Getting Things Out Of Order 
(Au LFG-Proposal for the Treatment of German Word Order) 
Klaus Netter 
Department of Linguistics 
University of Stuttgart 
West Germany 
0. INTRODUCTION 1) 
One of the most characteristic features of German word order 
seems to be a contrast between fixed ordering rules concerning 
the order of verbal elements and a much more variable ordering 
of their corresponding nominal arguments. As a consequence 
German word order seems to yield a large number of phenomena that 
may be classified as 'unbounded' or 'long-distance dependencies', 
without necessarily involving wh-constituents or 'movement' 
across sentence boundaries. Whereas in traditional LFG long dis- 
tance dependencies are treated by means of constituent control, 
we will follow a recent proposal by KAPLAN and ZAENEN (1986) 
to give up the constraint known as 'functional locality' and 
instead allow regular expressions to appear as functional schemata 
annotated to c-structure rules. Exploiting the principles of 
completeness and coherence we will thus be able to cope even with 
absolutely free word order without the need of generating empty 
terminal nodes at all. The empirical assumption, underlying the 
proposed analysis in its most radical form, is the hypothesis that 
(with very few exceptions) the nominal arguments have to appear to 
the left of the verb by which they are assigned case. We will 
restrict the discussion to sentences with one finite verb as well 
as to subcategorized nominal arguments, largely ignoring 
ADJuncts. 2) 
In chap. 1. we will introduce the mechanism of 'functional 
incertainty', applied to the example of extraposed infinitival 
constructions. Chap. 2. will deal with the sentence initial posit- 
ion(s) in verb-second clauses and chap. 3. will discuss some 
approaches to handle free word order in the topological field 
between COMP and the clause-final verb complex. 
1. EXTRAPOSED INFINITIVAL CONSTRUCTIONS 
To demonstrate the mechanism of 'functional iucertainty' as 
well as some basic assumption in the proposed fragment of 
German grammar we will start with a relatively simple problem 
involving the treatment of embedded and extraposed infinitival 
constructions. Consider the following data: 3) 
(1) dass \[der Mann \[\[ das Such gelesen \] haben\] wird\]. 
that the man the book read have wit l 
(2) dass \[der Mann \[\[das Such zu lesen\] versucht\] hat\]. 
that the man the book to read tried has 
(3) dass \[der Mann \[\[das Such gelesen\] zu haben\] scheint\]. 
that the man the book read to have seems 
(4) dass \[der Mann \[versucht\] hat\], \[das Buch zu lesen\]. 
that the man tried has , the book to read 
(5) *dass\[der Mann scheint\],\[\[ das Such gelesen\] zu haben\]. 
that the man seems , the book read ta have 
(6) *dass \[der Mann wird\], \[\[das Such gelesen\] haben\]. 
that the man witt, the book read have 
(7) dass \[der Mann \[ versucht\] hat\], \[zu behaupten\], 
\[\[das Such gelesen\] zu haben\] 
that the man tried has, to ctairn, the book read to have 
Sentence (1) contains a main verb and two auxiliary verbs, (2), 
(4) and (6) contain equi verbs (versuchen, behaupten), 'scheinen' 
in (3) and (5) represents a raising verb. (2) is the same struct- 
ure as (4) with the infinitival clause extraposed in the latter. 
The contrast between (2)/(4) and (3)/(5) or (1)/(6) illustrates 
two points: on the one hand extrapositiou is permitted iff the 
governing verb governs the 'zu' infinitive and is not a raising 
verb; on the other hand even with these conditions met 
extraposition is not obligatory. Informally extraposition can thus 
be characterised as an optional right-adjunction of an embedded 
constituent to a self-embedding structure. Sentence (7) is to 
show that this rule can be applied recursively. 
494 
In the treatment of equi- and raising-verbs, including (in out" 
approach) auxiliary and any kind of modal verbs, we will 
follow the traditional LFG approach; these verbs are sub- 
categorized for the 'open' grammatical function XCOMP and are cor- 
respondingly assigned a functional control equation that identifies 
the verb's own subject or object with the subject of this 
XCOMP. Raising verbs are distinguished from equi verbs by denoting 
that the 'raised' function will not be interpreted in the raising 
verb's semantic form. Beyond this it will have to be specified for 
each equi or raising verb what type of infinitive the verbal head 
of the governed XCOMP has to be (The feature referring to 
extraposition in (L3) will be attached only to the appropriate 
verbs). 4) 
it, l) wird: V, (t PRED) = "wel'den <(XCOMP)> (SUBJ) 'I 
((I TENSE) = future 
