The Linear Aspects of Syntax: Ideas for Your
Conlangs
Auteur: Douglas Ball
Date MS: 07-23-2012
Date FL: dix-01-2012
Numéro FL: FL-00000D-00
Citation: Ball, Douglas. 2012. “The Linear Aspects of
Syntaxe: Ideas for Your Conlangs.”
FL-00000D-00, Fiat Lingua,
droits d'auteur: © 2012 Douglas Ball. Ce travail est sous licence
sous une attribution Creative Commons-
Non Commercial-Pas de Dérivés 3.0 Licence non transférée.
!
http://creativecommons.org/licenses/by-nc-nd/3.0/
Fiat Lingua est produit et maintenu par la Language Creation Society (LCS). Pour plus d'informations
à propos du LCS, visitez http://www.conlang.org/
The Linear Aspects of Syntax:
Ideas for Your Conlangs⇤ †
Douglas Ball
Abstrait
As part of an effort to encourage conlangers to explicate the syntaxes of their languages,
this paper discusses several of the most common linear order generalizations found in natural
langues. Among those discussed are the linear order generalizations surrounding heads, le
order of verbal arguments, ordering of elements with certain information statuses, and ordering
by “weight”.
1
Introduction
In many descriptions of conlangs on the web, the section on word order (or syntax, more generally)
is rather short and not all that detailed. While this could have a number of different causes, ce
paper operates on the premise that providing more information about how (a subpart) of syntax
works in natural languages could help conlangers to find ways to extend and elaborate their syntax
sections.
As I see it, this paper could be utilized in one of two ways. For conlangers interested in creating
natural language-like conlangs, this paper provides details that are ripe for “importation” into your
conlang(s). For conlangers interested in stretching the boundaries of language—whether through
alien languages or interesting thought experiments—this work will provide you with a primer of
what not to do. De plus,, it gives you some language parameters that you can play with to alter
and stretch the boundaries of what we know and think about language.
⇤This paper is a non-trivially revised version of my talk “Conlanging and the Linear Aspects of Syntax” from the
first Language Creation Conference in 2006. I appreciate the questions and the comments raised by the participants
of that conference that ultimately improved this paper. Initialement, this talk also discussed the ideas of linearization
théorie, as found in frameworks like some versions of Head-driven Phrase Structure Grammar (HPSG) (Sag, Wasow,
Bender 2003 is a general introduction; Gazdar and Pullum [1981] offer an accessible introduction to linearization
théorie), some versions of Lexical-Functional Grammar (LFG) (Bresnan 2001), and Categorial Grammar (particularly
as proposed by Dowty 1995). While these theories still inform this paper, the original theoretical discussion has been
removed.
†This paper was originally put together in summer 2009.
In revising it in 2012, I have just corrected minor
errors, but otherwise left it as is. This also means that the examples of Skerre are a bit dated, but they remain as a
documentation of this particular version of the language.
1
1.1 Why Linearization?
The question arises why I should be focusing on the linear aspects of syntax. This focus derives
from what seem to be some basic properties of all natural languages. The limits on what natu-
ral languages can package into a single word (which exist for some currently unknown reason)
necessitate that the information be spread out temporally and, more often than not, across many
mots. Par conséquent, all natural languages have multiword expressions. Autrement dit, il y a
no language where a word like havem means “We were surprised that they came over to our house
last weekend for the party.” Certainly, there are languages like the Caddoan language Wichita that
can jam quite a bit into a word, as exemplified in (1):
(1)
Wichita
kiya:k´ıriw´a:cParasarikitaPah´ı:riks
(le)À-
kiya:à-
QUOT.AOR.3-
COLL-
‘He brought the large quantity of meat up to the top.’
riwa:c-
grand-
ri-
PORT-
Paras-
viande-
nous-
haut-
Pennsylvanie-
viens-
hir´ı:k-
ITER-
s
IPFV
(Rood 1976, 65)
But even in Wichita, while sentences can potentially be just one word, not all of the sentences are
just one word long. To express particular meanings, more than one word must be used. This can
be seen in (2), which includes the word in (1) in a sentence:
(2)
Wichita
de:cParPa kiya:k´ıriw´a:cParasarikitaPah´ı:riks niya:hkw´ırih.
riwa:c-
de:cParPa
grand-
écureuil
(h)irih
le-
de:k-
PL-
bois-
LOC
‘Squirrel brought the large quantity of meat up to the top of where the tree was.’ (Rood
1976, 266)
kiya:à-
QUOT.AOR.3-
wi-
stand.upright-
(le)À-
COLL-
ri-
PORT-
Pennsylvanie-
viens-
hir´ı:k-
ITER-
Paras-
viande-
déjà-
PTCP-
s
IPFV
nous-
haut-
So as long as there are sequences of words in natural language, there has to be some sort of
linear order to them. And in fact, the discipline of syntax, as modernly conceived, is at least
partly devoted to characterizing the linear orders in natural languages and trying to come up with
explanations for their patterning.
Et, En fait, most syntactic theories utilize more elaborated data structures than just the “pure”
linear order in their proposed understanding of natural language syntax. But to keep the discussion
at a more basic level, I will focus on linear order generalizations and leave a full elaboration
of why the generalizations are what they are for another venue. I encourage interested readers,
mais, to seek out works that do discuss possible explanations of generalizations; some places to
look include those given in the references section.
