Movement rules revisited" 
Eva Haji~ov~ 
Institute of Formal and Applied Linguistics 
Charles Univeristy 
Malos~'anskd n/un~st| 25 
118 00 Prague, Czech Republic 
hajicova@ufal.mff.cuni.cz 
0. Introduction 
The Praguian approach to formal description of 
language (in detail, see Sgall et aI. 1986), works 
with dependency syntax and includes the topic- 
focus articulation into the (underlying, 
tectogrammatical) syntactic representations of 
sentences, using the left-to-right order of the 
nodes of the dependency tree for that purpose. 
The edges of the tree denote the dependency 
relations and the nodes carry complex labels 
indicating their lexical and morphological values 
(e.g. Preterite). No nonterminals and no nodes 
corresponding to function words (auxiliaries, 
prepositions, conjunctions, articles) are present 
in the tree. Counterparts of function words (and 
function morphemes) are parts of the complex 
symbols of the nodes. Instead of using the notion 
of phrase, we work with subtrees (i.e. the 
governor and its dependents, or all its 
subordinate nodes, where "subordinate" is the 
transitive closure of "dependent"). 
The present paper concentrates on a 
motivating discussion pointing out the reasons 
for distinguishing three layers of the order of 
lexical occurrences (Sect. 1), presenting a 
characterization of movement rules (Sect. 2), 
discussing certain relevant relationships between 
the level of syntax and that of morphernics (Sect. 
3), and sketching a possibility how to proceed 
from surface (morphemics) to (underlying) 
syntax (Sect. 4). A way how to formalize the 
description of (underlying) syntactic slructure is 
briefly outlined in Section 5. 
1. Three layers of word order 
1.1 As has been argued in detail in Sgall et al. 
(1986), there are good reasons (concerning the 
semantic relevance as well as syntactic 
constraints of the topic-focus articulation, see 
below) to distinguish three layers of the order of 
the occurrences of lexical (autosemantic) items 
in the sentence: 
(i) the surface word order, which can be 
understood as the order of items on the level 
of morphemics; on that level the sentence 
representation has the shape of a string of 
word-forms (rather than that of a tree); 
(ii) the communicative dynamism (underlying 
word order), i.e. the order of nodes in an 
underlying (tectogrammatical) representation 
(TR), which has the shape of a dependency 
tree; the topic of the sentence is 
characterized as consisting of items which 
are less dynamic than those belonging to its 
focus; thus, if A belongs to the topic in all 
TRs of a given sentence S and B belongs to 
the focus in some of the TRs of S, then A is 
less dynamic than B; 
(iii)the systemic ordering of valency slots 
(kinds of dependency relation), which is 
specified in the valency frames (included in 
the lexical entries), see Sgall et al. 0986), 
Sgall (in press). 
Let us illustrate the mentioned three layers by the 
following example (with the capitals denoting 
the intonation center of the sentence): 
"The work on this paper was supported by the grant of the Czech Grant Agency GA~R 405/96/K214. 
49 
! 
i 
I 
I 
I 
I 
I 
I 
I 
i 
I 
i 
i 
(1) My sister was visited by a painter in PARIS 
last week. 
A difference between (i) and (ii) - the surface 
word order and the communicative dynamism - 
concerns here the fact that the most dynamic 
item (the bearer of the intonation center) does 
not occupy its prototypical position at the end of 
the sentence. Since the temporal adverbial at the 
end follows the intonation center, it can only be 
understood as belonging to the topic of the 
sentence, i.e. as less dynamic than the focus. Let 
us assume that the preferred reading of (1) - see 
Fig.1 - can be paraphrased as "As for the last 
week in my sister's life, (she) was (then) visited 
by a painter in Paris"; the order of the valency 
slots (the communicative dynamism) of the 
(tectogrammatical) syntactic tree representing 
this reading is, in ascending order, Patient 
(Objective) - Temporal-when - Actor - Locative 
(with the Patient being the least dynamic item, 
i.e. the topic proper, and the verb being less 
dynamic than the Actor and more dynamic than 
the temporal adverbial). 
visit.Pret 
sist~aris Loc / / 
I.Appurt last.Restr 
Fig. 1. A simplified TR of the sentences (1) and (2). 
