3. Theoretical discussions
3.4 Summary: Triangle relationship
spoken in Brazil is attested to have a gender system and three sub-types of classifiers (Aikhenvald, 1994). Other examples of genders and classifiers coexistence can also be observed in various languages such as Tidore (van Staden, 2000:77-81) or Ngan’gityemerri which shows the development from generic classifiers into genders (Reid, 1997). Further statistical evidence is provided in the GIS section.
3.3 Genders and plural markers
The convergence of genders and plural markers is motivated by the fact that they both carry different semantic functions. Genders fulfill semantic classification while plural markers represent countability marking. The two systems separately fulfill the two main classifications present in noun phrases, leading us to the expectation that they should co-occur in languages. This is actually what is observed in most languages of the world, taking French as an example, gender marking is existent as demonstrated in the previous sections. Furthermore, plural marking is also realized through suffixes, as demonstrated in Table 4.
Table 4. Genders and plurals co-occurrence in French
Singular Plural
Syntactically this is also possible, since genders and count/mass distinction are supposed to be assigned at the lexical level rather than by syntactic processes (Her, 2012). Moreover, the semantic functions and syntactic forms of the two elements are both different therefore no conflict occurs among them. As a resume, we realize that genders and plural markers separately fulfill the two main classification functions of noun phrases therefore they are expected to complement each other, as a contrast to numeral classifiers which merge these two functions into one entity.
3.4
Summary: Triangle relationship
We propose that despite their apparent divergence, the three elements display a particular distribution in languages due to their common features. As displayed in
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23
Table 5, the noun phrase normally requires: countability (mass/count distinction) and noun classification. First, numeral classifiers carry both simultaneously, as in the example where the numeral classifier assigns the noun to the category of ‘long-shape objects’ and highlights the fact that it is countable. If it was a mass noun, the noun phrase would use a measure word (quantifier) instead of a numeral classifier, e.g. shi bang rou ‘ten M-pound meat’ (Li & Thompson, 1981). Second, genders provide semantic classification to facilitate referent tracking, as demonstrated in the introduction (Luraghi, 2011; Contini-Morava & Kilarski, 2013). Finally, grammatical plural markers point out the countability of the noun, as in English where only count nouns can take plural marking, e.g. ‘some tables’ but ‘some water’ (Sanches & Slobin, 1973; Greenberg, 1990; Ghomeshi & Massam, 2012).
Table 5. Semantic functions of numeral classifiers, genders and plural markers Countability Classification Example
Num classifier Yes Yes Chinese ‘one CL-long fish’
Gender No Yes French ‘table, fem’/ ‘book, mas’
Plural marker Yes No English –s ‘three tables’
From features in Table 5 we can obtain the relationship graph in Fig 1. We hypothesize that in a language, numeral classifiers and plural markers are mutually exclusive (line 1), numeral classifiers and genders have the tendency to not co-occur (line 2) while genders and plural markers are expected to co-occur (line 3).
Figure 1. Interaction of numeral classifiers, genders and plural markers
In line 1, numeral classifiers and plural markers both mark countability in semantic function and syntactic form, therefore should not occur together (Borer, 2005; Hsieh, 2009; Her & Chen, 2013). In line 2, numeral classifiers and genders share the semantic function of classification so are expected not to co-occur (Dixon, 1986:111, Blench, 2012). However, since the overlap is in semantic function rather than syntactic form, the redundancy is acceptable but would occur in low frequency
‧
classifier. However, they have the same semantic function of classification, making the overlap redundant, thus dispreferred. In line 3, genders and plural markers bear separately the two types of information: countability and classification. Therefore, they should appear in the same environment, unless another carrier fulfilling the same purpose is found, e.g. noun classifiers, morphological case marking, among others.This exceptional situation is demonstrated in the typological section. A typological example would be that the lack of gender and number markers in Sino-Tibetan, Miao-Yao and Tai nouns is somewhat compensated for by numeral classifiers, which serve to individuate nouns and may be said to ‘agree’ with particular classes of nouns e.g. flat, round, elongated objects among others (Matisoff, 1991:496).