1.2 Plan of the Paper
This paper, however ambitious it may seem to be, cannot tackle the whole of the problem of
linear ordering in natural language in one fell swoop. Donc, I focus on four “case studies” of
linearization generalizations. These generalizations cover some of the most pervasive and common
linear order patterns in natural languages, et ainsi, I hope, will be of the most interest to conlangers.
2
The first concerns the linearization of the central word of a group of words—what is called the
“head”—and the other words related to it, its dependents. The second concerns the linearization
of the dependents, in particular a subject known as arguments, internal to themselves. Le troisième
concerns the role of information structure—how new or old to the discussion a linguistic expres-
sion is—in linear ordering generalizations. Quatrième, I discuss the interesting behavior of “heavy”
phrases—expressions with lots of words—within clauses.
2 Heads and Linear Ordering
A notion that has emerged as a particularly useful one for the study of syntax—one that occurs in
various guises in various frameworks—is the notion of “head.” And as it turns out, one use for the
notion of head is in linear ordering generalizations. So I begin the “case studies” by examining
heads and their linear order generalizations.
2.1 About Heads
What exactly, mais, is a head? As the metaphorical use of the body part term “head” indicates,
the head is the central part of a group of words. This term, like so many in linguistics, has its
controversies, and in some cases, analysts are divided about exactly what the correct head is (ou
if a single one should be recognized). Il y a, mais, some reasonably reliable heuristics for
determining what the head is (si seulement) among a group of words.1
Trask (1993, 125) defines a head as “[t]hat element of a constituent that is primarily responsible
for the syntactic character of the constituent.” This is probably not the most transparent definition,
so let me consider what it means in more depth. If this element is primarily responsible for the
syntactic character of its larger unit, it stands to reason that if a word requires other words around
it to have particular forms (ainsi, determining a part of the syntactic character of the unit), then it
is probably a head. (This criterion is called the criterion of subcategorization.) On this criterion,
verbs and prepositions (or postpositions), qui, in some languages, require other words to have
particular case forms, must be heads.
En outre, if one part of the phrase is primarily responsible for the syntactic character of the
phrase as a whole and some phrases are comprised of single words and some phrases are comprised
of multiple words, then the single words must be the heads of this kind of phrase (the criterion of
distributional equivalence). Ainsi, in phrases like swam and washed the car, the verb must be
the head, because only a single verb (like swam) can appear in the same syntactic locations as the
more elaborated phrase (par exemple. washed the car). Notice that the car, d'autre part, cannot appear
appear in the same syntactic locations as washed or swam. Ainsi, this criterion is almost paramount
to saying that the head is the only obligatory member of the phrase.
The above criteria of subcategorization and distributional equivalence give the most uniform
results. Toutefois, two more criteria have been put forth. One is the locus of morphosyntactic mark-
ing. Essentiellement, this criterion says that the word that is inherently inflected (c'est, it does not agree
with another word) for case, tendu, aspect, genre, etc.. is the head. Toutefois, this criterion must
be treated with caution because, in some instances in some languages, the morphological exponent
1This discussion here has been greatly informed by the discussion in Beavers 2003. En particulier, the criterion
names are all taken verbatim from this source.
3
of case, tendu, aspect, genre, etc.. appears in a location defined with the respect to the edge of the
phrase, not on the head (Perhaps the best known instance of this is the English possessive ’s, qui
goes at the end of an NP, not on the head noun). There is also the criterion of semantic charac-
terization. To apply this criterion, one looks at the semantics of the unit. Since washed the car
describes an event of washing and not a kind of car, the head must be washed. This last criterion
is most likely to give a result that differs from the earlier criteria.
On the whole, these criteria generally point to the same sort of words in various expressions.
Ainsi, the head of a sentence is a finite verb (John has written a book), the head of prepositional
phrase is a preposition (in the night). the head of a verbal unit is the verb (washed the car), et,
perhaps most controversially, the head of a nominal expression is a noun (the computer).2 And
it appears that roughly equivalent words behave as the heads of roughly equivalent phrases across
langues, insomuch as the same phrase types recur across languages.
2.2 Head Linear Order Generalizations
The interesting linear order generalization with regard to heads is that languages tend to consis-
tently place them at one peripheral part of the phrasal unit. Ainsi, phrases can either be character-
ized as head-initial or head-final. En outre, most languages are reasonably consistent across
the headed phrases within the language. The relatively strong consistency underlies several of the
famous word order universals proposed by Greenberg (1963).
To see the consistency, let me consider two examples. My conlang Skerre is an example of the
head-initial type, as shown in (3) (heads bolded):
(3)
Phrase Type
Clause
Adpositional Phrase
Noun Phrase
Skerre
Ehosi tsa keriyos a hantsi
PFV.eat ERG man ABS meat
te wiyet
LOC boat
yaak i arik-he
father GEN friend=1SG.POSS
Traduction
‘The man ate the meat’
‘on the boat’
‘father of my friend’
En revanche, Turkish is very consistently head-final, as shown in (4) (heads again bolded):
(4)
Phrase Type
Clause
Adpositional Phrase
Noun Phrase
Turkish
Ben sizi g¨ord ¨um.