SUBJ CASE) = n~l 
(t SUB J)= (t XCOMP SUB J) 
(t XCOMP INF) =c en 
(L2) haben: V, (~ PRED) = "haben <(XCOMP)> (SUB J)" 
((I I~IF) =en 
SUB J) = (~ XCOMP SUB J) 
(t XCOMP INF) =e ge 
(L3) versucht: V, (C PRED) = "versuchen <(SUBJ) (XCOMP)>" 
(t INF) = ge 
(~ SUBJ)= (~ XCONP SUB J) 
XCOMP INF) =C zu 
(t XCOMP EXTRAPOSII'ION) = +. 
Given a lexicon of the above format we could use the following 
rules to generate the embedded infinitival constructions (1)-(3): 
tRI) S' "> COMP ~ = 4 
s 4. 
(R2) S -> NP (~ SUB J) = 
vo t = t- 
(NP (t OBJ) = ~\] 
\[vP (t xcoMP) = 4\] 
v t:t- 
To cover the extraposition cases we can make efficient use of the 
notion of 'functional incertainty'. What we want to express is 
that the extraposed infinitival construction appears as the 
right-most daughter of the governing S-node, that it fulfills the 
function of an XCOMP and that it should be embedded in the 
resulting f-structure at an unpredictably deep level which is 
determined by the number of XCOMPs that have been generated by the 
self-embedding VP in (R3). To do this we will expand the rule (R2) 
by introducing a special VP' category that is annotated by a 
functional schema containing a regular expression involving the 
Kleene star on the grammatical function XCOMP (The constraining 
equation attached to the VP' nodes serves to make sure that only 
those XCOMPs will be dominated by this node that are governed by an 
appropriate verb satisfying this constraint through the correspond- 
ing defining equation in its lexical entry; see above (L3).): 
(R2,1) S -> NO (~ SUBJ) = 
vP t : t 
(~ EXTRAPOSITION) =c +\]. 
(R4) VP' -> VP ~ = 4 
\[VP' (t XCOMP* XCORp) = 
(t EXTRAPOSITION) =c +\]. 
These rules will produce the following c- and f-structures with the 
metavariables appropriately instantiated. Although the VP'-node 
appears in the c-structure as an immediate daughter of S 
it's corresponding f-structure (f4) is pushed down to the 
deepest level of embedding in tire global f-structure. 
(Sl) S fl 
(fl SUgJ) = f2 
NP f2 
der Mann 
fl = f3 
VP f3 /\ 
(f3 XO~MP) = f5 f3 : f6 
VP f5 V f6 
f5 :: f8 
V f8 
I 
versucht hat 
(fl XCOMp XCONP) = f4 
VP' f4 L 
f4 = f7 
VP f7 
(f7 OBJ) = f9 f7 = flO 
NP f9 V flO I I 
das Buch zu Lesen 
(F1) -- -, .. ._ 
fl/ XCOMp = f5/ XCOMP = f4/ \[ PRED : \[eson( \[SUBJ\], \[OBJ\] ) 
f3/ f8 iT/ | zu j 
f6/ f10 \[ INF = 
I I I EXTRAPOSITI°N = + 
LsunJ = f2 ~_ ~ J 
PRED versuchen( \[SUBJ\], \[XCOMP\] ) /) 
INF = ge 
.}u,J : f2 ~_~-- 
PRED = haben( \[XCOMP\] ) "~ 
susJ f2 I-pRE~ ..... "\] 
\[._CASE = nora J 
That the numbt~'r of iterations implied by the Kleene-star operator 
on the XCOMP in (R2.1) and (R4) will always be the correct 
one is guaranteed by principles that are central to traditional 
\],FG: The principle of coherence requiring that any goveruable 
function must he governed by a local predicate in its f-structure 
will make sure that the number of iterations will not exceed 
the number of XCOMPs generated by the left sister VP 
structure plus one; the requirement for a miuimal solution and 
the restriction that any predicate introduced by a lexical entry 
must not be trollied with any other predicate entry will make sure 
that the number of iterations will not be smaller than the 
appropriate number of XCOMPs. 
2. VERB-SECOND 
Until now we have considered only clauses with the finite verb in 
a final position, and a strict order of nontinal arguments, In 
the following we shall ntodify our rules to cover also verb 
second clauses, restricting ourselves to declaratives. 5) 
The rule determining the position of the finite verb in German 
can be informally expressed by the following generalization: 
The finite verb appears in a clause final position iff the first 
position of the sentence is taken by a conjunction, or by a 
relative or (subordinative) interrogative prononn. If these con- 
ditions do not apply we ",viii assume that the finite verb takes the 
position that is dominated by COMP in (Rl). In this case the 
position in front of the finite verb, which we will assume to be 
(laminated by S" , may be filled by any major constituent 
(including VP). As this position is by no nteans lestrieted to be a 
SUBJ~posltion, we will introduce a variable G ranging ovtu the 
full set of governable functions (including XCOMP). 6) 
Furthermore wt, ~ have to take into consideration, that only the 
surface realization S\[JBJ will always appear on the top-level 
of the global f-structure, whereas any other function may belong 
to a deeper level of f-structnre embedding. So we will again make 
use of an annotation involving the Kleene--star operator, yielding 