As a matter of fact, another surface order 
corresponding to this underlying wee may be that 
of(2): 
(2) Last week, my sister was visited by a painter 
in PARIS. 
The relationship between (ii) and (iii) is 
reflected in (1) by the fact that the above 
mentioned ascending order of valency slots is in 
accordance with the systemic ordering within the 
focus (Actor preceding Locative), while this is 
not the case with the .position of Actor with 
regard to the two valency slots in the topic: 
Actor follows both Temporal and Objective, 
although the systemic ordering of these three 
valemcy slots is Temporal - Actor - Objective 
(see Haji6owi and Sgall 1987; Haji6ovfi 1991). 
This difference is determined by the fact that the 
parts of the topic are less dynamic than those of 
the focus, even if the degrees of dynamism are 
then not in accordance with systemic ordering. 
The restriction that the communicative 
dynamism within the focus part of the sentence 
is in accordance with systemic ordering, is 
crucial for the specification of syntactic 
representations, as will be seen in paragraph 
(ii)Co) in Sect. 5.3. 
1.2 The differences in communicative dynamism 
are relevant not only for the appropriateness of 
the use of the given sentence in this or that 
context, but also semantically, even in the 
narrow sense of differences in truth conditions. 
This clearly concerns sentences with certain 
kinds of complex quantification, such as (3), but 
also sentences without such (overt) quantifiers, 
such as (4) - (7) (I mostly use examples taken 
fi, om earlier linguistic discussions): 
(3) (a) Two languages are known by everybody 
in this ROOM. 
Co) Everybody in this room knows two 
LANGUAGES. 
(4) (a) He wrote his dissertation on SUNDAYS. 
Co) On Sundays he wrote his 
DISSERTATION. 
(5) (a) I exercise in the MORNINGS. 
Co) In the mornings I EXERCISE. 
(6) (a) John saw an EXPLOSION. 
Co) Mary saw an EXPLOSION. 
(c) John and Mary saw an EXPLOSION. 
(7) (a) An explosion was seen by JOHN. 
Co) An explosion was seen by MARY. 
(c) An explosion was seen by JOHN AND 
MARY. 
(d) An EXPLOSION was seen by Johri and 
Mary. 
With (6c) John and Mary could have seen 
different explosions, which is not the case with 
(7c), at least on the preferred reading. It is worth 
noting that the mere change of the position of the 
intonation center in (7d) gives a reading different 
! 
! 
50 
from (7c), namely the same reading as that of 
(6c). 
Examples in which the degrees of dynamism 
within topic are relevant can be presented e.g. 
from Czech: 
(8) (a) Tady byla loni MILENA. 
here was last-year Milena 
E. Here was MILENA last year. 
(b) Byla tady loni MILENA. 
was here last-year Milena 
E. Last year, MILENA was here. 
While (8a), with 'tady' \[here\] in the position 
of topic proper (i.e. the least dynamic word), 
which may carry a (contrastive) phrasal stress, 
can be paraphrased as "The place we are now is 
where MILENA was last year", (8b) means "R 
happened that MILENA was here (not 
necessarily in the place we are now, but in a 
broader sense) last year." 
In the sequel, we argue first for the usefulness 
to work with movement rules between syntax 
and morphemics (Section 2), we characterize our 
approach to these two levels of grammar 
(Section 3), focussing then on the transition from 
morphemics to syntax (Section 4) and on a 
formal specification of syntactic 
(tectogrammatical) representations and their 
transduction to morphemic ones (Section 5). 
2. Movement rules 
The examples from Section I show that it is 
necessary to work with an order of items also in 
the (underlying) syntactic representation. This is 
much easier with dependency syntax than with 
the constituency based descriptions where order 
is handled as closely connected with the 
determination of syntactic re'\[ations. However, it 
is impossible to straightforwardly copy in these 
representations the surface order, due to the 
differences between these two layers, as 
illustrated in I.I above, in the discussion of 
example (I). 