1SG.NOM 2SG.ACC see.PRET
altı saatten sonra
six hour.ABL after
radyonun sahibi
radio.GEN owner
Traduction
‘I saw you’
‘after six hours’
‘owner of the radio’
(Examples from Thomas 1967, 42, 87, 60)
While a lot of languages are more or less reasonably consistent (Anglais, par exemple, is pretty
consistently head-initial), some languages are “mixed.” One of the more famous examples is Ger-
2Certain frameworks, most notably late Government and Binding Theory as well as the Minimalist Program,
adopt the view that the determiner is the head of nominal expressions like the computer. Beavers 2003 offers a nice
discussion of the issues and reasons to think that both the determiner and the noun exhibit some head-like properties.
4
homme (and other Germanic languages). Ici, the finite verb appears in second position in main
clauses,3 while main verb appears finally in these clauses, un péché (5):
(5)
Allemand
[hat
Er
3SG.M.NOM
have.3SG
‘He bought the book.’
[das
DEF.ACC
Buch
livre
gekauft]].
buy.PPTCP
(Graves 1990, 103)
Dans (5), hat is the finite verb. It is initial in its phrase, demarcated by the outside brackets. Toutefois,
the subject is even further to the left, making the verb second in the sentence. The main verb is
gekauft in (5). It is final in its phrase, marked by the inner brackets. Ainsi, in both cases, the head
appears on an edge of its phrase, but other parts of the sentence obscure this to a degree, et le
two kinds of verbal groupings differ in which edge is chosen.
While German shows that languages need not be precisely uniform in their placement of heads
(et, En fait, this “mixed pattern” has been quite diachronically stable, as these patterns have
existed in Germanic languages for well over 1,000 années), the fact remains that most languages are
relatively consistent in their placement. The places where head-initial/head-final splits occur are
not currently very well-understood, but it seems that a split between main and subordinate clauses
(like in German) and occasional splits between clausal and adpositional elements (see Dryer 2008
for discussion of data and some possible causes of the splits) are among the most frequent, dans
this relatively rare phenomenon.
3 Ordering of Verbal Arguments
Turning away from heads, let us look at the ordering generalizations related to the non-heads,
specifically the subset called “arguments”4. A working definition of an argument is an obligatory
dependent of some head. En particulier, I focus on arguments of verbs, because just about every
verb in every language has at least one argument, Et ainsi, the behavior of verbal arguments has
been well-studied.
3.1 Prominence Among Arguments
An important insight about verbal arguments is that, for many syntactic and semantic purposes,
they are not equal. A means of understanding their unequalities is via the notion of prominence
(see Levin and Rappaport Hovav 2005, ch. 6 for further discussion of this idea). The idea then is
that some arguments are more prominent—semantically-salient— than others and this allows them
more ‘privileges’, semantically and also syntactically.
The most prominent arguments are those that instigate the action: what roughly can be con-
sidered to be agents. The least prominent arguments are those that are altered or affected by the
action: what roughly could be called patients. An important thing to keep in mind about the notion
3German syntax adds further wrinkles in subordinate clauses and questions/imperatives. In subordinate clauses,
the finite verb is final and main verbs are penultimate. In questions and imperatives, the finite verb is strictly initial;
there is no initial subject. For the sake of brevity, I do not discuss these further wrinkles in depth.
4The linguistic term derives from the term from logic. This is due to the fact that logical formalisms were used
and continue to be used for describing semantics, and linguistic arguments are arguments—in the logical sense—of
logical predicates.
5
of prominence is that it is relative; even if a particular argument is not exactly the most prototypical
agent, it can still be said to be more prominent than another argument, such as a patient.
Agent and patient do not exhaust the possible relationships an argument can have with a pred-
icate. There are also the indirectly affected arguments in three-place predicates, like the recipient
in a giving-event. Within the linguistics literature, there is a fair amount of controversy about what
level of prominence they should be given. Recipients are sometimes treated as just being below
agents in prominence; d'autres fois, they are treated as being the least prominent, below even pa-
tients. The data that considered below will just feed this controversy: they give evidence in both
directions.
3.2 Linear Ordering Generalizations with Arguments
Unlike the ordering of heads, the linear ordering generalizations with arguments do not appear to
be quite so particular to individual languages. Plutôt, there is an overall tendency, given in (6):
(6)
More Prominent Arguments
“More prominent arguments appear to the left of less prominent arguments”
Less Prominent Arguments
This left-to-right order seems to hold regardless of where the head is placed. En outre, le
tendency in (6) plays out in different ways in different kinds of languages. Dans (most) langues
with fixed word order (along the lines of English), the order in (6) is the only order possible.
In languages with freer word order (along the lines of Russian), the order shown in (6) is the
most pragmatically unmarked order. Parce que (6) is a tendency, il y a, bien sûr, des exceptions.
Toutefois, many of the exceptions are cases where information status considerations come into play
(see section 4). Moins fréquemment, other considerations (such as those noted in section 5) also create
exceptions to (6).
Let me illustrate (6) with two brief case studies: one involving the orders found in the Germanic
famille, and the other concerning the orders found in the Taiwanese Austronesian language, Seediq.