('1' XCOMP* G) = "~. Thus we will be able to topicalize any 
governable function xto matter, how deep it be embedded in the VP or 
VP' structures. The principle of consistency will again guarantee 
that no governable function will appear in this position that is 
not subcategorized for by a verbal predicate that can be reached 
along an f-structure path consisting of XCOMPs only. Restricting 
the f-structure path to XCOMPs will make sure that no 
governable function be 'extracted' from inside a complex NP or some 
node annotated as ADJUNCT. 7) 
Another important modification concerns our S-rule: In order to 
avoid constituent control as well as a VP rule expanding to an 
ol)tional V constituent, we 'flatten' the S rule, making all 
its immediately dominated constituents optional. The principle of 
consistency will then make sure that one and only one position for 
the verb will be selected. The existential constraint requiring 
tense will hold for both the final and the front position. 
(Re) S'' -> \[XP (t XCOMP* G) = 
(~ MOOD) = doctarative\] 
s, t:t- 
(R1) S' -> CO/4P t = 
~(; TENSE) 
s % 
(R2.2) s -> \[NP (1 SUB J) = 
\[up (1 oBa2) = ~1 
tNP (10BJ) = /(\] 
\[vP (1 xcoMp) = t\] 
tv t = tl \[vp, (t xcoMp* xcoMp) = i' 
,,.\], 
(RS) COMP -> V t = 
(R6) COMP -> CONJ t ° 
(I MOOD) = subordinate, 
Given this rule system it will be no problem to generate any of 
the following declarative sentences: 
(8) Der Mann hat \[der Frau das Buch gegeben/. 
the alan lies tile Woalan the book given 
(9) Der t,'rau hat der Mann dos Buch gegeben. 
to the uoman has ttle man the book given 
(i0) Das Buch hat dec Mann der l,'rau gegeben. 
tile book has the man to the ~oman given 
(l I) Der Frau hat er w, rsprochen, das Buch zu lesen. 
to the woman has he pr~rlfised the book to read 
(12) Dos Buch hat er ihr versprochen zu lesen 
the book has he her promised to read 
(13) Dos Buch /tat er ihr zu lesen versprochen, 
the book has lie her to read promised 
(14) Dos Buch zu lesen hat er ihr versprochen. 
the book to read has he her promised 
(/5) Versueht hat er, das Buch zu lesen. 
tried has he tile book to read 
3. GERMAN AS A FREE WORD ORDER LANGUAGE 
fit the htst section we assumed that it will be only the sentence 
iuitial positions that can be filled by any major constituents. 
However there is little reason to assmne that the linear order of 
nominal arguments will always follow the pattern produced by 
the original rules (R2) and (R3): 
(16) dass dem Erfinder diese Entdeckung gehmgen ist. 
to the inventor thi!; discovery succeeded has 
(17) dass dem Mann jemand ein Buclt geschenkt hat. 
to the mnn somebody a book given has 
(18) d.ss ibm dos linch jemand geschenkt hat. 
hirn tile book soli\]ebody given has 
(19) dasses ihm jemand geschelt\]¢l h(zl. 
it hba sc~nobody given has 
(20) dass siclt der Mann zu bewegen begann, 
himself tile man to iilove started 
(21) dass ihn eine Studentin zu kuessen versuchte. 
him a student (femote) to kiss tried 
(22) dass einc Studentin ihn versucht hat zu kuessen. 
a student hilll tried has to kiss 
(23) dass ihn eine Studentin versucht hat zu kuessen, 
hinl a student tried has to kiss 
(24) dass es ihm jemand xu &,sen versproehen hat. 
it to him somebody to read promised has 
(25) dass sic" ihm dos Buch \[ hat \[\[geben\] wollen\]\]. 
she IHm the book has given uanted 
495 
(16) is representative for a class of verbs which require an un- 
marked linear order OBJ2 < SUBJ. (17)-(19) show some per- 
mutations of SUBJ, OBJ and OBJ2 which seem to obey some 
linear precedence rules relating to a definiteness-scale for noun 
phrases. In (20) and (21) an argument depending on a verb 
governed by an equi verb precedes the subject of this equi 
verb. In the extraposition variants (22) and (23) of the embedded 
structure (21) the verbs are extraposed while leaving behind their 
arguments. In (24) the linear order of nominal arguments has been 
'reversed' to yield a structure comparable to some Dutch 
infinitival constructions. In (25) the finite verb separates the 
all objects from heads appearing in a left-branching verb complex. 
All of these sentences have in common that tim arguments of some 
embedded verb(s) are separated from their head(s) by some 
constituent belonging to a higher level of f-structure embedding, 
i.e. they are obviously not derivable by a simple self-embedding c- 
structure rule. Basically three different approaches to handle 
these cases come to mind: 
a) One could stick to rule (R2.2) and modify it by introducing 
several NP-positions as left sisters of the SUBJ-NP annotated with 
a regular expression containing as its rightmost attribute a 
variable Go ranging over the set of various OBJects: 