It is necessary to explain that in line 1 the two elements are mutually exclusive since both numeral classifiers and grammatical plural markers share the same semantic function and syntactic form, while syntactic stacking is not possible. Moreover, they distinguish between two separate meanings, e.g. in English ‘book’ and ‘books’ are definitively different, but marking the mass/count distinction twice as in ‘*bookss’
would not add anything new to the word, making the redundancy useless. However, for line 2 the shared feature of noun classification is not in syntactic form but in semantic classification and does not distinguish a meaning, rather it is an optional system of noun classification for referent tracking. In this sense, one could add as many categories as one wants in order to obtain a maximum of precision, as it would be when adding adjectives to a noun, e.g. the big yellow French book. A similar process with noun classification would result in something like ‘the book, masculine/
inanimate/ square-shape’. This makes the overlap acceptable, but due to economy tradeoff in language, it is still expected to appear in low frequency. In other words, since the purpose of noun classification is to highlight a specific feature of the noun, adding more than one would cancel the stand-out effect of a feature, therefore it is dispreferred.
Combining the deduction mentioned previously, we may display the two functions via the concept of continuum: each language is searching for a balance between economy and expressiveness. For the speaker, the more economical the better, since less energy is invested in speaking. On the opposite side for the listener, the more expressive the better, since less energy is needed to interpret the message. In other words, countability and classification are two sources of information available for noun phrases, the most equilibrated situation occurs when both functions are fulfilled once
‧
each, while marking only one of it or neither of them leads to more economy but less expressiveness and vice-versa, as demonstrated in (22).
(22) Equilibrium continuum between economy and expressiveness within languages
Economy Expressiveness
<< --- >>
Countability ✓ ✓ ✓ ✓✓ ✓✓
Classification ✓ ✓ ✓✓ ✓ ✓✓
Language Usan Piraha Finnish Chinese/
French
Tidore Hungarian Marathi
The most equilibrated point is expected to occur in the middle, when both countability and classification are fulfilled once, e.g. Chinese has numeral classifiers which carry countability and classification marking once, French relies on genders for classification and plural markers for countability. Other distributions may occur but the frequency is predicted as low. As an example, when the language prefers economy over expressiveness, it would only include one function between countability and classification, e.g. the Piraha language from the Amazonas only shows countability marking, while Finnish only marks countability via plural markers. A most extreme example in this direction would be Usan, a language from Papua New Guinea, which does not mark either of the two functions. On the other hand, a language may also tend to enhance expressiveness. In this case it would mark one function twice, e.g. in Tidore noun classification is fulfilled once by numeral classifiers and a second time by genders. Most extreme cases also exist, as in Marathi which is attested to have numeral classifiers, genders and plural markers, apparently resulting in double marking for countability and classification. Our speculation regarding the high ratio of balanced languages in the middle of the continuum is further supported by our data in the GIS section, but it is also necessary to point out that beside countability marking and noun classification, it is possible that other functions may be required in the noun phrase. Moreover, the hierarchy between the two main functions involved in (21) also needs further discussion. Due to resource limitation and avoidance of a too large scale for this paper, these issues are left for further research to develop.
As a summary, we expect that the two functions of countability (mass/count) and classification may explain the distribution of numeral classifiers, grammatical genders and grammatical plural markers. First, countability marking is syntactic therefore it is not possible to show stacking, following this logic we propose that numeral classifiers
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
26
and grammatical plural markers should be mutually exclusive. Second, noun classification is semantic therefore it is possible to be represented in more than one form. Thus, to avoid losing the function of highlighting a noun’s feature it would still be dispreferred. Under this hypothesis, numeral classifiers and grammatical genders are expected to show the tendency of not co-occuring. Finally, since grammatical genders and grammatical plurals separately fulfill countability and noun classification, they are not affecting each other and are expected to co-occur. It is also necessary to point out that numeral classifiers, grammatical genders and grammatical plural markers may not be the only carriers fulfilling the function of countability and classification marking, e.g. noun classifiers, grammatical case marking are also attested to be plausible candidates. However, in this study we first focus on the most common three categories of numeral classifiers, genders and plural markers to obtain a general overview. A brief analysis of noun classifiers and case marking is still presented in the typological evidence section, nevertheless more detailed research in this subject is required for further studies.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
27
Chapter 4 Supporting evidence
This section is divided in three parts: first we briefly explain the methodology of data gathering. Second, we run statistical typological analysis to verify our hypothesis in the languages of the world. Third, based on GIS system, we display and explain the areal distribution and historical development of numeral classifiers, grammatical genders and grammatical plural markers in the world’s languages.
