Incremental Parsing, or Incremental Grammar? 
Matthew Purvery and Ruth Kempsonz
Departments of yComputer Science and zPhilosophy,
King’s College London, Strand,
London WC2R 2LS,
UK
fmatthew.purver, ruth.kempsong@kcl.ac.uk
Abstract
Standard grammar formalisms are de ned with-
out re ection of the incremental and serial na-
ture of language processing, and incremental-
ity must therefore be re ected by independently
de ned parsing and/or generation techniques.
We argue that this leads to a poor setup for
modelling dialogue, with its rich speaker-hearer
interaction, and instead propose context-based
parsing and generation models de ned in terms
of an inherently incremental grammar formal-
ism (Dynamic Syntax), which allow a straight-
forward model of otherwise problematic dia-
logue phenomena such as shared utterances, el-
lipsis and alignment.
1 Introduction
Despite increasing psycholinguistic evidence of
incrementality in language processing, both in
parsing (see e.g. (Crocker et al., 2000)) and in
production (Ferreira, 1996), there is almost uni-
versal agreement that this should not be re-
 ected in grammar formalisms which constitute
the underlying model of language (for rare ex-
ceptions, see (Hausser, 1989; Kempson et al.,
2001)). Constraint-based grammar formalisms
are accordingly de ned neutrally between either
of these applications, with parsing/generation
systems (whether incremental or not) de ned
as independent architectures manipulating the
same underlying system.1 Such assumptions
however lead to formal architectures that are
relatively poorly set up for modelling dialogue,
for they provide no basis for the very rich de-
gree of interaction between participants in dia-
logue. A common phenomenon in dialogue is
 Related papers from the point of view of generation
rather than parsing, and from the point of view of align-
ment rather than incrementality, are to be presented at
INLG’04 and Catalog’04 respectively.
1Authors vary as to the extent to which these archi-
tectures might be de ned to be reversible. See (Neu-
mann, 1994).
that of shared utterances (Clark and Wilkes-
Gibbs, 1986), with exchange of roles of parser
and producer midway through an utterance:2
(1)
Daniel: Why don’t you stop mumbling
and
Marc: Speak proper like?
Daniel: speak proper?
(2) Ruth: What did Alex . . .Hugh: Design? A kaleidoscope.
Such utterances clearly show the need for a
strictly incremental model: however, they are
particularly problematic for any overall archi-
tecture in which parsing and generation are in-
dependently de ned as applications of a use-
neutral grammar formalism which yields as out-
put the set of well-formed strings, for in these
types of exchange, the string uttered by one and
parsed by the other need not be a wellformed
string in its own right, so will not fall within
the set of data which the underlying formalism
is set up to capture. Yet, with the transition be-
tween interlocutors seen as a shift from one sys-
tem to another, each such substring will have to
be characterised independently. Many other di-
alogue phenomena also show the need for inter-
action between the parsing and generation pro-
cesses, among them cross-speaker ellipsis (e.g.
simple bare fragment answers to wh-questions),
and alignment (Pickering and Garrod, 2004), in
which conversational participants mirror each
other’s patterns at many levels (including lexi-
cal choice and syntactic structure).
The challenge of being able to model these
phenomena, problematic for theorists but ex-
tremely easy and natural for dialogue partici-
pants themselves, has recently been put out by
Pickering and Garrod (2004) as a means of eval-
uating both putative grammar formalisms and
2Example (1) from the BNC,  le KNY (sentences
315{317).
models of language use. In response to this chal-
lenge, we suggest that an alternative means of
evaluating parsing implementations is by eval-
uation of paired parsing-generation models and
the dialogue model that results. As an illustra-
tion of this, we show how if we drop the assump-
tion that grammar formalisms are de ned neu-
trally between parsing and production in favour
of frameworks in which the serial nature of lan-
guage processing is a central design feature (as
in Dynamic Syntax: (Kempson et al., 2001)),
then we can de ne a model in which incremental
sub-systems of parsing and generation are nec-
essarily tightly coordinated, and can thus pro-
vide a computational model of dialogue which
directly corresponds with currently emerging
psycholinguistic results (Branigan et al., 2000).