3.2.1 Case Study 1: the Germanic Family
The Germanic family is one that has both fixed word order languages and freer word order lan-
langues, so it provides an excellent opportunity to illustrate how (6) plays out in both kinds of
langues. In this subsubsection, the examples will all be of giving-events, so there will be three
arguments to worry about: the giver or Agent NP, the thing given or Patient NP, and the entity
receiving the gift or Recipient NP.5
Within the Germanic family, the fixed word order languages require the order in (7):
(7)
Agent NP
“The Agent NP precedes the Recipient NP, qui, à son tour, precedes the Patient NP”
Recipient NP
Patient NP
Dans (8), there is an Agent NP, Jan; À
One such fixed word order Germanic language is Dutch.
recipient NP, zijn vader; and a Patient NP, het boek. Comme (8) shows, these three arguments appear in
the order expected from (7); the head of the subordinate clause—bolded—is final:
5All examples in this subsubsection are given as subordinate clauses, to illuminate the differences that would be
obscured by using main clause examples.
6
(8)
Dutch
dat
…
…
ce
'… that Jan gives his father the book’
Jan
(nom)
vader
père
boek
livre
het
DET
zijn
his
geeft.
gives
Paul Kiparsky, class handout
Fitting with the characterization of Dutch as a fixed word order language, all other orders of Jan,
zijn vader, and het boek are unacceptable.
Moving to another fixed word order Germanic language, Swedish, we can see that the place-
ment of the head does not matter for (7). In contrast to Dutch, Swedish allows for (et, En fait,
requires) verb-medial subordinate clauses. Encore, when we consider the translational equivalent of
(8) in Swedish, we get the same order of arguments as Dutch, even though the head is in a different
position (second in the clause—bolded). This is shown in (9):
(9)
Swedish
att
…
ce
…
'… that Jan gives his father the book ’
Jan
(nom)
loin
père
ger
gives
boken.
book-DET
sin
his
Paul Kiparsky, class handout
Comme (9) shows, Swedish has the order that (7) leads us to expect. All other orders of Jan, sin far,
and boken are unacceptable in Swedish.
Turning to a freer word order Germanic language, we consider German. Befitting German’s
status as a freer word order language, the order Agent NP
Patient NP from (7) est
not required. But this order is the least pragmatically marked order in German. This means that the
order Agent NP
Patient NP can appear in the largest number of contexts and this
arrangement of arguments has the fewest “extra” meanings; examples of the “extra” meanings I
mean here would include the sense of “emphasis” given to a particular argument in English when it
is intonationally stressed (JAN gave his father the book) or occurs after the verb BE in constructions
like It is JAN that gave his father the book.
Recipient NP
Recipient NP
The translational equivalent of (8) et (9) also exists in German; it is given in (dix) (note
that German patterns with Dutch and has the phrase-final head—bolded in (dix)—in subordinate
clauses):
(dix)
Allemand
seinem
dass
…
…
his.DAT
ce
'… that Jan gives his father the book’
Jan
(nom)
Père
père
das
DET.NEUT.ACC
Buch
livre
gibt.
gives
Paul Kiparsky, class handout
Toutefois, in contrast to Dutch and Swedish, all other orders of Jan, seinem Vater, and das Buch
are possible. But only the order shown in (dix) is the contextually unmarked “neutral” word order:
neither Jan, seinem Vater, nor das Buch is being contrasted with other words and this order can
appear in the largest number of contexts.
Ainsi, this brief Germanic case study seems to suggest two things about linear order in natural
langues. D'abord, the order of arguments is not particularly dependent on where the head verb
est. Deuxième, the required orders in fixed word order languages appear to mirror the pragmatically
“neutral” orders of freer word order languages.
7
3.2.2 Case Study 2: Seediq
The Germanic data are interesting, but taken by itself, it could just be an interesting fact about
that family. Ainsi, to show that some of the general principles discussed above in section 3.2.1 apply
elsewhere in the world’s languages, I move to a discussion of Seediq, an Austronesian language
spoken in Taiwan. This is not to suggest that the syntax of Seediq is particularly close to that of the
Germanic languages; quite a few of the details are significantly different. En fait, the differences
here will also allow us to see an example of one general type of exception that I mentioned earlier:
exceptions to (6) often are due to “conflicting” information structure considerations.
Seediq is a typical western Austronesian language. Its heads are rigidly initial in their phrases.
En plus, all of its verbs have different forms that indicate that a particular argument NP has a
special syntactic status. The particular NP is called the trigger6 and the verb forms are known as
foci in the literature on Seediq.7 In Seediq, the trigger is marked by the prenominal word ka. Le
ka-phrase is required to be clause-final, regardless what focus form the verb is in. This is shown in
(11); le (À) example has the Agent Focus (AF) form while the (b) example has the Patient Focus
(PF) formulaire:
(11)
Seediq
À.
huling
Qmita
see.AF
chien
‘Pawan sees a dog’
le
LE
Pawan.
(nom)
b. Wada=mu
PRET.AUX=1SG.GEN
‘Pawan was seen by me’
qtaun
see.PF
le
LE
Pawan.
(nom)
(Holmer 1996, 58)
(Holmer 1996, 58)
Aside from the ka phrase, Seediq NPs follow the generalization from (6) that more prominent
arguments appear to the left of less prominent arguments. Ainsi, in Seediq, the prominence-based
order holds in just a particular region of the clause: between the verb and the trigger (et, nec-
essarily, excludes the trigger). A further quirk (mentioned earlier) is that the recipient and patient
appear in Seediq in the reverse order from the Germanic family. Donc, au lieu de (7), Seediq
has the order in (12):
(12)
(Patient NP)
(Agent NP)
“The Agent NP precedes the Patient NP, which precedes the Recipient NP,
which precedes, finally, the Trigger.
