Gapping and Frame 
A fresh look from a cognitive 
Semantics: 
perspective 
Andreas Stolcke 
Computer Science Division 
University of California, Berkeley 
and 
International Computer Science Institute 
1947 Center Street, Suite 600 
Berkeley, CA 94704, U.S.A. 
stolcke@icsi.berkeley.edu 
1 Intro'duction 
The topic of this paper is one of the pet sub- 
jects of trar~gformational syntax--gapping. Sev- 
ern 'exa~ffple's belOw will show that gapping is 
actually affected 'by' a, combination of syntactic, 
semantic and pragmatic factors, thus being far 
from a purely syntactic phenomenon. By look- 
ing at some basic properties of gapping from the 
point of view of interpretation and cognitive pro- 
,cessing, it is possible to arrive at a global princi- 
ple tha t can provide an integrated approach for 
these Various intluences on. gapping. The princi- 
ple proposed has the added advantage of being 
ndt justian arbitrary formal rule but being mo- 
tivated by eogni{ivh cOnsidei'ations. Thus, tile 
view taken here is compat!ble with tire basic 
tenets o\[ Cognitive' !,inguist,ics. ' 
2 Gapping and Interpretation 
2.1 Some examples 
Gapping constructions ifivolve two or more co- 
ordinated.clauses where the verb is missing in 
all but one clause, such as 
(1) Bill is reading a book, John (D a newspaper, 
and Kathy @ her favorite cartoon. 
Here I use '0 ' to indicate the position of the gap, 
i.e. the position the missing constituent would 
presumably have taken inthe clause if it had 
been spelled out in, full. This notation is merely 
a convenience and should not be taken to imply 
that some sort of empty constituent is present. 
Also, for historical reasons, I will refer to the 
constituents 'left-over' in the gapping clause as 
~rrmants, without implying that some sort of 
deletion has actually taken place. 
The more interesting cases of gapping are 
those where the verb is missing along with other 
constituents. In these cases the @ -notation is es- 
sential to disambigu~te the intended reading. 
(2) (a) Bill gave John the book and ~) (D Kathy 
the newspaper 
(b) ?Bill gave John the book and Kathy ~) 
the newspaper 
(c) *Bill gave John the book and i(aLhy 
the newspaper {? 
This example also shows that, in general, 
'grammaticality' or 'acceptability' is defined 
only relative to an intended interpretation. 1 
Thus, whereas sentence/interpretation (2a.) is 
fine, most speakers don't get (2c) and accc.pt 
(2b) only when given a particular intonation, i.e. 
topic-focus structure. 
The examples above all illustrate for'uJar'd 9ap .. 
ping, i.e. the gapping clause follows the non- 
gapping clause. The converse, backward gap- 
ping, also occurs, but the question of what de- 
termines the direction of gapping will not be ad- 
dressed here. In fact, all the examples discussed 
inthis paper are forward gapping constructions 
from English and German where backward gap- 
ping is not nearly as varied and flexible as the 
forward variety. 
1Langa.cker (1985) makes the point that accept~tbility 
is a judgement about the appropriateness of a pairing 
bel, ween a phonological form and the conceptualiz~Lt.ion 
it expresses, rather than some purely formal property of 
the phonological form per se, 
341 
The main reason for focusing on forward gap- 
ping at this point is that we want to consider 
gapping phenomena from the point of view of 
the cognitive processing demands they pose for 
the speaker/hearer. As will be discussed in sec- 
tion 2.2, it seems safe to make some very ba- 
sic assumptions about processing in the case of 
forward gapping, so that its implications can be 
studied without the burden of wildly speculative 
premisses fl'om a still largely unknown domain. 
Most theories to date have tried to account 
for gapping on the basis of purely formal aspects 
of syntax alone (see van Oirsouw (1987) for an 
overview), with the notable exception of Kuno 
(1976), who explicitly considers functional and 
perceptual aspects. Such approaches, mostly 
in a transformational framework, try to explain 
gapping as the result of some form of 'deletion 
under identity.' tlence these accounts are unable 
to consider questions such as why (2b) becomes 
possible only when given the appropriate focus 
structure (stressing Bill, the book, and Kathg, 
the newspaper). In these frameworks focus is 
considered a matter of pragmatics and simply 
doesn't figure in a syntax-based account. Sec- 
tion a.3 will show that the effects of focus on 
gapping are a natural consequence of the ap- 
proach taken here. 
