1

th
7 Summer School of the German Linguistic Society/ Walter Bisang, Mainz
7. Sommerschule der DGfS [email protected]

Typology 4
___________________________________________________________________________

Universals II: Hawkins, Dryer

4. John A. Hawkins
4.1. Hawkin’s (1983) sample and his language types
Hawkin’s (1983) is based on three samples:
- Greenberg’s 30 language sample
- Greenberg’s expanded sample (142 languages)
- His own sample, which he calls Expanded Sample and consists of 336 languages

Type Nr Values of the 4 parameters Number of languages

1 V-1/Pr/NG/NA 38
2 V-1/Pr/NG/AN 13
3 V-1/Pr/GN/AN 1
4 V-1/Pr/GN/NA 0
5 V-1/Po/NG/NA 0
6 V-1/Po/NG/AN 0
7 V-1/Po/GN/AN 1
8 V-1/Po/GN/NA 0
9 SVO/Pr/NG/NA 56
10 SVO/Pr/NG/AN 17
11 SVO/Pr/GN/AN 7
12 SVO/Pr/GN/NA 4
13 SVO/Po/NG/NA 0
14 SVO/Po/NG/AN 0
15 SVO/Po/GN/AN 12
16 SVO/Po/GN/NA 13
17 SOV/Pr/NG/NA 10
18 SOV/Pr/NG/AN 0
19 SOV/Pr/GN/AN 2
20 SOV/Pr/GN/NA 0
21 SOV/Po/NG/NA 11
22 SOV/Po/NG/AN 0
23 SOV/Po/GN/AN 96
24 SOV/Po/GN/NA 55

Table 8: Frequency table for 24 language types (Hawkins 1983: 288)

 2

The present study is based on a language sample of some 350 languages, taken
from all the major language families of the globe ... This sample has incorporated
Greenberg’s (1966) samples as a starting point, and has considerably expanded
them. The particular emphasis within these languages is on approximately one
dozen word orders consisting of pairs of modifier + head: adjective and noun,
direct object and verb, preposition and NP, etc. For five of these pairs I have data
from all 350 languages (adjective and noun, genitive and noun,
preposition/postposition and noun phrase, object and verb, and subject and verb).
For the remainder (demonstrative determiner and noun, relative clause and noun,
etc.), I have data from between one-third and one-half of the sample. The same
methodological problems apply in this work which apply in all large-scale
typological studies: Only a limited number of properties (here word order) can be
studied; the grammars upon which we rely are not always dependable; the
categories that we study are not always readily comparable across languages ...;
and some word orders exhibit variant orders. (Hawkins 1983: 9)

4.2. The properties of Hawkin’s universals

Hawkin’ s universals are:
- They are absolute (that is, non-statistical)
- Implicational
- Multitermed
The implicational universals to be proposed in this chapter have the following
properties: they are almost all NONSTATISTICAL (I.E., EXCEPTIONLESS)
RATHER THAN STATISTICAL, relative to the data base; they are all
UNILATERAL (if P then Q) RATHER THAN BILATERAL (if P then Q and vice
versa, as in Vennemann' s NSP); and a significant number are MULTITERMED
RATHER THAN BITERMED (i.e., defined in terms of at least three rather than
just two properties. (Hawkins 1983: 59)

The reason for only accepting absolute/non-statistical universals:

Consider just the word orders exemplifying the operator-operand orderings listed
in the discussion of Vennemann' s theory. There are 17 in all. If we add indirect
object + verb, oblique NP + verb, and also subject + verb ..., we have a total of 20
word order pairs, each of which can be serialized with the operator either before or
after the operand. As these orders are for the most part logically independent of one
another, we have some 220, or over a million possible co-occurrences; and these 20
do not begin to exhaust all the definable word orders of a language.
Now assuming ... that the number of currently existing languages are in the region
of 4000-8000, even if each of them were to have a different co-occurrence array
from every other (i.e., different with respect to at least one word order), the total
number of attested co-occurrence types could not exceed 8000, which would
represent just .76% of the mathematical possibilities for 20 varying word orders.
And if ... the number of possible co-occurrence types is more like many hundreds
of millions, the actually attested co-occurrences could be no more than a tiny
fraction of 1% of these.
There is therefore a trivial sense in which a huge discrepancy must exist between
the mathematical possibilities and the actually attested co-occurrences - there are
not enough languages to go around. (Hawkins 1983:61)
 3

Because Hawkin’ s operates with multitermed universals, he can only use absolute
implicational universals. If statistical implicational universals were combined into
implicational chains (multitermed universals) the multiplication of their probability rate would
soon lead into statistical irrelevance.