The only requirement is that there must be a Trigger at the end of the sen-
tence; all other NPs are, en principe, optional”
(Recipient NP)
Trigger NP
Seeing the full pattern in (12) is a bit tricky, because verbs in Seediq that would have the requisite
four arguments—an Agent, a Patient, a Recipient, and a Trigger—are either rare or non-existent.
But we can see the pattern composited from two examples. The first, given in (13), has the Agent
NP (the giver, ka Awi) as the trigger. The other two NPs are in order Patient NP (italicized)
Recipient NP (underlined):
6So called because it appears, Au moins, to trigger a certain verb form.
7Other Austronesian languages have the same phenomenon, but in the discussions of them, the term ‘voice’ is used
plutôt.
8
(13)
Seediq
Wada
PRET.AUX
‘Awi gave Pawan a house.’
mege
give.AF
sapah
maison
Pawan
(nom)
le
LE
Awi.
(nom)
(Holmer 1996, 79)
In the second (14), the Recipient NP (the gift-getter) is the trigger. The other two NPs are in the
Patient NP (both parts underlined):8
order Agent NP (italicized)
(14)
Seediq
Bniqan=mu
give.PRET.LF=1SG.GEN
‘I gave my clothes to him’
lukus
vêtements
dans
1SG.GEN
heya
3SG.NOM
(Holmer 1996, 79)
These patterns are not unique to Seediq; the same sort of ordering facts (including the placement of
the trigger) are found in other rigid word order Austronesian languages, such as Malagasy (Pearson
2005).
The above brief exploration of the grammar of Seediq reveals two further things. D'abord, le
patterns found in the Germanic family do not seem to be isolated to the Germanic family; le
general tendency found in (6) is also found in Seediq. Deuxième, cependant, the grammar of Seediq
has particular complications not found in the Germanic family: the interaction between the verb
form and the trigger, and the rigidity in where the ka-phrase occurs.9
4
Information Status Ordering
Jusqu’à présent, I have discussed the linear ordering tendencies found with respect to heads and with argu-
ments. But another common basis of linear order generalizations in natural languages is informa-
tion status. What is information status, you might wonder? It is a property of phrases, having to
do with their informational position in the current conversation, discours, or written text. This boils
down to whether the phrase is new information, old information, re-introduced old information,
etc.. The study of the information status is fraught with controversies over terms and the proper
classification of the requisite information status categories. For the present purposes, I will try to
avoid the bulk of the controversies (see Chafe 1976; Prince 1981; Gundel 1988; Lambrecht 1994;
Engdahl and Vallduv´ı 1996, entre autres, for more discussion) and deal with the simple facts
surrounding only two information statuses. I will define them as in (15):
(15)
Rubrique: What has been under discussion, discourse old (aka “theme”)
Focus: New information, often contrastive (aka “rheme,” “comment”)
The linearization generalization, which appears to be universal,10 is given in (16):
8Ideally, l'exemple dans (14) would have the agent and recipient as non-pronominals, parce que (À) pronominals
independently have a tendency to appear early in clauses and (b) in Seediq, the so-called “genitive” pronouns must
appear after the first item in the clause. But Holmer 1996 does not give examples that would avoid these potential
confounds.
9In the interest of full disclosure, the ka-phrase ordering phenomenon in Seediq does bear some resemblance to the
topic-first phenomenon found in at least some Germanic languages.
10There is another commonly discussed information status often called antitopic, which usually appears at the end
of sentences. An example would be He is a good linguist, that guy, where that guy is the antitopic.
9
(16)
Focus
Rubrique
“Phrases with a topic information status precede those with the focus information status”
This pattern presents itself in several languages. It seems especially common in languages where
verb-final order is either the most frequent or is required in declarative main clauses. I will use the
Basque language of the Pyrenees to illustrate.
In Basque, it is most usual to end a clause with a main verb followed by an auxiliary. An
example of this is given in (17):
(17)
Basque
Hemen
ici
bizi
en direct
Verb
et.
AUX.3SG
Auxiliary
‘She lives here.’
(King and Olaizola Elordi 1996, 18)
Moving backwards in the clause, the focus information status element (including all content ques-
tion words like the translational equivalents of who, what, où, quand, etc.) must immediately
precede the verb. An example showing both a question word and a non-question NP focus is given
dans (18):
(18)
Basque
Nor
zen
who
AUX
‘Who came? John came.’
– Jon
John
etorri
viens
etorri
viens
zen.
AUX
(King and Olaizola Elordi 1996, 204)
In the first clause of (18), nor ‘who’ is the focus element, since the question word has to be the
new element or it won’t be a sensical question. So nor has to come before the main verb, etorri. Dans
the second clause of (18), Jon is the new information, since it is the answer to the question. Ainsi,
Jon must come before the main verb in this clause.
Continuing our Benjamin Button-esque trip through the Basque clause, the last generalization
is precisely the one given already in (16): the topic element must precede the focus. An example
of this sort of ordering in Basque is given in (19):
(19)
Basque
zidan.