4.1 Methodology
To gather sufficient data and test our hypothesis, we choose to expand our coverage as much as possible to get a big picture of the languages in the world, and then we go into details for each language. Two main criteria are set in terms of language coverage:
speaker population and language genealogy. For speaker population, we aim at including in our sample a quantity of speakers reaching 60% of the world population, as to follow the actual tendency of the actual world. The second condition of language genealogy is motivated by the fact that language diversity must also be taken into consideration. As an example, English is one of the most spoken languages in the world however it may be resulting from the influence of domains such as economy and politics rather than being solely dependent on a well-structured language system.
Therefore, we need to cover different language groups to test if our hypothesis does predict a correct tendency. Our preliminary data contained 80 languages with their information on numeral classifiers, genders and plural markers. It was obtained via the merge of three biggest open-source database: Gil (2013) for numeral classifiers, Corbett (2013) for genders and Haspelmath (2013) for plural markers, as displayed in Figure 2.
Figure 2. Preliminary database of 80 languages
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
28
Then, we checked which main language groups or geographical regions were not sufficiently covered in these 80 languages to enlarge our inventory obtaining a database of 155 languages, as displayed in Map 7 with each dot representing a language. This result is obtained through the combination of the biggest available open source database: Gil’s (2013) survey on numeral classifier languages, Corbett’s (2013) discussion on genders, Haspelmath’s (2013) research on plural markers, SIL (Ethnologue) language records and information from professor Her research team in the syntax lab of National Chengchi University, e.g. data on 454 numeral classifier languages. The detailed list is provided in Appendix 1.
Map 7. Display of 155 surveyed languages (by dots)
In terms of world population the speakers of the 155 languages we gathered reached 65% of the world’s population according to SIL data, while in terms of language genealogy we covered 30% (39/129) of the main language groups in the world, e.g.
Indo-European, Austronesian, Sino-Tibetan among others. Our genealogical coverage may seem insufficient, however we already included in our study 90% (18/20) of the top 20 biggest main language groups in the world (Indo-European, Sino-Tibetan, Afro-Asiatic, Niger-Congo, Austronesian, Dravidian, Japonic, Altaic, Austro-Asiatic, Tai-Kadai, Creole, Nilo-Saharan, Uralic, Quechuan, Hmong-Mien, Mayan, North Caucasian, Language isolates), which account for 99% of the world speakers population according to SIL, therefore we estimate it sufficient. Following the distribution on the map in Map 5, we may realize that Africa, Europe and Asia are covered however Australia and the Americas would need more representative languages. The main reason for this point is that the literature is quite controversial on whether the languages in these areas rely on numeral classifiers, noun classifiers, genders (noun classes) or other types of systems, therefore to avoid incorrect
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
29
judgment of categorization, we do not include languages with different structures attested. Finally, to assure that our coverage in terms of language speakers is sufficient, in Map 8 we also display the language speaker distribution via polygons of the QGIS software. The black area represents the presence of speakers from the 155 languages, showing that in demographic terms we do cover the main parts of the globe, since every continent is colored.
Map 8. Display of 155 surveyed languages (by speaker population)
Before going into the details of the typological evidence it is also necessary to highlight that some records of previous studies database have been modified within our coding. One example would be Chinese. In Gil (2013), Chinese is attested to have numeral classifiers, and noted by Corbett (2013) as genderless which is indeed supported by our data. However, Haspelmath (2013) database on nominal plural categorize Chinese as having optional plural marking on human nouns. As a reminder, we defined that our study only targets grammatical plural markers, this criteria automatically rules out Chinese since it is a language without number agreement.
Therefore in our database we count Chinese as without plural markers. A possible source of confusion could be the Chinese collective marker men, but since it is more analyzed as a clitic without grammatical agreement rather than a grammatical plural marker attached to the noun, we do not count it as a plural marker (Rijkhoff, 2000:240). Further evidence on agreement and definiteness are shown in (23).