(R2.3) S -> NP* (~ XCOMP* Go) = 
\[NP ((I SUBJ) = I\]\] 
\[NP OBJ2) 
\[NP ((I OSJ) = ~\],3 \[VP XCOMP) 
tv t = *~. 
Apart from potentially yielding several different c-structure 
analyses in eases where the SUBJ has been topicalized, this 
rule would only cover tim sentences (16)-(21). 
b) In addition to the modifications under a) the V constituent in 
the VP-rule is made optional and complemented by introducing a 
V' rule: 9) 
\[NP (t OBJ) = ~\] 
\[vP (-~ XCOMP) = t\] 
iv, i = tl. 
(R7) V' -> iV' (~ XCOMP) = V(\] 
v t:t rv' (t XCOMP) = f 
(~ INF =c en\]. 
A rule system of this format (if annotated appropriately) seems to 
handle the double infinitive cases as well as some of the 
extraposition structures quite nicely, however it has several 
major disadvantages: As the clause final verb-complex can be 
generated by the VP or the V' rule (or both), in many cases this 
could lead to multiple c-structure ambiguities for identical input 
strings with identical f-structure outputs. To make sure that the 
verbs appear in a 'coherent' structure obeying the constraints 
on possible topicalization and extraposition, quite a number of 
additional constraining rules presumably will be necessary. 
Finally, as in this rule system c-structure derivations may not be 
avoidable, that follow a rule like 
vp -> vP (I XCOMp) = f. 
(i.e. violate the constraint for valid c-structure derivations 
Kaplan/Bresnan (1982: 266f)), a slight increase in the generative 
power of the system may be the consequence. 
c) The VP is given up altogether in non-peripheral positions. We 
define a variable ranging over the set of grammatical functions 
that may be fulfilled by NPs (let's call it Gs) and modify our 
rules in the following way: 
Range of Variebtes: 
G =: { SUBJ, OgJ, OBJ2, XCOMP ) 
GS { SUBJ, ogJ, OBJ2 \]' 
Go =: C ogJ, OgJ2 } 
(NO), (RI), (RS), (R6) 
(R2.4) S -> 
(R3.2) VP "> 
(R7.1) V' "> 
UP* (1 XCOMP* Gs) = ~,1 
iv t : t 
It XCOMP INF =c en\] iV ~ xcoNP) = ~,\] 
tv t = t\] 
\[VP (1 XCORP* XCOMP) = 
iV(EXTRAPOSITION) =c +\]. 
,,P. II %= \[V' XCOMP) = 
v t=t \[vp (I' XCOMP* XCOMp) = f 
(~ EXTRAPOSITION) =c +\]. 
V 
4. CONCLUSION 
As we mentioned in the introduction, the analysis proposed in 4.c) 
is a very radical solution which in its present form pre- 
supposes that German is a (relatively) free word order language. 
Its advantages should be obvious from the examples given above: 
Free word order in the domain of nominal arguments, 
extraposition and topicalization of incomplete VP-nodes etc. can be 
easily covered, its disadvantages might be limitable by the 
introduction of linear precedence rules. Still, the idea of 
functional incertainty certainly represents an interesting 
mechanism, that could make it much easier to get things out of 
order. 
Footnotes: 
I) Tllis paper has been based on research presently carried out at the 
Department of Linguistics, Stuttgart, sponsored by the BUNDESHINISTER 
FUER FORSCHUNG UND TECHNOLOGiE (BMFT) (grant no. 1013207 0). 
2) For lack of space we will have to presuppose the theory of LFG as it 
was developed in Kaplan / Bresnen (1982) and Sresnan (1982B). The paper on 
'functions( incerteinty' by KAPLAN / 2AENEN to our knowledge has not been 
published yet; our use of this mechanism is for the moment restricted 
insofar as we will allow only iterations on XCOMPs and that we do not 
permit regular expressions to appear in functional centre( equations. 
3) The infinitival constructions in German ere described in great detail 
in BECH (1955), which has been claimed to be an early version of Bresnan's 
theory of control. 
4) The syntax of our grammar rules follows the syntax accepted by the 
Stuttgart LFG-system (see EISELE / DOERRE this vol.): It differs from 
traditional LFG notations only in \[\] for optional constituents and ! for 
existential constraints. The Kteene star used in functions\[ schemata is 
net yet implemented and is 'simulated' at the moment by annotating 
appropriate disjunctions. 
5) See HAIDER (1985) for some recent analysls of the sentence initial 
positions in the framework of G&B. 
6) For the use of variables ranging over sets of grammatical functions 
see WEDEKIND (1986). 
7) This prediction is relevant for the difference between the following 
sentence pairs: 
(i) Er hat versucht dieses Such zu \[esen. 
(i*) Dieses guch hat er versucht, zu tesen. 
(ii) Er ist gekommen, dieses Buch zu tesen. 
(ii') *Dieses Such ist er gekommen, zu lesen. 
\]he extraposed VP in (i) is to be analysed as XCONP, the VP in (ii) as 
XADJUNCT. 
9) See JOHNSON (1985) whose description has been m(~delied after the 
approach given in BRESNAN et el. (1982). 