In particular, by adding a shared model of con-
text to previously de ned word-by-word incre-
mental parsing and generation models, we show
how the switch in speaker/hearer roles during
a shared utterance can be seen as a switch be-
tween processes which are directed by di erent
goals, but which share the same incrementally
built data structures. We then show how this
inherent incrementality and structure/context
sharing also allows a straightforward model of
cross-speaker ellipsis and alignment.
2 Background
Dynamic Syntax (DS) (Kempson et al., 2001) is
a parsing-directed grammar formalism in which
a decorated tree structure representing a se-
mantic interpretation for a string is incremen-
tally projected following the left-right sequence
of the words. Importantly, this tree is not a
model of syntactic structure, but is strictly se-
mantic, being a representation of the predicate-
argument structure of the sentence. In DS,
grammaticality is de ned as parsability (the
successful incremental construction of a tree-
structure logical form, using all the information
given by the words in sequence): there is no cen-
tral use-neutral grammar of the kind assumed
by most approaches to parsing and/or gener-
ation. The logical forms are lambda terms of
the epsilon calculus (see (Meyer-Viol, 1995) for
a recent development), so quanti cation is ex-
pressed through terms of type e whose complex-
ity is re ected in evaluation procedures that ap-
ply to propositional formulae once constructed,
and not in the tree itself. With all quanti cation
expressed as type e terms, the standard grounds
for mismatch between syntactic and semantic
analysis for all NPs is removed; and, indeed, all
syntactic distributions are explained in terms of
this incremental and monotonic growth of par-
tial representations of content. Hence the claim
that the model itself constitutes a NL grammar
formalism.
Parsing (Kempson et al., 2001) de nes pars-
ing as a process of building labelled semantic
trees in a strictly left-to-right, word-by-word in-
cremental fashion by using computational ac-
tions and lexical actions de ned (for some natu-
ral language) using the modal tree logic LOFT
(Blackburn and Meyer-Viol, 1994). These ac-
tions are de ned as transition functions be-
tween intermediate states, which monotonically
extend tree structures and node decorations.
Words are speci ed in the lexicon to have as-
sociated lexical actions: the (possibly partial)
semantic trees are monotonically extended by
applying these actions as each word is consumed
from the input string. Partial trees may be un-
derspeci ed: tree node relations may be only
partially speci ed; node decorations may be de-
 ned in terms of unful lled requirements and
metavariables; and trees may lack a full set of
scope constraints. Anaphora resolution is a fa-
miliar case of update: pronouns are de ned to
project metavariables that are substituted from
context as part of the construction process. Rel-
ative to the same tree-growth dynamics, long-
distance dependency e ects are characterised
through restricted licensing of partial trees with
relation between nodes introduced with merely
a constraint on some  xed extension (following
D-Tree grammar formalisms (Marcus, 1987)),
an underspeci cation that gets resolved within
the left-to-right construction process.3 Quanti-
fying terms are also built up using determiner
and noun to yield a partially speci ed term e.g.
( ; y; Man0(y)) with a requirement for a scope
statement. These scope statements, of the form
x < y (‘the term binding x is to be evaluated
as taking scope over the term binding y’), are
added to a locally dominating type-t-requiring
node. Generally, they are added to an accu-
mulating set following the serial order of pro-
cessing in determining the scope dependency,
but inde nites (freer in scope potential) are as-
signed a metavariable as  rst argument, allow-
3In this, the system is also like LFG, modelling long-
distance dependency in the same terms as functional un-
certainty (Kaplan and Zaenen, 1989), di ering from that
concept in the dynamics of update internal to the con-
struction of a single tree.