These differences may be described by means 
of movement rules, or perhaps e.g. of numerical 
indices for underlying positions and of specific 
topic and focus slots, as Starosta (1993; in press) 
proposes. It would deserve a systematic 
investigation to find whether such an approach 
allows to account in a natural way not only for 
51 
the subtle relationships between surface word 
order, communicative dynamism and the 
systemic ordering, but also for those concerning 
the different positions of quantifiers or of 
'focalizers', cf. Haji~ov~i et al. (in press). Of 
course, an important methodological requirement 
is to minimize the number of movement rules, 
and of all distinctions between morphemic 
strings and syntactic representations, be they 
treated as constituting two layers of description 
or as corresponding to different subsets of 
symbols within complex node values. In any 
case, at least the following issues have to be 
accounted for, and we indicate below how this 
can be done by means of movement rules (cf. 
Sect. 5.4): 
(a) Sentences with a secondary (other than 
rightmost) placement of the intonation 
center, which marks the position of focus, 
cf. (1) and its TR in Fig. 1 above; see also 
(7d); for Czech, the rule may transfer the 
righmost item of the TR to the leftmost 
position, marking the moved item as the 
bearer of the intonation center of the 
sentence (which then, in phonemics, gets a 
falling stress). 
(b) Grammatically determined positions of 
verbs, adjectives, clitics, and so on, in 
different languages; the adjective primarily 
is more dynamic than its head noun, but it 
is placed to its left in English surface word 
order; in Czech clities usually are placed in 
the "second" (Wackemagel's) position, even 
if they depend on an infinitive standing 
after the main verb: 
(9) Martin mi ji tu chtel UKJ~AT. 
Martin to-me her here wanted to-show 
E. Martin wanted to SHOW me her 
here. 
chtel.Pret 
Martin.Act tddzat.Pat n'nSLoc 
Fig. 2. A simplified TR of sentence (9). 
I 
I 
I 
I 
I 
! 
i 
! 
I 
i 
I 
i 
i 
I 
i 
I 
I 
This also concerns other dependents of an 
infinitive, if they belong to the topic: 
(l 0) Martin mi Milenu zam3~lel ~T. 
Martin to-me MilenaA,~ intended to-show 
E. Martin intended to SHOW me Milena. 
Wackernagers position, into which the 
moved items usually are placed, is the 
position just after the lefanost node of the 
uppermost part of the tree (and, if the 
lei~most node differs from the verb, also 
after all nodes subordinated to the lefanost 
one). 
(c) Positions of function words (articles, 
prepositions, conjunctions, auxiliaries), 
which need not have any nodes as their 
underlying counterparts (just indices within 
node labels, see Section 4 below): in the 
morphemic sizing (i.e. in the surface word 
order) they are usually placed at the 
beginning of their word groups; the rules 
describing these movements are to be 
combined with changing the indices in the 
complex node labels into symbols of 
articles, prepositions, etc. 
(l l) Jim visited me, since he wanted to ask 
me for ADVICE. 
(d) Also other cases of apparent non- 
projectivity (i.e. peripheral, strongly limited 
cases) can be described by movement rules 
concerning morphemics: 
(12) I met a man yesterday, who asked me 
for your ADDRESS. 
The order in the TR: I - yesterday - met - a 
man, who... (see Fig. 3). 
(13) a largcr town than Boston: 
The order in the TR: larger - than Boston - 
town. 
The use of movement rules, rather than of 
numerical indices for underlying positions 
enables us to work with perspicuous syntactic 
representations reflecting all the differences 
illustrated in the above quoted examples, 
including an economical handling of systemic 
ordering and dynamism. 
Even when attempting at a'monostratal 
description, it is always necessary to distinguish 
between morphemics and syntax, e.g. by using 
abstract symbols in the complex labels of nodes; 
52 
thus, also with such an approach the problem 
stressed by Starosta (1993) is still present, 
namely the danger that abstract symbols might 
get too far from the empirical properties of 
language. As mentioned in Section 2 above, the 
requirement of the minimalization of the number 
of rules (or symbols) is to be given due attention. 
meet.Pret 
I.Act yesterday.Temp-when man.Pat 
ask.Pret.Res~ /2 
who.Act I.Addr address.Pat / 
you.Appurt 
Fig.3. A simplified TR of sentence (12). 