Ni-ri,
I-DAT
AUX
‘JON explained that to me.’
Jonek
Jon.ERG
azaldu
expliquer
(Hualde and Ortiz de Urbina 2003, 460)
Dans (19), niri ‘to me’ appears in an initial, intonationally-separate phrase, a common place for topics.
The sense of (19) is probably more like “As for me, Jon explained that to me;" c'est, returning to
something of relevance to the speaker, it was Jon that explained that particular thing to that speaker.
Patterns very similar to the ones found in Basque are also found in the rigidly verb-final Turkish
(Hoffman 1998), the generally verb-medial Hungarian (Kiss 1995), and the reasonably free word
order Australian Aboriginal language Warlpiri (Legate 2002).
The generalization in (16) also manifests itself in verb-initial languages. Tzotzil, a Mayan
language spoken in Mexico exemplifies this. In Tzotzil clauses, as in many Mayan languages,
the arguments usually follow the initial verb. Toutefois, in some specialized constructions, Tzotzil
allows both topic elements and focus elements to appear before the verb. When both a topic and
dix
a focus appear before the verb, they appear in that order, in line with (16). An example of this is
given in (20). Note that all Tzotzil initial topics require the pre-phrase word a. The context (dans
English translation) is given in (20a) so the relative oldness and newness of the topic and focus of
(20b) can clearly be seen:
(20)
Tzotzil
À.
b.
Context:
Once there was an orphan. The orphan suffered greatly. Whatever the master’s
children ate, they ate first. They drank first.
de
DET
UN
ch’ak’bat.
was.given
TOP
‘It was leftovers that the poor girl was given’
sovra
leftovers
tzeb-e
girl-ENC
prouver
pauvre
(Aissen 1992, 51)
As the context shows, the orphan is introduced early in this (court) discourse, so the coreferential
phrase ti prove tzeb ‘the poor girl’ clearly qualifies as a topic. Sovra ‘leftovers’ is clearly new to the
discourse in (20b), so it qualifies as a focus. Given these information statuses, the order in (20b) est
precisely expected from (16).
Ainsi, the Tzotzil data reveals that the generalization in (16) seems to hold when topics and
foci are clearly not in their canonical positions (as you may recall, normally, all arguments follow
the verb in Tzotzil), in addition to when the topics and foci seem like they are in more “normal”
positions, as in the Basque examples above.
The Tzotzil data also shows the information status-based ordering can be found in all kinds of
langues. En outre, contrasting the patterns in Tzotzil with Basque brings out an important
point about the nature of information status-based ordering. In a language such as Basque with
focus order
fairly rich morphology (especially either case or agreement morphology), the topic
appears “unadorned”, that is without any additional specialized collection of words. When special
information statuses need to be highlighted in more analytical languages, there is an additional
collection of words. Besides Tzotzil, English offers an example, aussi, in the form of cleft sentences.
In a cleft sentence like It is linear ordering that we’re talking about, the “extra” it is is needed to
contrast the newer information linear ordering with other possible bits of information.
5 Ordering by “Weight”
The final linear ordering phenomenon I want to address in this paper is what I will call ordering by
“weight”. The notion of “weight” deals with the amount of linguistic material in a given phrase.
The key unit of “weight” is “heavy”: heavy phrases either consist of a large number of words or
involve a complex structure, such as several relative clauses or other subordinate clauses. For a
great many of the cases, it is difficult to decide if the large number of words or the structural com-
plexity is the defining quality of these phrases, because the two are highly correlated. (Deciding
on this matter will not end up being relevant for our purposes.) In addition to heavy phrases, quelques
linguists have also explored generalizations about light phrases, which are usually defined as sin-
gle word phrases, perhaps phonologically dependent on another phrase. Toutefois, it is not so clear
what is going on with light phrases and, aside from noting that they often occur near heads, I will
not discuss them further in this paper.
11
The most well-established ordering by weight generalization is that heavy phrases appear at
Heavy, where X is any non-heavy phrase.11 This occurs in English.
the end of clauses; that is X
Comme (21) shows, a complex NP like some friends that John had brought to the party is much better
at the end of the sentence, comme dans le (À) exemple, as opposed to in the usual object “slot” between
the verb and the prepositional phrase (shown in the (b) exemple):
(21)
Anglais
À.
b.
I introduced to Mary some friends that John had brought to the party.
?I introduced some friends that John had brought to the party to Mary.
(Hawkins 1990, 228)
Toutefois, this linear order generalization is not specific to English or the Germanic family. Il
has instantiations around the world, in languages with otherwise very different word order patterns.
One such language is the Boumaa dialect of Fijian. In this language, the canonical order has the
verb first and the doer of the action at the end; ainsi, Boumaa Fijian has Verb–Object–Subject
(VOS) ordre des mots. This is shown in (22):
(22)
Boumaa Fijian
[À
rai-ca
E
3SG
see-TR
ART
‘The old person saw the child.’
gone]
enfant
[À
ART
qase].
old.person
(Dixon 1988, 243)
Yet if the verb takes a complex argument, such as a complement clause, that argument must appear
at the end of the clause. This is exactly what happens in (23), where the complement clause ni o
ira sana mai ’abati Boumaa appears at the end of a clause:
(23)
Boumaa Fijian
V-S-Heavy
tu’u-na
E
sa-na
INCP-FUT
tell-TR
3SG
‘Tui Waini’eli said that they would come and invade Boumaa.’