4.3. Consequences of Hawkin’s research

1. SVO is no longer a type indicator; that is, nothing correlates with SVO in a
unique and principled way, according to our evidence. There are, of course, many
languages with SVO, but there is no "SVO-type."
2. VSO and SOV are type indicators, but limited ones. Much better and more
general type indicators are prepositions and postpositions.
3. The whole notion of a "word order type" becomes more abstract. The set of
languages comprising a common type no longer share all of a given set of word
order properties. ... The set of languages which are considered to belong to a
common type (e.g., prepositional languages) may now vary in many (even most)
of their basic word orders. But they must all conform to a relatively restricted set
of word order subtypes obeying certain general regularities (e.g. the subtypes of
the Prepositional Noun Modifier Hierarchy, ...), each of which contains a shared
common property functioning as the typological indicator (here prepositions)."
(Hawkins 1983:16f.)

4.4. Some examples

(I’) If a language has OV word order, then if the adjective precedes the noun, the genitive
precedes the noun; i.e., OV ⊃ (AN ⊃ GN).
(II’) If a language has verb-first order, then if the adjective follows the noun, the genitive
follows the noun; i.e., V-1 ⊃ (NA ⊃ NG)
(III’) If a language has Prep and any verb position other than SVO, then if the adjective
follows the noun, the genitive follows the noun; i.e., Prep & -SVO ⊃ (NA ⊃ NG)
(IV) If a language has Postp word order, and if the adjective precedes the noun, then the
genitive precedes the noun; Postp ⊃ (AN ⊃ GN).

(XIV): PREPOSITIONAL NOUN MODIFIER HIERARCHY (PrNMH)

Prep ⊃ ((NDem » NNum ⊃ NA) & (NA ⊃ NG) & (NG ⊃ NRel))

This hierarchy reflects the relative instability with which the five modifiers of the
noun (Demonstrative, Numeral, Adjective, Genitive, Relative Clause) keep
modified-modifier word order. It states that Dem and Num are more unstable than
Adj, Adj is more unstable than Gen and Gen is more unstable than Rel. Thus, if in a
prepositional language, NP structure deviates from the modified-modifier word
order, it is Dem and/or Num that occur first in the prenominal position. If they are in
the prenominal position the next category that can occur prenominally is Adj, etc.
This yields the following hierarchy (cf. handout 3, § 1.2):
(33) Rel < Gen < Adj < {Dem, Num}
Universal (XIV)/Hierarchy (33) predicts that only 7 of the logically possible 32 (=
25) structural word-order types are attested:
 4

1. Prep & NDem & NNum & NA & NG & NRel 0 modifiers preposed: N1
2a. Prep DemN NNum NA NG NRel 1 modifier preposed: N2
2b. Prep NDem NumN NA NG NRel 1 modifier preposed: N2
3. Prep DemN NumN NA NG NRel 2 modifiers preposed: N3
4. Prep DemN NumN AN NG NRel 3 modifiers preposed: N4
5. Prep DemN NumN AN GN NRel 4 modifiers preposed: N5
6. Prep DemN NumN AN GN RelN 5 modifiers preposed: N6
Table 9: The 7 word-order types in terms of the Prepositional Noun Modifier Hierarchy (Hawkins 1983:75)

5. Matthew S. Dryer
The aim of Dryer’ s (1992) studies:
- What are the pairs of elements whose order correlates with that of the verb and object? (cf. §5.1)
- Why do these correlations exist? (cf. §5.2)
For this purpose, none of the criteria postulated by Hawkins (absolute, implicational,
multitermed universals) is necessary. What is needed for analysing the above question is a
sound statistical method.

5.1. Dryer’s correlation pairs

Unlike Hawkins (1983), Dryer (1992) operates with statistical tendencies. On the basis of his
method (cf. handout 2, § 3) he wants to find out what pairs of elements correlate with VO/OV
word order.

Africa Eurasia SE-As & Ocean Austr-Neu Gui N-Am S-Am Average
OV & Po 15 26 5 17 25 19 107
OV & Pr 3 3 0 1 0 0 7
VO & Po 4 1 0 0 3 4 12
VO & Pr 16 8 15 6 20 5 70
Table 10: The order of adposition and NP in correlation with VO and OV (Dryer 1992: 83)
Since the type OV & Po invariably dominates over OV & Pr through all the six large areas and
since VO & Pr does the same with regard to VO & Po, the two parameters of VO/OV and
Pr/Po covary, that is, they are a correlation pair.