Figure 1: Parsing \john likes mary" . . . . . . and generating \john likes mary"
fg
fjohn0g f}g
john
fg
fjohn0g fg
flike0g f}g
likes
flike0(mary0)(john0);}g
fjohn0g flike0(mary0)g
flike0g fmary0g
mary
fg
fjohn0g f}g
FAIL FAIL
john likes mary
fg
fjohn0g fg
flike0g f}g
FAIL
likes
mary
flike0(mary0)(john0);}g
fjohn0g flike0(mary0)g
flike0g fmary0g
mary
ing selection from any term already added, in-
cluding temporally-sorted variables associated
with tense/modality speci cations. The gen-
eral mechanism is the incremental analogue of
quanti er storage; and once a propositional for-
mula of type t has been derived at a node with
some collection of scope statements, these are
jointly evaluated to yield fully expanded terms
that re ect all relative dependencies within the
restrictor of the terms themselves. For example,
parsing A man coughed yields the pair Si < x,
Cough0( ; x; Man0(x)) (Si the index of evalua-
tion), then evaluated as Man0(a) ^ Cough0(a)
where a = ( ; x; Man0(x) ^ Cough0(x)).4
Once all requirements are satis ed and all
partiality and underspeci cation is resolved,
trees are complete, parsing is successful and the
input string is said to be grammatical. Central
to the formalism is the incremental and mono-
tonic growth of labelled partial trees: the parser
state at any point contains all the partial trees
which have been produced by the portion of the
string so far consumed and which remain can-
didates for completion.5
4For formal details of this approach to quanti cation
see (Kempson et al., 2001) chapter 7; for an early imple-
mentation see (Kibble et al., 2001).
5Figure 1 assumes, simplistically, that linguistic
names correspond directly to scopeless names in the log-
Generation (Otsuka and Purver, 2003;
Purver and Otsuka, 2003) (hereafter O&P)
give an initial method of context-independent
tactical generation based on the same incre-
mental parsing process, in which an output
string is produced according to an input
semantic tree, the goal tree. The generator
incrementally produces a set of corresponding
output strings and their associated partial trees
(again, on a left-to-right, word-by-word basis)
by following standard parsing routines and
using the goal tree as a subsumption check.
At each stage, partial strings and trees are
tentatively extended using some word/action
pair from the lexicon; only those candidates
which produce trees which subsume the goal
tree are kept, and the process succeeds when
a complete tree identical to the goal tree is
produced. Generation and parsing thus use
the same tree representations and tree-building
actions throughout.
3 Contextual Model
The current proposed model (and its imple-
mentation) is based on these earlier de nitions
but modi es them in several ways, most signif-
icantly by the addition of a model of context:
ical language that decorate the tree.
while they assume some notion of context they
give no formal model or implementation.6 The
contextual model we now assume is made up not
only of the semantic trees built by the DS pars-
ing process, but also the sequences of words and
associated lexical actions that have been used
to build them. It is the presence of (and as-
sociations between) all three, together with the
fact that this context is equally available to both
parsing and generation processes, that allow our
straightforward model of dialogue phenomena.7
For the purposes of the current implementa-
tion, we make a simplifying assumption that
the length of context is  nite and limited to the
result of some immediately previous parse (al-
though information that is independently avail-
able can be represented in the DS tree format,
so that, in reality, larger and only partially or-
dered contexts are no doubt possible): context
at any point is therefore made up of the trees
and word/action sequences obtained in parsing
the previous sentence and the current (incom-
plete) sentence.
Parsing in Context A parser state is there-
fore de ned to be a set of triples hT; W; Ai,
where T is a (possibly partial) semantic tree,8
W the sequence of words and A the sequence
of lexical and computational actions that have
been used in building it. This set will initially
contain only a single triple hTa;;;;i (where Ta
is the basic axiom taken as the starting point of
the parser, and the word and action sequences
are empty), but will expand as words are con-
sumed from the input string and the corre-
sponding actions produce multiple possible par-
tial trees. At any point in the parsing process,
the context for a particular partial tree T in
6There are other departures in the treatment of linked
structures (for relatives and other modi ers) and quan-
ti cation, and more relevantly to improve the incremen-
tality of the generation process: we do not adopt the
proposal of O&P to speed up generation by use of a re-
stricted multiset of lexical entries selected on the basis
of goal tree features, which prevents strictly incremental
generation and excludes modi cation of the goal tree.