Similarly, semantic constraints and contextual 
factors may have an inttuence on the interpreta- 
tion of gapping clauses. Compare the sentences 
in (2) above to those in (3). 
(3) Mary got a lot of nice clothes on her six- 
teenth birthday. 
(a) IIer parents gave her an expensive 
evening gown and her godmother ~ 
a fine pair of shoes. 
(b) ?tier parents gave her an expensive 
evening gown and ~ @ her godmother 
a fine pair of shoes. 
(2a) and (3a), and (2b) and (3b), respectively, 
have identical phrase structure, yet the preferred 
interpretations of the gapping clauses are re- 
versed. Some of the ways in which semantics can 
influence gapping are discussed in section 3.4. 
2.2 I'¥ame Semantics 
It is natural that researchers try to confine lin- 
guistic phenomena to subtheories (such as syn- 
tax), feeling that not doing so would inevitably 
342 
face them with an overwhelming range of data 
for which no adequate principles and rules are 
available. 
Cognitive Linguistics proposes that all levels 
of linguistic processing are embedded in the gen- 
eral human cognitive apparatus and that, there- 
fore, linguistic generalizations can only be de- 
rived by studying language as the product of 
cognition rather than of some autonomous for- 
mal system (Langacker, 1985; Lakoff, 1986). If 
this view is correct then the study of processing 
constraints and applicable cognitive principles 
(e.g., of categorization) should provide us with 
a handle on some of the linguistics phenomena 
that apparently span a variety levels. 
Of course nothing is won if we have to re- 
sort to uncertain or unjustified claims about the 
cognitive domain itself, or have no certainty of 
its relation to the linguistic problems in ques- 
tion. Fortunately, in the special case of gapping 
as discussed here, the assumptions we have to 
make about underlying semantics and process- 
ing structures are minimal. I will adopt an infor- 
mal version of frame semantics, along the lines 
of Filhnore (1982). In this view, the main verb 
of a clause determines a set of roles to be filled by 
the (meaning of) remaining constituents. Thus, 
interpreting the first clause of (2a) resuIts in a 
structure of the form 
give 
agent : Bill 
(4) patient : book 
recipient : John 
It is a matter of debate how 'deep' the roles or 
cases making np the frame representation should 
be. 15br our purposes it is not crucial whether 
one prefers several levels of representation, such 
as grammatical roles vs. semantic roles, or gen- 
eral roles (like agent) vs. frame-specific roles 
(like giver). The only assumption here is that 
interpretation requires a process equivalent to 
the assignment of constituent meanings to slots 
in the frame, and that this assignment has to be 
derived based on whatever linguistic and con- 
textual clues are provided. 
In the case of gapping, then, several such 
frames are obtained by partly retaining slot 
fillers from previous instances of the same type 
of frame (i.e. generated by the same verb). Thus 
a sequence of multiple frames with shared fillers 
can be linguistically realized in a single, compact 
construction where multiple fillers are expressed 
only once. In the case of example (2a) above, 
the second, gapping clause refills the 'patient' 
and 'recipient' slots of the give frame, while re- 
taining the 'agent' slot. 
In non-gapping sentences a combination of 
word order, case marking, prepositional mark- 
ers, selection restrictions, etc. is usually more 
than sufficient to assign fillers unambiguously to 
frame slots. In gapping constructions the role 
assignment problem becomes potentially more 
difficult since some of this information is lost, al- 
though in many cases the loss is marginal. Thus, 
in example (1) interpretation of the second 
clause is unproblematic since two constituents 
have to fill two slots, and the usual rules of con- 
stituent order in English unambiguously deter- 
mine the assignment. In (2), on the other hand, 
there are three potential role assignments con- 
sistent with the constituent order given. 
The problem, then, is to identify the syntac- 
tic, semantic and pragmatic factors that ren- 
der certain gapping structures (and their corre- 
sponding interpretations) acceptable, while oth- 
ers axe ruled out (or at least made less felici- 
tous). 
3 A Conjecture 
We can now state a tentative principle regarding 
the relation between the intcrI)retation of gap- 
ping constructions and their acceptability. 
(5) Role Assignment Principle: A gapping 
construction with a specific interpretation 
will be acceptable only if it provides suffi- 
cient clues for the role assignment implied 
by that interpretation. 