(34) a. V O b. O V c. <Adposition, NP>

Pr NP NP Po

The position of the adposition covaries with the position of V

=> the adposition is a verb patterner
The position of the NP covaries with the position of O
=> the NP is an object patterner
 5

The correlation pair <Adposition, NP> is consistent through all six large areas. Are there
also correlation pairs which are to some extent inconsistent?
Africa Eurasia SE-As & Ocean Austr-Neu Gui N-Am S-Am Average
OV & RelN 5 11 2 2 3 3 26
OV & NRel 9 5 2 6 12 3 37

VO & RelN 0 0 1 0 0 0 1
VO & NRel 21 8 12 3 11 5 60

Table 11: Order of noun and relative clause (Dryer 1992:86)

From looking at the first two lines in table 11 (OV & RelN, OV & NRel) we may conclude
that the position of Rel relative to N does not form a correlation pair with VO/OV. However, if
we look at the second two lines (VO & RelN & VO & NRel) we observe a very strong and
areally consistent preference of NRel in VO languages (60 genera against 1, the language with
VO & RelN is Chinese). The statistical significance of these findings can be shown as follows:
Africa Eurasia SE-As & Ocean Austr-Neu Gui N-Am S-Am Average
OV 0.36 0.69 0.50 0.25 0.20 0.50 0.42
VO 0.00 0.00 0.08 0.00 0.00 0.00 0.01

Table 12: Proportions of genera containing RelN languages as opposed to NRel

In table 12, the percentage of genera of type OV & RelN is compared to the percentage of ge-
nera of the type VO & RelN. From table 11 we can see that in Africa, there are 5 genera of the
type OV & RelN and 9 genera of the type OV & NRel. The percentage of OV & RelN in OV
languages of Africa is thus 0.36 (= 5 : [5 + 9]). If we proceed like this throughout table 11, we
get at table 12. As we can see, the percentage of RelN is higher in all six large areas with type
OV than with type VO. With this areally consistent result the higher frequency of RelN in OV
is statistically significant. Dryer (1992) treats cases like these as correlation pairs too: <N,
Rel>.
If a pair of elements X and Y is such that X tends to precede Y significantly more
often in VO languages than in OV languages, than <X,Y> is a CORRELATION
PAIR and X is a VERB PATTERNER and Y an OBJECT PATTERNER with
respect to this pair. (Dryer 1992: 87)

With this method, Dryer (1992) finds the following correlation pairs:
VERB PATTERNER OBJECT PATTERNER EXAMPLE
(1) verb object ate + the sandwich
(2) verb subject (there) entered + a tall man
(3) adposition NP on + the table
(4) copula verb predicate is + a teacher
(5) ’want’ VP wants + to see Mary
(6) tense/aspect auxiliary verb VP has + eaten dinner
(7) negative auxiliary VP
(8) complementizer S that + John is sick
(9) question particle S
(10) adverbialer subordinator S because + Bob has left
(11) article N’ the + tall man
 6

(12) plural word N’

(13) noun genitive father + of John
(14) noun relative clause movies + that we saw
(15) adjective standard of comparison taller + than Bob
(16) verb PP slept + on the floor
(17) verb manner adverb ran + slowly
Table 13: Complete list of correlation pairs (Dryer 1992: 108)

Some noncorrelation pairs:

DEPENDENT HEAD EXAMPLE
(1) adjective noun tall + man
(2) demonstrative noun that + man
(3) intensifier adjective very + tall
(4) negative particle verb not + go
(5) tense/aspect particle verb
Table 14: Noncorrelation pairs (Dryer 1992: 108)

5.2. Dryer’s explanation for the correlation pairs

Dryer compares two potential explanations for his results concernig correlation pairs:
(35) The HEAD-DEPENDENT THEORY (HDT):
Verb patterners are heads and object patterners are dependents. That is, a pair of elements
X and Y will employ the order XY significantly more often among VO languages than
among OV languages if and only if X is a head and Y is a dependent. (Dryer 1992: 87)
(36) The BRANCHING DIRECTION THEORY (BDT):
Verb patterners are nonphrasal (nonbranching, lexical) categories and object
patterners are phrasal (branching) categories. That is, a pair of elements X and Y
will employ the order XY significantly more often among VO languages than
among OV languages if and only if X is a nonphrasal category and Y is a phrasal
category.

As it turns out, the Head-Dependent Theory makes wrong predictions concerning the question
of what is a correlation pair and what is not. The Branching Direction Theory does not.

5.2.1. The Head Dependent Theory

Elements that can be explained by the Head Dependent Theory

VERB PATTERNER OBJECT PATTERNER EXAMPLE

(=head) (=dependent)
(1) verb object ate + the sandwich
(3) adposition NP on + the table
(13) noun genitive father + of John
(14) noun relative clause movies + that we saw
(14) noun relative clause movies + that we saw
(15) adjective standard of comparison taller + than Bob
 7

(16) verb PP slept + on the floor

(17) verb manner adverb ran + slowly
(4) copula verb predicate is + a teacher
(5) ’want’ VP wants + to see Mary
Table 15: correlation pairs that can be explained by the BDT
Instances which do not form correlation pairs against the predictions of the BDT:

Cf. table 14.