7In building n-tuples of trees corresponding to
predicate-argument structures, the system is similar to
LTAG formalisms (Joshi and Kulick, 1997). However,
unlike LTAG systems (see e.g. (Stone and Doran, 1997)),
both parsing and generation are not head-driven, but
fully (word-by-word) incremental. This has the ad-
vantage of allowing fully incremental models for all
languages, matching psycholinguistic observations (Fer-
reira, 1996).
8Strictly speaking, scope statements should be in-
cluded in these n-tuples { for now we consider them as
part of the tree.
this set can then be taken to consist of: (a) a
similar triple hT0; W0; A0i given by the previous
sentence, where T0 is its semantic tree repre-
sentation, W0 and A0 the sequences of words
and actions that were used in building it; and
(b) the triple hT; W; Ai itself. Once parsing is
complete, the  nal parser state, a set of triples,
will form the new starting context for the next
sentence. In the simple case where the sentence
is unambiguous (or all ambiguity has been re-
moved) this set will again have been reduced
to a single triple hT1; W1; A1i, corresponding to
the  nal interpretation of the string T1 with its
sequence of words W1 and actions A1, and this
replaces hT0; W0; A0i as the new context; in the
presence of persistent ambiguity there will sim-
ply be more than one triple in the new context.9
Generation in Context A generator state
is now de ned as a pair (Tg; X) of a goal tree
Tg and a set X of pairs (S; P), where S is a
candidate partial string and P is the associated
parser state (a set of hT; W; Ai triples). Ini-
tially, the set X will usually contain only one
pair, of an empty candidate string and the stan-
dard initial parser state, (;;fhTa;;;;ig). How-
ever, as both parsing and generation processes
are strictly incremental, they can in theory start
from any state. The context for any partial tree
T is de ned exactly as for parsing: the previ-
ous sentence triple hT0; W0; A0i; and the cur-
rent triple hT; W; Ai. Generation and parsing
are thus very closely coupled, with the central
part of both processes being a parser state: a set
of tree/word-sequence/action-sequence triples.
Essential to this correspondence is the lack of
construction of higher-level hypotheses about
the state of the interlocutor. All transitions
are de ned over the context for the individ-
ual (parser or generator). In principle, con-
texts could be extended to include high-level
hypotheses, but these are not essential and are
not implemented in our model (see (Millikan,
2004) for justi cation of this stance).
4 Shared Utterances
One primary evidence for this close coupling
and sharing of structures and context is the ease
with which shared utterances can be expressed.
O&P suggest an analysis of shared utterances,
9The current implementation of the formalism does
not include any disambiguation mechanism. We simply
assume that selection of some (minimal) context and at-
tendant removal of any remaining ambiguity is possible
by inference.
Figure 2: Transition from hearer to speaker: \What did Alex .../ ...design?"
Pt =
D f+Qg
fWHg falex0gf?Ty(e ! t);}g
;fwhat;did;alexg;fa1; a2; a3g
E
Gt =
 f+Q; design
0(WH)(alex0)g
falex0g fdesign(WH)g
fWHgfdesign0g
;
 
;;
D f+Qg
fWHg falex0gf?Ty(e ! t);}g
;fwhat;did;alexg;fa1; a2; a3g
E !
G1 =
 f+Q; design
0(WH)(alex0)g
falex0g fdesign0(WH)g
fWHgfdesign0g
;
 
fdesigng;
D
f+Qg
fWHgfalex0g f?Ty(e ! t)g
f}gfdesign0g
;f: : :;designg;f: : :; a4g
E !
and this can now be formalised given the cur-
rent model. As the parsing and generation pro-
cesses are both fully incremental, they can start
from any state (not just the basic axiom state
hTa;;;;i). As they share the same lexical en-
tries, the same context and the same semantic
tree representations, a model of the switch of
roles now becomes relatively straightforward.