This principle is similar in spirit to tile R,e- 
coverability Constraint proposed by I\[ankamer 
(1973), although it is far more generM as it po- 
tentially encompasses all kinds of linguistic phe- 
nomena that play a role in interpretation, not 
just syntax. 2 
Note that (5) does not claim that in- 
terpretability alone will render a sentence 
acceptable, a To prevent the conjecture from be- 
2IIankamer's principle was concerned with the recov- 
ery of 'deep structure' after having undergone deletions 
and movement. 
3If someone used apple green as a noun phrase refer- 
ring to a green apple we will in general be able to make 
sense of (i.e., interpret) that description while still recog- 
nizing it as ungramnlatical. 
ing vacuous two points have to be addressed. 
First, we have to show that there is a systematic 
repertoire of linguistic devices available for the 
interpretation of gapping constructions. Sec- 
ondly, it must be shown that these devices sys- 
tematically affect the acceptability of gapping 
sentences in accordance with the principle. 
In many cases gapping constructions will not 
be strictly 'good' or 'bad', but felicitous to a cer- 
tain degree, or acceptable to only some fraction 
of speakers. This is consistent with the principle 
stated above, since (5) can easily be rephrased 
in terms of degrees of acceptability: 
(6) Role Assignment Principle (graded 
version): A gapping construction with a 
specific interpretation will tend to become 
more acceptable as more clues for the in- 
tended role assignment are made available. 
Note also that this formulation effectively 
characterizes interpretation of gapping construc- 
tions as an evidential cognitive phenomenon, 
i.e. one where multiple sources of evidence con- 
tribute to a perception or interpretation in a cu- 
mulative manner. 
3.1 Syntactic clues 
Despite the definite non-syntactic influences on 
gapping, syntax (together with the rules govern- 
ing its relation to semantics) is the major source 
of information for interpretation, even when the 
syntactic structure is 'incomplete'. Clues such 
as linear order, case marking, and prepositions 
are still potentially available for the remnant 
constituents. 
Van Oirsouw (1987) has recently suggested 
a theory in terms of constituent deletion un- 
der identity which probably covers as much 
ground as is possibl e in a purely syntax-oriented 
framework. 4 Interestingly, he was able to state 
some surprisingly far-reaching generalizations 
by ignoring all fine points of phrase structure 
and simply considering the top-level ordering of 
constituents. 
In a similar fashion, it seems that for forward 
gapping, the following simple rule captures a 
considerable range of cases. It is a straightfor- 
ward variant of the MinimM Distance Principle 
proposed by Kuno (1976). 
4His generalizations can be rephrased without neces- 
sarily assuming deletion transformations. 
343 
(7) Principle of Surface Proximity: A rem- 
nant constituent tends to fill a role whose 
previous filler is close in surface constituent 
order. (Fillers of the same role tend to be 
(:lose.) 
Consider the following series of examples which 
have increasing amounts of material intervening 
between the two filler constituents. 
(8) (a) John gave Peter a bagel and a banana 
(b) :John gave Peter a bagel yesterday, 
and a banana 
(c) :?John gave Peter a bagel yesterday 
before lunch, and a banana 
(d) :::John gave Peter a bagel yesterday 
before lunch after he had begged him, 
and a banana 
(9) (a) John lent Peter some money, and 0 0 
0 some clothes 
(b) ?John lent his brother some money, 
and 0 0 his girlfriend 0 
(c) ?John lent me some money, and Paul 
000 
Again sentences marked '?' require appropriate 
contrastive intonation to become acceptable (cf. 
section a.a). 
Note that (7) is not arbitrary; it makes sense 
from a cognitive point of view. If we assume that 
roles are sequentially processed and have to be 
retrieved when refilled, it seems natural that the 
retrieval tends to become more difficult as the 
temporal distance from the previous processing 
event increases. 
At a more abstract level, syntagmatic prox- 
imity can be viewed as symbolizing conceptual 
relatedness (Langacker 1985). Hence the frame 
unit at the conceptual level corresponds (proto- 
typically) to a clause unit at the syntactic level. 
Gapping constructions relate not just the fillers 
of the roles in a single frame but also the var- 
ious fillers of the same role in different frames. 
Therefore we would expect those fillers to be 
syntagmatically close to the extent that the two 
relational aspects can be accommodated simul- 
taneously. 