(23) Collective marker men in Chinese a. Phrase with one teacher
一 位 老師 出去 了 yi wei laoshi chuqu le one CL-person teacher go out PFV
‘One teacher went out.’
‧
c. Conflict of indefinite marker and men
*那裡 有 一些 老師 們 nali you yixie laoshi men there is some teachers Collective
‘There is some the teachers there.’
As demonstrated in (23a-b), when men is attached to the noun lao shi ‘teacher’, it seems to reflect plural however it does not have grammatical agreement with the verb.
Moreover, men shows definiteness, as shown in (23c) with the impossible co-occurrence with indefinite marker such as yi xie ‘some’, which is different from the conventional plural markers, e.g. with -s in English, teachers would be indefinite and some teachers entirely grammatical. Therefore, the men is interpreted here as a collective marker which highlights the homogenous feature of the group members, rather than being a grammatical plural marker. As a reminder, the detailed coding and references for each language are provided in Appendix 1.
4.2 Typological evidence
Following our assumptions in Table 5, we know that mathematically eight combinations are possible for the presence (＋) or absence (－) of numeral classifiers, genders and plural markers. According to our concept of continuum in (22), only two of them should be optimal in the world languages, with countability and classification fulfilled once either by numeral classifiers alone (Chinese) or the combination of genders and plural markers (French). Three are under-marked: if the language favors economy, one or both of the two functions may not be marked, resulting in under-marking of noun information, e.g. languages only marking countability with plural markers, just marking classification via genders or not marking either of the two. Finally, three are over-marked: if the language prefers expressiveness, one or both of the two functions may be marked twice, e.g. languages having numeral classifiers and genders would experience an overlap of noun classification, while languages with numeral classifiers and plural markers display double marking of countability. The last extreme situation would be languages attested with numeral
‧
countability and classification. The result of categorizing the 155 languages of our database by the eight combinations mentioned is displayed in Table 6.Table 6. Different combinations of numeral classifiers, genders and plural markers Plural Num CL Gender
Combination 1 and 2 both have the two main marking of noun phrases: countability and noun classification represented once therefore are optimal. For combination 3, it only has the marking of countability (plural markers), however since semantic classification (genders) is rather optional, its absence can be tolerated. The similar logic applies to combination 4, which only has the noun classification represented by genders. Combination 5 includes languages with the most economic but least expressive system without either marking. On the other hand, combination 6 has a semantic redundancy of noun classification through the co-occurrence of numeral classifiers and genders, thus is also possible but marked. Combinations 7 and 8 have redundancy of countability marking (numeral classifiers and plural markers) which is also marked, as explained in the theoretical discussions.
In terms of typological distribution, the results show that 87% of the data is in accordance with our predictions. Furthermore, the anomalies can be explained when analyzing the details. We provide a quick overview first in the following while further details are demonstrated in the incoming sections. For category 7 and 8, two situations can generally be found: first, the languages are attested to have numeral classifiers but their inventory is very limited, as in German which only has one optional numeral classifier attested (Sussex and Cubberley, 2006) or Tuvaluan with just some classifier-like elements (Besnier, 2000; Gil, 2013). Second, even though the language does have numeral classifiers and plural markers, they do not co-occur on the noun, as
‧
in Turkish (Kornfilt, 1997; Göksel and Kerslake, 2011) and Hungarian (Csirmaz and Dekany, 2014). Therefore it does not create a major violation and most of these languages could actually belong to category 1 and 3. For category 3 and 6, the absence or redundancy of semantic classification is possible, as explained in the theoretical discussion. Finally for category 4 and 5, the languages involved all have restricted numeral systems, as in Usan up to ten, or Piraha, which is a rare language attested not to have the concept of numerals (Everett, 1986; Corbett, 2000), explaining the lack of grammatical plural markers.
As a summary, the special cases are not real exceptions if applying strict terms. We show this fact in the following sections by analyzing the correlation of numeral classifiers, grammatical genders and grammatical plural markers among category 1-8 and decorticating the evidence and possible violations via language samples.
Following our categorization in Table 6, this step is further divided into three sections:
optimal marking, under-marking and over-marking of countability & noun
optimal marking, under-marking and over-marking of countability & noun