Transition from Hearer to Speaker Nor-
mally, the generation process begins with
the initial generator state as de ned above:
(Tg;f(;; P0)g), where P0 is the standard initial
\empty" parser state fhTa;;;;ig. As long as a
suitable goal tree Tg is available to guide gen-
eration, the only change required to generate a
continuation from a heard partial string is to
replace P0 with the parser state (a set of triples
hT; W; Ai) as produced from that partial string:
we call this the transition state Pt. The initial
hearer A therefore parses as usual until transi-
tion,10 then given a suitable goal tree Tg, forms
a transition generator state Gt = (Tg;f(;; Pt)g),
from which generation can begin directly { see
 gure 2.11 Note that the context does not
change between processes.
For generation to begin from this transition
state, the new goal tree Tg must be subsumed
by at least one of the partial trees in Pt (i.e.
the proposition to be expressed must be sub-
sumed by the incomplete proposition that has
been built so far by the parser). Constructing
10We have little to say about exactly when transitions
occur. Presumably speaker pauses and the availability
to the hearer of a possible goal tree both play a part.
11Figure 2 contains several simpli cations to aid read-
ability, both to tree structure details and by show-
ing parser/generator states as single triples/pairs rather
than sets thereof.
Tg prior to the generation task will often be a
complex process involving inference and/or ab-
duction over context and world/domain knowl-
edge { Poesio and Rieser (2003) give some idea
as to how this inference might be possible { for
now, we make the simplifying assumption that
a suitable propositional structure is available.
Transition from Speaker to Hearer At
transition, the initial speaker B’s generator
state G0t contains the pair (St; P0t), where St is
the partial string output so far, and P 0t is the
corresponding parser state, the transition state
as far as B is concerned.12 In order for B to
interpret A’s continuation, B need only use P 0t
as the initial parser state which is extended as
the string produced by A is consumed.
As there will usually be multiple possible par-
tial trees at the transition point, A may con-
tinue in a way that does not correspond to B’s
initial intentions { i.e. in a way that does not
match B’s initial goal tree. For B to be able
to understand such continuations, the genera-
tion process must preserve all possible partial
parse trees (just as the parsing process does),
whether they subsume the goal tree or not, as
long as at least one tree in the current state does
subsume the goal tree. A generator state must
therefore rule out only pairs (S; P) for which P
contains no trees which subsume the goal tree,
rather than thinning the set P directly via the
subsumption check as proposed by O&P.
It is the incrementality of the underlying
grammar formalism that allows this simple
switch: the parsing process can begin directly
12Of course, if both A and B share the same lexical
entries and communication is perfect, Pt = P0t, but we
do not have to assume that this is the case.
from a state produced by an incomplete gener-
ation process, and vice versa, as their interme-
diate representations are necessarily the same.
5 Cross-Speaker Ellipsis
This inherent close coupling of the two incre-
mental processes, together with the inclusion
of tree-building actions in the model of con-
text, also allows a simple analysis of many cross-
speaker elliptical phenomena.
Fragments Bare fragments (3) may be anal-
ysed as taking a previous structure from con-
text as a starting point for parsing (or genera-
tion). WH-expressions are analysed as partic-
ular forms of metavariables, so parsing a wh-
question yields a type-complete but open for-
mula, which the term presented by a subsequent
fragment can update:
(3) A: What did you eat for breakfast?B: Porridge.
Parsing the fragment involves constructing an
un xed node, and merging it with the contex-
tually available structure, so characterising the
wellformedness/interpretation of fragment an-
swers to questions without any additional mech-
anisms: the term ( ; x; porridge0(x)) stands in a
licensed growth relation from the metavariable
WH provided by the lexical actions of what.