3. Syntax and morphemics 
Using movement rules to describe 
relationships between syntax and morphemics 
makes it possible to treat these two levels apart, 
with simple strings of (complex) symbols as 
morphemic representations, and trees (or more 
complex networks) as syntactic ones. It is 
necessary to distinguish these two levels at least 
for two reasons: 
(i) the relation between a function word and its 
autosemantic 'partner' differs essentially 
from the dependency relation: a preposition 
or article cannot occupy any other position 
than that of accompanying a noun; an 
auxiliary or (typically) a conjunction always 
accompanies a verb; in this point they are 
similar to endings, rather than to adjuncts; it 
is essentially more economical to have just 
i! 
io 
(ii) 
indices in complex node labels than large 
numbers of nodes; 
to work with a node for every surface 
lexical occurrence (and to use no other 
nodes) is difficult, not only with respect to 
the description of deletions of different 
kinds, but first of all with surface word 
forms such as French du, au, or German 
rum, beim; these forms show that the 
boundary between surface word forms is not 
reliable. Such cases can be found in many 
languages, of. e.g. in Czech the conjunction 
-li 'if (occurring only after a verb form and 
attached to it by a hyphen, so that the 
intermediate position between a function 
word and a suffix of the verb is clearly 
manifested), or the clitic s (auxiliary of the 
2nd Person Sing. Preterite), written together 
with (any!) preceding word form, not only 
after a verb as in dostals 'you got' or after 
the subject pronoun as in tys DOSTAL 'you 
GOT', but also e.g. after the object noun as 
in knihus DOSTAL 'you GOT the book'. 
4. From surface (morphemics) to 
(underlying) syntax 
In the primary unmarked situation, 
morphemes (endings, prepositions, other 
function words) express dependency relations 
and values of morphological categories (number, 
definitness, tense, aspect, modalities, degrees of 
comparison), whereas surface (morphemic) word 
order expresses the basic scale of the topic-focus 
articulation, i.e. that of communicative 
dynamism. Prototypically these two orders are 
identical, the differences - described by 
movement rules - are relatively rare and can be 
described by a limited number of rules. 
The core of the transition from morphemic 
strings to syntactic representations (preliminary 
formulations of implemented procedures for 
which can be found with Kirschner 1987; 
Haji~ovfi et al. 1994) then is to describe the 
transition from endings and function words to 
their functions. Thus, a Genitive typically 
depends on the adjacent noun, a prepositional 
group usually expresses an adverbial (dependent 
on the verb), prepositions have either their 
primary 'meanings', as e.g. E. with: 
Accompaniment or Means, from: Directional.I, 
through: Directional.3, or they have their 
syntactic functions (they express certain kinds of 
dependency relation; perhaps it would be more 
exact to see dependency as a family of relations) 
- to: Addressee, by: Actor with Passive. 
Another part of the task is to proceed from 
surface word order and certain properties of 
sentence prosody (positions of the intonation 
center or the main stress of the sentence, and of 
phrasal stress) to the scale of dynamism. In non- 
prototypical (not just marginal) cases the word 
order expresses syntactic (dependency) relations. 
Among such cases there is the English 
'configurationality' with the order Actor, verb 
and Objective; however, even in English, as soon 
as other kinds of the dependency relations are 
encountered, it is necessary to take morphemes 
into account (in English - prepositions, see 
above; in most Continental languages - case 
endings; in the agglutinating languages of Asia 
and other regions - suffixes; verb categories are 
practically always expressed by morphemes, 
etc.). 
Furthermore, it is necessary to restore the 
deleted items; this task has often been discussed 
with different frameworks, but an adequate 
solution still seems to be connected with many 
open questions. One of them is that it is not easy 
to draw the boundary line between those cases in 
which the deleted lexical unit is to be inserted as 
such, and cases in which just an anaphoric or 
indexical item is appropriate. The former case 
clearly concerns at least the basic kinds of 
coordination deletion (e.g. with blue and yellow 
flag it is appropriate to restore blue flag in the 
TR). The latter case can be illustrated by the 
deleted Directional with George has arrived on 
Sunday - either here or there is present in the 
TR, i.e. the sentence is ambiguous in this point. 
5. Formalisation of dependency syntax 
5.1 A descriptive framework meeting the 
conditions specified in the preceding sections 
can now be briefly characterized. We present 
first an illustration of how the complex 
tectogrammatical networks can be formally 
described (Sect. 5.1), then we characterize the 
form of a lexical entry (5.2) and proceed to a 
brief sketch of a specification of the set of TRs 
(5.3), adding a partial illustration of movement 
53 
! 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
i 
I 
I 
I 
i 
I 
i 
I 
rules used in transducing TRs to morphemic 
representations (5.4). 