Example: Adjective / Noun:

Africa Eurasia SE-As & Ocean Austr-Neu Gui N-Am S-Am Average
OV & AdjN 7 24 2 4 10 8 55
OV & NAdj 18 4 5 15 18 14 74

VO & AdjN 3 6 4 5 19 3 40
VO & NAdj 25 3 12 2 8 5 55
Table 16: Order of noun and adjective (Dryer 1992: 95)

Africa Eurasia SE-As & Ocean Austr-Neu Gui N-Am S-Am Average
OV 0.28 0.86 0.29 0.21 0.36 0.36 0.39
VO 0.11 0.67 0.25 0.71 0.70 0.38 0.47
Table 17: Proportions of genera containing AdjN languages as opposed to NAdj
(Dryer 1992: 95)

Controversial pairs (correlation pairs where head and dependent are controversial)

VERB PATTERNER OBJECT PATTERNER EXAMPLE

(6) tense/aspect auxiliary verb VP has + eaten dinner
(7) negative auxiliary VP
(8) complementizer S that + John is sick
(9) question particle S
(10) adverbialer subordinator S because + Bob has left
(11) article N’ the + tall man
(12) plural word N’
(2) verb subject (there) entered + a tall man

5.2.2. The Branching Direction Theory

Correlation pairs Noncorrelation pairs
(14) <noun, relative clause> (A) noun/adjective
(15) <adjective/ standard of comparison> (C) Adjektiv/Verstärker
 8

(16) <verb, PP> (D) Verb/Negativpartikel

From the perspective of the Head-Dependent Theory, the noncorrelation pairs should also turn
out to be correlation pairs. Given the different status of phrasality in the correlation pairs and
the noncorrelation pairs, the BDT predicts the above differences:
On (14) vs. (A):
relative clauses are phrasal categories that are fully recursive, adjectives are not.
On (15) vs. (C):
The standard of comparison is phrasal (= object patterner), the intensifier is not.
On (16) vs. (D):
The PP is phrasal (= object patterner), the negative particle is not.

6. Outlook: Word order typology in terms of minimalism

Different word order accross languages is the result of two different types of features which
need to be checked in a tree structure of the following type (cf. next page) and thus trigger
movement.
As we have seen in handout 1 (pp. 3f.), features (grammatical features) need to be ckecked
before spell-out for a derivation to converge (not to crash). The basic idea is that there are two
different types of features:
- Strong features: the lexical unit moves together with ist phonological features
- Weak features: only grammatical and logical/semantic features move, the
phonological features remain in situ.
Verbal features are checked at a V adjoined to a functional head, nominal features are checked
at a DP in the specifier position of AgrP (maybe TP).

(37) CP
C’
C AgrSP
AgrS’
AgrS TP
T’
T vP
DP v’
v AgrOP
DP AgrO’
AgrO VP
DP V’
 9

V DP

The combination of weak and strong nominal and verbal features in Agr and T allows for a
formal explanation of word-order types. The number of word possible word orders is much
larger than the six options with S, V, and O. From the formalist perspective, the six basic
word-order types in terms of Greenberg (1966) are only epiphenomenal. They are the
product of innate properties of UG which are much more complex.
An example:

In English, Agr features are weak. Thus, object-DPs do not move before spell-out and we get
OV. The T-features for subjects, however, are strong. It is for that reason that the subject DP
needs to move before spell-out and we get SV. The verb does not move before spell-out. Thus,
in English adverbs such as often occur in front of V, whereas in French souvent ‚often‘ occurs
after the verb.
(38) Elmer lave souvent son chat.
Elmer washes often his cat
(39) Elmer often washes his cat. (Marantz 1995: 372)

In French, V features are strong in both positions, Agr and T.

References
Dryer, Matthew S. 1992. "The Greenbergian word order correlations." Language 68,81-138.
Hawkins, J.A. 1983. Word Order Universals. New York: Academic Press.Dryer, Matthew S. 1992. "The
Greenbergian word order correlations." Language 68,81-138.
Chomsky, Noam. 1995a. The Minimalist Program. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 1995b. "Bare phrase structure", in: Webelhuth, Gert. ed. Government & Binding Theory and the
Minimalist Program, 383 - 439. Oxford: Blackwell.
Marantz, Alec. 1995. "The Minimalist Program", in: Webelhuth, Gert. ed. Government & Binding Theory and the
Minimalist Program, 351 - 382. Oxford: Blackwell.