Tui Waini’eli]
(title)
[Ni
COMP
Peut
ici
o
ART
[o
ART
ira
3PL
Peut
viens
’aba-ti
invade-TR
(Dixon 1988, 243)
Boumaa].
(lieu)
This phenomenon also occurs in Basque, providing an interesting twist to its word order. Comme
mentioned in section 4, the main verb and auxiliary usually end a Basque clause (I call this the
verbal complex in the examples below). The other arguments follow the Germanic order of (7), comme
shown in (24):
(24)
Basque
Ordonnance: Agent–Recipient–Patient–Verbal Complex
[ekarri
[Ene
apporter
mon
‘My father brought mother a red skirt.’
[amari]
mother.DAT
aitak]
father.ERG
gorria]
red.DET
[gona
skirt
dio].
AUX
(Hualde and Ortiz de Urbina 2003, 448)
Yet when a Basque verb takes a complement clause, the complement clause appears at the end of
the sentence. This is shown in (25), with the complement clause Mikelek erlojua galdu duela:
11This phenomenon sometimes is called Heavy-NP Shift, a term that dates from early transformational grammar,
where it was proposed that these phrases underwent a transformation that moved them to the end of the clause.
12
(25)
Basque
Ordonnance: Agent–Verbal Complex–Heavy Phrase
[Mikelek
[Jonek]
(nom).ERG
(nom).ERG
‘Jon said that Mikel lost the watch.’
erlojua
montre
[esan
dire
du]
AUX
galdu
perdre
duela].
AUX.COMP
(Hualde and Ortiz de Urbina 2003, 452)
Basque, alors, is an interesting case. In English and Boumaa Fijian, when a heavy phrase appears at
the end of a clause, it is still appearing where other arguments appear, even if it is not the ordinary
spot for that kind of argument. Toutefois, in Basque, the heavy phrase appears at the end of the
clause, which is not where other arguments appear.
The apparent switch from verb-final order to verb-medial order with a heavy phrase raises the
question of whether the Basque pattern is always how verb-final languages work and whether the
opposite of X
Heavy ever occurs. In Japanese, some ostensibly heavy phrases are claimed to
appear clause-initially, un péché (26):
(26)
Japanese
À.
b.
À]
ce
kekkonsita
married
John-ga
(nom)-NOM
[Kinoo
yesterday
‘Mary said that John got married yesterday.’
?[Mary-ga]
kekkonsita
(nom)-NOM
married
‘Mary said that John got married yesterday.’
John-ga
(nom)-NOM
[kinoo
yesterday
[Mary-ga]
(nom)-NOM
[itta].
dit
À]
ce
[itta].
dit
(Hawkins 1990, 231)
But there is some controversy over whether this is the same sort of “heavy shift” that occurs in
Anglais, Boumaa Fijian, and Basque or an independent information status-related phenomenon.
Depending on what is ultimately decided about these Japanese examples (see Yamashita and Chang
2001 for an interesting discussion), the generalization about having heavy phrases at the end might
not be a strict universal. Néanmoins, it appears that “heavy at the end” is an extremely common
phenomenon.
6 Concluding Thoughts
In this paper, I have discussed four different areas of notable linear ordering generalizations in
langues naturelles: ordering of heads, ordering of arguments, ordering of information statuses, et
ordering by weight. Not all of these proved to have universal patterns, but I hope to have illus-
trated that the existing variation in natural languages appears to be relatively small. Heads seem
to always be at the edge of their phrase. They most often are at the same edge throughout a given
langue, although some languages allow different head locations in different kinds of phrases.
The arguments, d'autre part, typically are ordered in the same way—in order of decreasing
prominence—regardless of the location of the head. Et, interestingly, the most pragmatically un-
marked orders in freer word order languages and the required orders in rigid word order languages
appear to be the same (except for the issue regarding the order of recipients and patients). Le
information status order also exhibits a kind of prominence asymmetry, with the older, peut-être
more accessible, information appearing before the newer information. Enfin, heavy multiword
phrases have a preponderance for appearing at the end of clauses.
13
This relatively narrow window of variation in natural languages, donc, leaves a lot open
for some interesting (if not completely naturalistic) constructed languages. A language that always
puts its heads in the second position of the clause? A language where focus must precede topic?
A language where all the light elements go to the edges, leaving the heavy stuff in the middle? Tous
these things—and more—await possible exploration!
Encore, even if your conlanging style keeps you pretty close to natural languages, the topics cov-
ered here still leave many things to decide: where will your language(s) put its heads, will your
langue(s) have special topic and focus constructions, which order of patients and recipients will
your language(s) employer? And as the many footnotes and hedges indicate, this paper also has
just provided an introduction to the linear order phenomena in natural languages: there’s more to
uncover and once the facts have been sorted out, more details to decide about and to implement.
The linguistics literature is one place to look for more, but I would not count out learning much
about languages from primary (and near primary) sources: grammars, teaching materials, et, de
course, speakers themselves. Ainsi, I hope this work has done its part in intriguing you about linear
order generalizations and provides motivation for increasing the description of your conlang’s (ou
conlangs’) syntax.
Les références
Aissen, Judith. 1992. “Topic and Focus in Mayan.” Language 68:43–80.