Note that the case of simple constituent coor- 
dination can be seen as the optimal solution to 
multiple frame realization as far as (7) is con- 
cerned, and is generally preferred. 
344 
(d) John and Paul lent me some money 
It represents the case where clause unity has 
been traded off for for the conceptualized unity 
of the two fillers of the same slot. (Accordingly 
we get a reading where John and Paul act as a 
single 'financial entity' only in (9d), but not in 
the corresponding gapping construction (9c).) 
In (8a), on the other hand, we have a case 
where both clause unity and filler proximity can 
be integrated perfectly. Note that the view 
presented here obliterates the question whether 
(8a) is the result of constituent coordination or 
a case of gapping. A formally oriented theory 
might be troubled by a syntactic construction 
that can be 'redundantly' accounted for by two 
different rules or principles. For a cognitively 
oriented account such a coincidence presents no 
contradiction. 5 
One might object that (7) makes the wrong 
predictions when comparing cases with one rem- 
nant to those with multiple remnants. 
(10) (a) John offered me money and 0 0 0 good 
advice 
(b) John offered me money, and Paul 0 0 
good advice 
Clearly the additional remnant in (10b) in- 
creases the distance between the two fillers of 
the recipient role of offer, yet both sentences 
in (10) seem equally unproblematic. However, 
note that as soon as more than one remnant 
is present their role assignments mutually con- 
strain each other, thus effectively making more 
5Van Oirsouw (1987) has tried to develop a unified 
rule of deletion in coordinations, covering cases otherwise 
treated as separate rules, such as gapping, conjunction re- 
duction, right-node raising, VP deletion, etc. One of the 
motivations for such an attempt is that these rules have 
a certain amount of overlap, i.e. there are constructions 
which fall into the scope of more than one rule, yet no 
single rule covers all cases. This is considered undesirable 
from a metatheoretical point of view since it apparently 
introduces ambiguity and redundancy. 
A different view may be derived frmn the notion of 
radial categories as it arises in human categorization be- 
havior (Lakoff, 1986). The sentences to be accounted 
for are apparently clustered around a small number of 
constructions (or rules). These constructions, however, 
are not completely unrelated; rather they are linked to 
each other by those special cases where more than one 
rule overlap. Thus the overlap between rules is not un- 
desirable but cognitively advantageous since it allows the 
complete set of constructions to be perceived as a unified 
phenomenon. 
information available to the interpretation pro- 
cess. (()lie fundamental constraint that comes 
into play is that every filler can fill only one 
role. Furthermore linear order information is 
now available.) 
3.2 Case marking and prepositions 
In languages that case-mark their verb argu- 
ments we would expect this device to provide 
very strong (in fact, mandatory) clues for role 
assignment. This can be verified in German, 
where noun phrases are case-marked, except for 
those consisting only of single proper nouns. 
(11) (a) Ich sah, wie der Englitnder den Fran- 
zosen begriigte und der Deutsche 0 (3 
I saw how the Englishman-NOM the 
Frenchman-ACC welcomed and the 
GermanCNOM 13 13 
(b) ?:Ich sah, wie Peter Paul begriil3te und 
Hans 13 (3 
I saw how Peter Paul welcomed and 
Hans 13 0 
(c) Ich sah, wie Peter Paul begriiflte und 
(3 Hans I3 
I saw how Peter Paul welcomed and 0 
Hans 13 
(11a) and (llb) have identical constituent struc- 
ture (up to the noun phrase level) and are both 
non-optimal by proximity considerations. (1 la) 
can be made acceptable with appropriate con- 
trastive intonation, while (11b) remains awk- 
ward. Instead, the interpreta.tion corresponding 
to (11c) is easily obtained, again in accordance 
with proximity. 
In English, where case-marklng is generally 
not available, prepositions can play a similar 
role. 
(12) (a) :Bill visited Jim yesterday after din- 
ner, and 0 13 John 0 13 
(b) Bill talked to Jim yesterday after din- 
ner, and {3 13 to John {3 
Again, both examples are awkward due to the 
great distance between successive role fillers, but 
in (12b) the prepositional marker to gives a 
strong clue for the intended attachment of John 
thus enhancing acceptability. 