Functional questions (Ginzburg and Sag,
2000) with their fragment answers (4) pose
no problem. As the wh-question contains a
metavariable, the scope evaluation cannot be
completed; completion of structure and evalu-
ation of scope can then be e ected by merg-
ing in the term the answer provides, identifying
any introduced metavariable in this context (the
genitive imposes narrow scope of the introduced
epsilon term):
(4) A: Who did every student ignore?B: Their supervisor.
fSi < xg
f( ; x; stud0(x))g fg
fWH;}g fignr0g
! fSi < x; x < yg
f( ; x; stud0(x))g fg
f( ; y; sup0(x)(y)gfignr0g
VP Ellipsis Anaphoric devices such as pro-
nouns and VP ellipsis are analysed as decorating
tree nodes with metavariables licensing update
from context using either established terms, or,
for ellipsis, (lexical) tree-update actions. Strict
readings of VP ellipsis result from taking a suit-
able semantic formula directly from a tree node
in context: any node n 2 (T0 [ T) of suitable
type (e ! t) with no outstanding requirements.
Sloppy readings involve re-use of actions: any
sequence of actions (a1; a2; : : :; an) 2 (A0 [ A)
can be used (given the appropriate elliptical
trigger) to extend the current tree T if this pro-
vides a formula of type e ! t.13 This latter
approach, combined with the representation of
quanti ed elements as terms, allows a range of
phenomena, including those which are problem-
atic for other (abstraction-based) approaches
(for discussion see (Dalrymple et al., 1991)):
(5)
A: A policeman who arrested Bill read
him his rights.
B: The policeman who arrested Tom
did too.
The actions associated with A’s use of read
him his rights in (5) include the projection of a
metavariable associated with him, and its res-
olution to the term in context associated with
Bill. B’s ellipsis allows this action sequence to
be re-used, again projecting a metavariable and
resolving it, this time (given the new context) to
the term provided by parsing Tom. This leads
to a copy of Tom within the constructed predi-
cate, and a sloppy reading.
This analysis also applies to yield parallellism
e ects in scoping (Hirschb uhler, 1982; Shieber
et al., 1996), allowing narrow scope construal
for inde nites in subject position:
(6) A: A nurse interviewed every patient.B: An orderly did too.
Resolution of the underspeci cation in the
scope statement associated with an inde nite
can be performed at two points: either at the
immediate point of processing the lexical ac-
tions, or at the later point of compiling the re-
sulting node’s interpretation within the emer-
gent tree.14 In (6), narrow scope can be as-
signed to the subject in A’s utterance via this
late assignment of scope; at this late point in the
13In its re-use of actions provided by context, this ap-
proach to ellipsis is essentially similar to the glue lan-
guage approach (see (Asudeh and Crouch, 2002) and
papers in (Dalrymple, 1999) but, given the lack of in-
dependent syntax /semantics vocabularies, the need for
an intermediate mapping language is removed.
14This pattern parallels expletive pronouns which
equally allow a delayed update (Cann, 2003).
parse process, the term constructed from the ob-
ject node will have been entered into the set of
scope statements, allowing the subject node to
be dependent on the following quanti ed expres-
sion. The elliptical word did in B’s utterance
will then license re-use of these late actions, re-
peating the procedures used in interpreting A’s
antecedent and so determining scope of the new
subject relative to the object.
Again, these analyses are possible because
parsing and generation processes share incre-
mentally built structures and contextual pars-
ing actions, with this being ensured by the in-
crementality of the grammar formalism itself.
6 Alignment & Routinization
The parsing and generation processes must both
search the lexicon for suitable entries at ev-
ery step (i.e. when parsing or generating each
word). For generation in particular, this is a
computationally expensive process in principle:
every possible word/action pair must be tested {
the current partial tree extended and the result
checked for goal tree subsumption. As proposed
by O&P (though without formal de nitions or
implementation) our model of context now al-
lows a strategy for minimising this e ort: as
it includes previously used words and actions,
a subset of such actions can be re-used in ex-
tending the current tree, avoiding full lexical
search altogether. High frequency of elliptical
constructions is therefore expected, as ellipsis
licenses such re-use; the same can be said for
pronouns, as long as they (and their correspond-
ing actions) are assumed to be pre-activated or
otherwise readily available from the lexicon.