A dependency tree is not sufficient, since it 
cannot reflect the complex relationships between 
dependency and coordination, if we do not want 
to neglect the difference between the binary 
dependency relations and the coordinated (and 
appositional) constructions having an indefinite 
number of members. However, even a network 
with a greater number of dimensions, which in 
this sense can serve as the shape of a TR, can be 
formally described in the form of its one-to-one 
linearization (see Petkevi6 1995; Sgall 1997), cf. 
examples (14) and (15); every dependent item or 
group is enclosed in a pair of parentheses and the 
subscripts, indicating the dependency relations, 
are placed at the parenthesis that is closer to the 
head and can be read as Restrictive (adjunct), 
Directional, Appurtenance (broader than just 
possession); Conjunction is understood as a kind 
of coordination: 
(14) My younger brother arrived there 
YESTERDAY. 
(14') ((mY)Restr (younger)Restr brother)Act 
arrive.Pret (Dir there) (Temp yesterday) 
(15) Mary and John, our neighbors, who are a 
nice pair, ARRIVE. 
(15') (((Mary John)Con j (we)Appur t 
neighbor)Appos (Restr (Rel)Act be.ires 
(Obj (nice)Restr pair)))Act (here)Dir 
arrive.Ires 
5.2 The relationship between lexical and 
grammatical information can be characterized by 
including much of grammatical information in 
the lexicon (especially in the valency frames or 
grids). The lexical entry contains the following 
parts: 
(a) the lexical meaning, which itself has its 
internal structure; in the cases of lexical 
ambiguity there are several representations, 
i.e. several lexical units; 
(b) the slots for values of relevant grammatical 
categories, i.e. of grammatemes; while e.g. 
gender with nouns is specified here, other 
grammatemes (number and definiteness with 
nouns, or tense, aspect, modalities with 
verbs, degrees of comparison with 
(c) 
54 
adjectives) get their values only for 
individual occurrences of the word forms in 
discourses; 
the valency frame: optional and obligatory 
dependents are ordered in accordance with 
systemic ordering; arguments or 'inner 
participants' (occurring at most once with a 
head node) and obligatory adjuncts are 
indicated (e.g. arrive at a place, behave 
somehow, last how long,...); optional 
adjuncts may be indicated by lists 
concerning individual word classes; surface 
deletability is indicated (e.g. Directional.2 
with to arrive is deletable, Objective with to 
meet is not (We met there is a case of 
reciprocity, rather than of deletion); further 
data concern an optional or obligatory 
controller (e.g. Actor is an obligatory 
controller with to try, an optional one with to 
decide; Addressee is an optional controller 
with to advise, to forbid), the dependent's 
ability to occupy certain syntactic positions 
(e.g. of Subject with Passivization, of a wh- 
element) or to constitute barriers for 
movement, and subcategorization 
conditions. 
5.3 The class of syntactic representations can be 
specified by means of a small number of general 
principles describing the core of grammar and by 
specific rules for peripheral patterns. 
The general part of this specification can have 
the shape of a generative procedure (first 
formulated by Haji~ov6 et al. 1991): 
(i) generate a node: 
(a) create the node either as the root of the 
representation, or as a node dependent on 
another one and placed to the right of all its 
sister nodes; 
Co) choose its lexical value and the values of its 
grammatemes, taking into account (by 
unification) the subcategorization conditions 
of the mother node and the restrictions on 
the combinations of grammatemes as 
specified in the lexical entry of the head or 
in the data concerning the respective word 
class; 
(c) if a root is being generated, its 
grammatemes determine it as a finite verb 
form; it may be denoted as CB, contextually 
bound, in topic (otherwise it is non-bound, 
in focus); the node that has just been 
generated is considered as node n in the 
next step; 
(ii) if the frame of node n contains a 
complementation, then it is possible to generate 
(ii) (a) either a left daughter of n, assign it 'CB' 
and a complementation value chosen 
from the frame of n, 
(ii) Co) or a right daughter of n, and assign it a 
complementation value from the left end 
of n's frame (this ensures that the 
underlying order of the 
complementations in the focus part of the 
sentence will obey the systemic ordering, 
see the end of Sect. 1.1 above). 