Beavers, John. 2003. “More Heads and Less Categories: A New Look at Noun Phrase Struc-
ture.” In Proceedings of the 10th International Conference on Head-Driven Phrase Structure
Grammaire, edited by Stefan M¨uller, 47–67. Stanford, Calif.: Publications du CSLI.
Bresnan, Jeanne. 2001. Lexical-Functional Syntax. Oxford: Éditeurs Blackwell.
Chafe, Wallace L. 1976. “Givenness, contrastiveness, sujet, sujet, and point of view.” In
Subject and Topic, edited by Charles N. Li, 25–55. New York: Academic Press.
Dixon, R. M. O. 1988. A Grammar of Boumaa Fijian. Chicago: Presse de l'Université de Chicago.
Dowty, David. 1995. “Toward a Minimalist Theory of Syntactic Structure.” In Discontinuous
Constituency, edited by Harry Bunt and Arthur van Horck, 11–62. Berlin and New York:
Mouton de Gruyter. (based on 1990 Tilburg Conference on Syntactic Discontinuity).
Dryer, Matthew S. 2008. “Relationship between the Order of Object and Verb and the Order of
Adposition and Noun Phrase.” Chapter 95 of The World Atlas of Language Structures Online,
edited by Martin Haspelmath, Matthew S. Dryer, David Gil, and Bernard Comrie. Munich:
Max Planck Digital Library. Available online at http://wals.info/feature/95.
Accessed on 2009-06-17.
Engdahl, Elisabet, and Enric Vallduv´ı. 1996. “Information Packaging in HPSG.” Edited by
Claire Grover and Enric Vallduv´ı, Edinburgh Working Papers in Cognitive Science, Volume
12: Studies in HPSG. University of Edinburgh, 1–32.
Gazdar, Gerald, and Geoffrey K. Pullum. 1981. “Subcategorization, Constituent Order, et le
Notion “Head”.” In The Scope of Lexical Rules, edited by M. Moortgat, H. van der Hulst, et
J. Hoekstra, 107–123. Dordrecht: Foris.
Graves, Paul G. 1990. German Grammar. Hauppage, New York: Barron’s Educational Series.
14
Greenberg, Joseph H.. 1963. “Some Universals of Grammar with Particular Reference to the
Order of Meaningful Elements.” In Universals of Human Language, edited by Joseph H.
Greenberg, 73–113. Cambridge, MA: AVEC Presse.
Gundel, Jeanette K. 1988. “Universals of topic-comment structure.” In Studies in Syntactic
Typology, edited by Michael Hammond, Edith A. Moravcsik, and Jessica Wirth, 209–239.
Amsterdam: John Benjamins.
Hawkins, John A. 1990. “A Parsing Theory of Word Order Universals.” Linguistic Inquiry
21:223–261.
Hoffman, Beryl. 1998. “Word Order, Information Structure, and Centering in Turkish.” In
Centering Theory in Discourse, edited by Marilyn A. Marcheur, Aravind K. Joshi, and Ellen F.
Prince, 251–272. Oxford: Presse Clarendon.
Holmer, Arthur J. 1996. A Parametric Grammar of Seediq. Lund, Sweden: Lund University
Presse.
Hualde, Jose Ignacio, and Jon Ortiz de Urbina, eds. 2003. A Grammar of Basque. Berlin: Mouton
de Gruyter.
King, Alan R., and Begotxu Olaizola Elordi. 1996. Colloquial Basque: A Complete Language
Course. London and New York: Routledge.
Kiss, Katalin ´E. 1995. “Discourse configurational languages: introduction.” In Discourse Con-
figurational Languages, edited by Katalin ´E. Kiss, 3–27. Oxford: Presse universitaire d'Oxford.
Lambrecht, Knud. 1994. Information Structure and Sentence Form: Rubrique, Focus and the Mental
Representation of Discourse Referents. Cambridge, ROYAUME-UNI: la presse de l'Universite de Cambridge.
Legate, Julie. 2002. “Microparametric Non-Configurationality: The Case of Warlpiri.” Invited
parler, Department of Linguistics Colloquia Series, University of Wisconsin, Madison.
Levin, Beth, and Malka Rappaport Hovav. 2005. Argument Realization. Cambridge, ROYAUME-UNI: Cam-
bridge University Press.
Pearson, Matthew. 2005. “The Malagasy Subject/Topic as an A0-Element.” Natural Language
and Linguistic Theory 23:381–457.
Prince, Ellen F. 1981. “Towards a taxonomy of given-new information.” In Radical Pragmatics,
edited by Peter Cole, 223–256. New York: Academic Press.
Rood, David. 1976. Wichita Grammar. New York: Garland Publishin.
Sag, Ivan A., Thomas Wasow, and Emily M. Bender. 2003. Syntactic Theory: A Formal Intro-
duction. 2e édition. Stanford, Calif.: Publications du CSLI.
Thomas, Lewis V. 1967. Elementary Turkish. New York: Publications de Douvres. Revised and
Edited by Norman Itzkowitz.
Trask, R.L. 1993. A Dictionary of Grammatical Terms in Linguistics. London and New York:
Routledge.
Yamashita, Hiroko, and Franklin Chang. 2001. “‘Long Before Short’ Preference in the Production
of a Head-Final Language.” Cognition 81:B45–B55.
15