As suggested by O&P, this can now lead di-
rectly to a model of alignment phenomena, char-
acterisable as follows. For the generator, if there
is some action a 2 (A0 [A) suitable for extend-
ing the current tree, a can be re-used, generat-
ing the word w which occupies the correspond-
ing position in the sequence W0 or W. This re-
sults in lexical alignment { repeating w rather
than choosing an alternative word from the lex-
icon. Alignment of syntactic structure (e.g. pre-
serving double-object or full PP forms in the use
of a verb such as give rather than shifting to
the semantically equivalent form (Branigan et
al., 2000)) also follows in virtue of the procedu-
ral action-based speci cation of lexical content.
A word such as give has two possible lexical
actions a0 and a00 despite semantic equivalence
of output, corresponding to the two alternative
forms. A previous use will cause either a0 or a00
to be present in (A0 [ A); re-use of this action
will cause the same form to be repeated.15
A similar de nition holds for the parser: for a
word w presented as input, if w 2 (W0[W) then
the corresponding action a in the sequence A0 or
A can be used without consulting the lexicon.
Words will therefore be interpreted as having
the same sense or reference as before, modelling
the semantic alignment described by (Garrod
and Anderson, 1987). These characterisations
can also be extended to sequences of words {
a sub-sequence (a1; a2; : : :; an) 2 (A0 [ A) can
be re-used by a generator, producing the cor-
responding word sequence (w1; w2; : : :; wn) 2
(W0 [ W); and similarly the sub-sequence of
words (w1; w2; : : :; wn) 2 (W0 [ W) will cause
the parser to use the corresponding action se-
quence (a1; a2; : : :; an) 2 (A0 [ A). This will
result in sequences or phrases being repeatedly
associated by both parser and generator with
the same sense or reference, leading to what
Pickering and Garrod (2004) call routinization
(construction and re-use of word sequences with
consistent meanings).
It is notable that these various patterns of
alignment, said by Pickering and Garrod (2004)
to be alignment across di erent levels, are ex-
pressible without invoking distinct levels of syn-
tactic or lexical structure, since context, content
and lexical actions are all de ned in terms of the
same tree con gurations.
7 Summary
The inherent left-to-right incrementality and
monotonicity of DS as a grammar formalism al-
lows both parsing and generation processes to
be not only incremental but closely coupled,
sharing structures and context. This enables
shared utterances, cross-speaker elliptical phe-
nomena and alignment to be modelled straight-
forwardly. A prototype system has been im-
plemented in Prolog which re ects the model
given here, demonstrating shared utterances
and alignment phenomena in simple dialogue
sequences. The signi cance of this direct re-
 ection of psycholinguistic data is to buttress
the DS claim that the strictly serial incremen-
tality of processing is not merely essential to
the modelling of natural-language parsing, but
15Most frameworks would have to re ect this via pref-
erences de ned over syntactic rules or parallelisms with
syntactic trees in context, both problematic.
to the design of the underlying grammar formal-
ism itself.
Acknowledgements
This paper is an extension of joint work on the
DS framework with Wilfried Meyer-Viol, on ex-
pletives and on de ning a context-dependent
formalism with Ronnie Cann, and on DS gen-
eration with Masayuki Otsuka. Each has pro-
vided ideas and input without which the cur-
rent results would have di ered, although any
mistakes here are ours. Thanks are also due to
the ACL reviewers. This work was supported
by the ESRC (RES-000-22-0355) and (for the
second author) by the Leverhulme Trust.

References
A. Asudeh and R. Crouch. 2002. Derivational
parallelism and ellipsis parallelism. In Pro-
ceedings of WCCFL 21.
P. Blackburn and W. Meyer-Viol. 1994. Lin-
guistics, logic and  nite trees. Bulletin of the
IGPL, 2:3{31.
