ ], <base>)], in which <base> represents the skeleton of the base to be embedded. The derivational prefix
re-, on the other hand, adds the feature [+CI] to certain types of base verbs to produce repetitive verbs, as in
redescend from
descend,
rebuild from
build,
rename from
name (
Lieber 2004, p. 147):
3 | -ery/-age | [+B, +CI ([ ], <base>)] |
4 | re- | [+CI ([ ], <base>)] |
For example,
redescend has the composed skeleton in which the <base> slot of the prefix is saturated with the skeleton of
descend: [+CI ([ ], [+dynamic ([ ])])]. While
Lieber (
2004, chap. 5) does not discuss compounding in this context, below, we argue that a similar analysis can be extended to dvandvas; that is, dvandva compounding extrinsically adds the feature [+CI] to the semantic contributions of the component lexemes.
First, the co-occurrence with the collective classifier
kumi ‘group’ shows that additive dvandvas refer to spatially bounded objects; that is, they carry the [+B] feature.
(8) | a. | hitokumi | no | oya-ko |
| | one-group | gen | parent-child |
| | ‘parents and children as one group’ |
| b. | hutakumi | no | oya-ko |
| | two-group | gen | parent-child |
| | ‘two groups of parents and children’ |
| c. | hyakkumi | no | oya-ko |
| | one-hundred-group | gen | parent-child |
| | ‘100 groups of parents and children’ |
The fact that
oya-ko is also associated with the feature [+CI] is confirmed by various tests. For example, compound verbs headed by
aw- ‘meet’ require a plural subject (
Yumoto 2005, p. 201), as shown by the following minimal pair (
atta is the final realization of the combination of
aw- and the past-tense suffix
-ta):
(9) | a. | Taroo | to | Hanako | ga | {hure-atta | / warai-atta} |
| | | and | | nom | touch-meet.pst | smile-meet.pst |
| | ‘Taro and Hanako {touched each other/smiled at each other}.’ |
| b. | *Taroo | ga | | | {hure-atta | / warai-atta} |
| | | nom | | | touch-meet.pst | smile-meet.pst |
In this construction, the syndetic coordinative phrase can be replaced by an additive dvandva compound, as follows:
(10) | a. | Hitokumi | no | oya-ko | ga | {hure-atta | / warai-atta}. |
| | one-group | gen | parent-child | nom | touch-meet.pst | smile-meet.pst |
| | ‘A parent and her child {touched each other/smiled at each other}.’ |
| b. | Hanako | no | te-asi | ga | hure-atta. | |
| | | gen | hand-foot | nom | touch-meet.pst | |
| | ‘Hanako’s hand and foot touched each other.’ |
In (10), a singular number or possessive phrase is added to the dvandva subject to ensure that there is only one collective set.
Next, while the non-additive types of dvandva nouns do not naturally co-occur with a collective classifier such as kumi ‘group’, it is certain that they are conceived of as being composed of separable similar internal units. This suggests that non-additive dvandvas are like lexically plural nouns with the [−B, +CI] features, such as cattle and sheep.
Studies on number-marking languages report that countable dvandvas exhibit lexical plural marking, as illustrated below by (11) Sanskrit examples from
Whitney (
[1879] 1962, p. 485), (12) Modern Greek examples from
Ralli (
2019, p. 7), and (13) Mordvin examples from
Wälchli (
2005, pp. 137, 139).
(11) | a. | hastyaśvau | Sanskrit |
| | elephant (hastin-)-horse (aśva-).du | |
| | ‘elephant and horse’ | |
| b. | hastyaśvāḥ | |
| | elephant-horse. pl | |
| | ‘elephants and horses’ | |
(12) | a. | jinek-o-peδa | Modern Greek |
| | woman-CM-child.pl | |
| | ‘women and children’ | |
| b. | maxer-o-piruna | |
| | knife-CM-fork.pl | |
| | ‘cutlery’ | |
| c. | ader-o-sikota | |
| | intestine-CM-liver.pl | |
| | ‘intestines and livers’ | |
(13) | a. | t’et’a.t-ava.t | Mordvin |
| | father.pl-mother.pl | |
| | ‘parents’ | |
| b. | ponks.t-panar.t | |
| | trousers.pl-short.pl | |
| | ‘clothing, clothes’ | |
The number markings in (11–13) are lexical rather than grammatical because they do not count the number of the referent of the whole construction, as is usually observed in countable endocentric NN compounds. That is, the grammatical number marking of
prayer books, for example, refers to the plurality of prayer books. In contrast, in (11a), the dual suffix does not mean that there are two elephant-horse sets; rather, it counts the number of homogenized set members: an elephant as one such member + a horse as another such member = two members. Similarly, in (13a), where multiple exponence (
Harris 2017) is observed, the plural suffixes are concerned with the plurality of set members, not sets (dual is already lost at this stage of the language (
Corbett 2000, p. 203)). In English,
Anglo-Saxon presents a similar case; the following paraphrase of this dvandva by
Renner (
2008, p. 609) suggests that the suffix
-s refers to the summation of Angles and Saxons:
(14) | a. | Anglo-Saxons are Angles plus Saxons. |
| b. | *An Anglo-Saxon is an Angle plus a Saxon. |
In conclusion, it seems safe to say that the feature [+CI] is the overarching onomasiological characteristics of dvandvas. This finding can be explained if, as mentioned above, dvandva compounding extrinsically adds the feature [+CI] to the semantic contributions of the component lexemes. For example, the additive dvandva in (4a),
oya-ko ‘a parent and child’, would be produced as follows:
(15) | [+material ([ ])] | + | [+material ([ ])] | → | [+CI ([ ], [+material ([ ])], [+material ([ ])])] |
| oya | | ko | | | oya | ko |
The output compound
oya-ko is exocentric precisely because there is no overt morph matching with the added feature [+CI] (although the lexical plural markers in (11–14) may be such morphs).
If this analysis is on the right track, the category-neutral nature of the quantity-related features (see the definitions above) and the impossibility of [+B, +CI] in the verbal domain (see
Table 1) lead to the following predictions: (i) there should be VV dvandvas that behave like [−B, +CI] words; (ii) there should be no VV dvandvas that behave like [+B, +CI] words.