REFERENCES 

BECH, Gunnar (1955), Studien ueber des Deutsche Verbum Infinitum. 2 vol. 
Kopenhagen 

BRESNAN, Joan (1982a), Centre{ and Con~olementatien. In: BRESNAN (1982b), 
282-390 

BRESNAN, Joan led.) (1982b), The Mental Representation of Grarl~atica{ 
Relations. Cambridge, Mass. 

BRESNAN, J./R.KAPLAN/S.PETERS/A.ZAEHEN (1982), Cross-Serial Dependencies 
in Dutch. In: Linguistic Ir~quiry 13.4 

EISELE, A./J. D~RRE (this vol.), A Lexical Functional Grammar System in 
Prolog. 

HAIDER, Hubert (1985), Verb second in German. To appear in: NAIDER, H. / 
PRINZHORN, M. (ads.), Verb Second. Derdrecht. 

JOHNSON, Mark (1985), An LFG Description of the Double Infinitive 
Construction in Dutch and German. unpubl. Ms., Stanford 

KAPLAN, R./J.BRESNAN (1982), Lexica\[ Functional Grammar. A Formal System 
for Grammatical Representation. In: BRESNAN (1982b)~ 173-281 

KAPLAN, R./A. ZAENEN (1986), Functional Incertainty in LFG. unpubl, ms., 
Stanford 

WEDEKIND, Juergen (1986), Ein Vorschleg zur nicht - konfiguretionalen 
Kodierung grammatischer Funktionen in der LFG. in:forum angewandte 
Linguistik, voI. 14, Tuebingen 