H. Branigan, M. Pickering, and A. Cleland.
2000. Syntactic co-ordination in dialogue.
Cognition, 75:13{25.
R. Cann. 2003. Semantic underspeci cation
and the interpretation of copular clauses in
English. In Where Semantics Meets Pragmat-
ics. University of Michigan.
H. H. Clark and D. Wilkes-Gibbs. 1986. Re-
ferring as a collaborative process. Cognition,
22:1{39.
M. Crocker, M. Pickering, and C. Clifton, ed-
itors. 2000. Architectures and Mechanisms
in Sentence Comprehension. Cambridge Uni-
versity Press.
M. Dalrymple, S. Shieber, and F. Pereira. 1991.
Ellipsis and higher-order uni cation. Linguis-
tics and Philosophy, 14(4):399{452.
M. Dalrymple, editor. 1999. Syntax and Se-
mantics in Lexical Functional Grammar: The
Resource-Logic Approach. MIT Press.
V. Ferreira. 1996. Is it better to give than to
donate? Syntactic  exibility in language pro-
duction. Journal of Memory and Language,
35:724{755.
S. Garrod and A. Anderson. 1987. Saying what
you mean in dialogue. Cognition, 27:181{218.
J. Ginzburg and I. A. Sag. 2000. Interrogative
Investigations. CSLI Publications.
R. Hausser. 1989. Computation of Language.
Springer-Verlag.
P. Hirschb uhler. 1982. VP deletion and across-
the-board quanti er scope. In Proceedings of
NELS 12.
A. Joshi and S. Kulick. 1997. Partial proof trees
as building blocks for a categorial grammar.
Linguistics and Philosophy, 20:637{667.
R. Kaplan and A. Zaenen. 1989. Long-
distance dependencies, constituent structure,
and functional uncertainty. In M. Baltin and
A. Kroch, editors, Alternative Conceptions of
Phrase Structure, pages 17{42. University of
Chicago Press.
R. Kempson, W. Meyer-Viol, and D. Gabbay.
2001. Dynamic Syntax: The Flow of Lan-
guage Understanding. Blackwell.
R. Kibble, W. Meyer-Viol, D. Gabbay, and
R. Kempson. 2001. Epsilon terms: a la-
belled deduction account. In H. Bunt and
R. Muskens, editors, Computing Meaning.
Kluwer Academic Publishers.
M. Marcus. 1987. Deterministic parsing and
description theory. In P. Whitelock et al., ed-
itor, Linguistic Theory and Computer Appli-
cations, pages 69{112. Academic Press.
W. Meyer-Viol. 1995. Instantial Logic. Ph.D.
thesis, University of Utrecht.
R. Millikan. 2004. The Varieties of Meaning.
MIT Press.
G. Neumann. 1994. A Uniform Computa-
tional Model for Natural Language Parsing
and Generation. Ph.D. thesis, Universit at des
Saarlandes.
M. Otsuka and M. Purver. 2003. Incremental
generation by incremental parsing. In Pro-
ceedings of the 6th CLUK Colloquium.
M. Pickering and S. Garrod. 2004. Toward a
mechanistic psychology of dialogue. Behav-
ioral and Brain Sciences, forthcoming.
M. Poesio and H. Rieser. 2003. Coordination in
a PTT approach to dialogue. In Proceedings
of the 7th Workshop on the Semantics and
Pragmatics of Dialogue (DiaBruck).
M. Purver and M. Otsuka. 2003. Incremental
generation by incremental parsing: Tactical
generation in Dynamic Syntax. In Proceed-
ings of the 9th European Workshop in Natural
Language Generation (ENLG-2003).
S. Shieber, F. Pereira, and M. Dalrymple. 1996.
Interactions of scope and ellipsis. Linguistics
and Philosophy, 19:527{552.
M. Stone and C. Doran. 1997. Sentence plan-
ning as description using tree-adjoining gram-
mar. In Proceedings of the 35th Annual Meet-
ing of the ACL, pages 198{205.