NOTE: If the chosen complementation value 
is an inner participant, it is deleted in the 
frame of n (as saturated); "from the left end" 
means that optional complementations can be 
skipped and deleted in the frame of n. 
(iii) if the frame of n contains no 
complementation, then the mother node of n 
is considered as node n; if no mother node is 
present; the procedure (generating the tree 
from the top down and from the left to the 
right) is finished; 
(iv) only representations containing a focus are 
accepted, i.e. only those whose branch going 
from the root to the rightmost daughter of 
the rightmost daughter of ... of the root 
includes a non-bound node. 
A declarative specification of the syntactic 
representations is well possible, if the concept of 
unification is so complemented as 
(a) to distinguish saturated items (cf. the Note 
above) and 
(b) to check the correspondence of the order of 
non-bound sister nodes with the systemic 
ordering. 
Then it can be checked whether the tree 
unifies with conditions stated in the lexicon, i.e. 
whether all obligatory complementations of the 
heads present in the representation are among the 
daughters of the heads, whether only 
complementations contained in the frame of any 
head (or in the relevant list of free modifications) 
occur as the bead's dependents, and whether 
subcategorization (and similar) restrictions are 
met. 
As we mentioned, this specification covers 
the core of sentence syntax; it can be completed 
to cover coordinated structures (see Petkevi~ 
1995), negation and other focalizers ('only', 
'also', 'willingly', etc.), in specific positions 
(prototypically they occur at the beginning of the 
focus; see Haji~ovk, Partee and Sgall, in press). 
The relatively very simple form of the 
representations (strings of bracketted and 
indexed symbols) may be helpful in explaining 
the easiness of the child's acquisition of the 
mother tongue, although the representations 
contain all semantically relevant information 
immediately comprised in the structure of the 
sentence. 
5.4 TRs of the form described in 5.3 can be 
transduced to morphemic representations 
(sUings) in that prototypically the left-to-right 
order of their nodes is retained, being changed 
only in specific cases. 
One type of such movement rules takes care 
of placing the most dynamic item (the bearer of 
the intonation center of the sentence) more to the 
left. In English this concerns first of all those 
cases in which the syntactic functions of 
participants (arguments) and adjuncts dete,~-~-~ine 
certain restrictions of their surface positions. 
Thus, e.g. sentence (1), repeated here for 
convenience, can be handled with the use of the 
following rule (presented here with many 
simplifications): 
(1) My sister was visited by a painter in PARIS 
last week. 
Rule MI: If x.Loc as the rightmost item ofa TR 
is preceded by y.Temp (which is placed between 
x and the verb), then y.Temp will be moved to 
the right of xF.Loc in the morphemic string; with 
the chosen linearization of TRs (see Sect. 5.1) 
the rule can be technically written (with many 
simplifications) as follows: 
X (U y.Temp W) Y (Z x.Loc) --> 
X Y (Z xF.Loc) (U y.Temp W) 
where X and Y are well parenthesized strings, x 
stands for a noun, pronoun or verb, y stands for a 
noun, adverb or verb (again, with their 
dependents), the superscript F denotes the 
placement of the intonation center of the 
sentence, Z, W, U, V are (possibly empty) 
strings of well parenthesized items (dependents 
of x and y), X (U y.Temp W) Y (Z x.Loc) is a 
55 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
I 
i 
I 
i 
well formed TR (according to Sect. 5.3), and the 
parentheses are supposed to be removed by one 
of the last steps of the procedure transducing 
TRs into morphemic strings. 
Rules of a similar kind can determine the 
movement of the subject (be it derived from 
Actor, or, with passivization, from Patient or 
Addressee) to the left of the verb in case its 
tectogrammatical source is more dynamic than 
the verb and thus stands to the right of the verb 
in the underlying order. 
Another kind of rules concerns the changes of 
tectogrammatieal values such as Plural, Preterite, 
Feminine, Comparative into morphemic items 
(endings, alternations, affixes, function words) 
and the placement of these morphemic items. 
Thus, e.g. for the treatment of nouns rules as the 
following one are needed: 
Rule M2: If x in a node label x.Num.Def is a 
noun, Num the value of Number (Sg or P1) and 
Def that of Definiteness (+ or-), then 
(i) rewrite + (-) into Artdf (Arfin) and place this 
symbol to the left of the nodes subordinated 
to node x (including x itself); recall that 
'subordinated' is the transitive closure of 
'depending'; Artdf will be transduced into the 
phonemic and phonetic (or graphemic) 
shapes of the, and Artin into a, an, or (with 
plural) into zero, 
(ii) rewrite PI (Sg) into -s (0) (this certainly 
requires to take into account also several 
specific sets of exceptions before Rule M2 is 
applied, cf. forms like mice, oxen, loaves); -s 
will be then changed into the surface shape of 
the Plural (or 3rd Person Singular) ending by 
the subsequent components of the procedure. 
The simplified technical form of M2 is: 
(Z x.Num.Def W) -, (Art Z x.End W) 
where x stands for a noun or a pronoun, Z and W 
are (possibly empty) slrings of well 
parenthesized items (dependents of x), Art is 
Artdef (Artin) if Defis + (-), End is -s (0) ifNum 
is PI (Sg). 
Still another type of movement rules is 
needed for apparently non-projective 
con.structions, illustrated by ex. (13), repeated 
here for convenience: 
(13) a larger town than Boston 
M3: ((Art 2 X A.Comp (than U Art I N1.End) W 
N2.End Z) --~ Art 2 X A.Comp W N2.End 2 
Z than Art 1 U N 1.End 1 
where N 1 and N 2 are nouns, A is an adjective, 
X, Z, U and W are (possl~oly empty) strings of 
well parenthesized items, Art i is Artdef or Artin 
(or zero, with a proper name), End i is -s or 0 for 
i-- 1,2. 
Rules of the last mentioned type are supposed 
to be applied after rules of the types M 1 and M2. 
Many technical as well as empirical issues are 
still open and constitute interesting topics for 
further research. 

References
Haji~,ovfi E. (1991): "Free" word order described 
without unnecessary complexity. Theoretical 
Linguistics 17:99-106. 
Hajifovfi E., Panevovfi J. and P. Sgall (1991): 
Why do we use dependency grammar? In: 
Buffalo Working Papers in Linguistics, Special 
Issue for Paul Garvin (ed. by W. W61ck). 
Buffalo:State University of New York, 90-93. 
Hajifovfi E., Partee B. and P. Sgall (in press): 
Topic-focus articulation, tripartite structures, 
and semantic content. To be published by 
Kluwer, Amsterdam. 
Hajifovfi E. and P. Sgall (1987): The ordering 
principle. Journal of Pragmatics 11,435-454. 
Hajifovfi E., P. Sgall and H. Skoumalovfi (1994): 
An automatic procedure for topic-focus 
identification. Computational Linguistics 
21:81-94. 
Kirschner Z. (1987): APAC3-2: An English-to- 
Czech machine translation system. Explizite 
Beschreibung der Sprache mad automatische 
Textbearbeitung XIII. Prague: Faculty of 
Mathematics and Physics, Charles University. 
Petkevi6 V. (1995): A new formal specification 
of underlying structures. Theoretical 
Linguistics 21:7-61. 
Sgall P. (1997): Valency and underlying 
structure. An alternative view on dependency. 
In: L. Wanner (ed.): Recent trends in Meaning- 
Text Theory. Amsterdam/ Philadelphia: 
Benjamins, 149-166. 
Sgall P. (in press): On the usefulness of 
movement rules. In: Caron B. (ed.), Acres du 
16e Congr~s International des Linguistes 
(Paris 20-25 juillet 1997), Oxford: Elsevier 
Sciences. 
Sgall P., Haji~ovA E. and J. Panevov/l (1986): 
The meaning of the sentence in its semantic 
and pragmatic aspects, ed. by J. L. Mey, 
Dordrecht:Reidel - Prague:Academia. 
Starosta S. (1993): Word order and focus in 
constrained dependency grammar. In: 
Haji~ov~ E. (ed.), Functional description of 
language. Prague: Charles University, 237- 
252. 
Starosta S. (in press): Comments on Petr Sgall's 
"On the usefulness of movement rules". In: 
Caron B. (ed.), Acres du 16e Congr s 
International des Linguistes (Paris 20-25 
juillet 1997), Oxford: Elsevier Sciences. 