3.3 Topic-Focus structure 
In most of the cases seen so far a typical in- 
tonation pattern will naturally accompany the 
gapping construction, focusing on and contrast- 
ing the various fillers of same frame slot. In fact, 
as the distance between remnants and previous 
filler constituents increases, the interpretation 
has to be supported by increasing amounts of 
stress. Note that such a trade-off between syn- 
tactic proximity and focus structure makes sense 
in view of the evidential character of gapping 
proposed in (6). 
(13) (a) Linda returned the textbook to the lio 
brary, but not (3 (3 0 to the one it be- 
longs to. 
(b) Linda returned the textbook to the li- 
brary, but not 13 13 the lecture tape 13 
(c) Linda returned the textbook to the li- 
brary, but not the other students in 
the class 0 0 
(Italics indicate intonation centers.) 
This behavior fits well into the general prin- 
ciple of role assignment proposed. In terms of 
our frame semantics account, focus ett~ctively 
'foregrounds' a subset of the roles implying that 
those are the ones likely to be refilled. 
There is evidence that the converse process 
works as well, i.e. that circumstances which 
'background' certain roles provide support for 
gapping structures where those roles are not re- 
filled. It seems that this is at least part of what 
happens in (3). The preceding sentence sets up 
Mary (i.e. her) as a topic, relative to which the 
other roles are foregrounded. 
Sgall et al. (1986) have argued for incorpo- 
rating descriptions of topic-focus structures into 
linguistic representations (rather than throwing 
them into the 'pragmatic wastebasket'). Clearly 
more work on the relation between syntactic 
configuration, focus and gapping is needed. To 
get an idea of the possibihties Mong these lines 
consider the following: Let us assume, following 
Sgall et al., that in the absence of special into- 
nation in English declarative sentences elements 
towards the end of the sentence are increasingly 
'focussed.' Then the gapping cases in (8) and 
(9) might be subsumed by focus structure: Re- 
filled roles tend to occur towards the end of the 
clause because that is where default focus is lo- 
cated. It would be jumping to conclusion, how- 
ever, to say that all gapping phenomena can be 
345 
accounted for in term of focus structure, given 
the other cases discussed here. 
3.4 Semantic constraints 
From the data presented so far we would ex- 
pect any type of constraint which could poten- 
tially guide role assignment to have an impact 
on gapping. Therefore, semantic compatibility 
between the verb semantics and its arguments 
should make a difference in gapping contexts. 
This is illustrated in the following example pairs. 
(14) (a)John likes ice-cream, but not ~ 0 
chocolate 
gapping, VP deletion, conjunction reduction, 
right-node raising) lend themselves to a similar 
analysis, hopefully using compatible principles. 
Still, gapping provides a striking case where 
a number of linguistic processes --traditionally 
perceived as operating at different levels and 
largely independently--seem to follow a general 
principle. More importantly, the principle can 
be understood as arising from cognitive con- 
straints on sentence processing. Forward gap- 
ping is a lucky case in point because ~he role 
assignment problem can easily be identified as a 
major subtask in the interpretation, thus giving 
us a handle on a cognitive analysis. 
(b) *John likes ice-cream, but not choco- 
late 0 0 Acknowledgements 
(15) (a) ?John likes ice-cream, but not O 0 his 
little sister 
(b) John likes ice-cream, but not his little 
sister @ @ 
I would like to thank George Lakoff for intro- 
ducing me to the view of linguistics that led to 
this work, and many helpful cornments along the 
way. The author is supported by an IBM Grad- 
u.ate Fellowship. 
Note how (15a) is perceived as 'funny' because 
the verb semantics are not kept exactly identi- 
cal; rather a different sense (and frame) associ- 
ated with like is invoked in the second clausefi 
4 Conclusion 
The discussion so far leaves a number of im- 
portant problems. The syntactic constraints on 
gapping are obviously more complex than the 
simple rule proposed here, and similarly for the 
non-syntacticphenomena (semantic constraints, 
focus) I appealed to. 
New questions arise in the framework pro- 
posed here, regarding the precise nature of in- 
teractions between the processing of the various 
clues available. It would certainly be nice if some 
kind of simple ordering or hierarchy between the 
various factors could be established, although 
this may not be possible. The fact that gapping 
seems to be a phenomenon involving so many 
processes and influences also implies that the in- 
teractions between them may be inherently com- 
plex and not compactly describable. 
Another open question is to what extent other 
types of 'deletion transformations' (backward 
6The same humoristic effect is more pronounced in 
sentences such as "First the president cut the salami, 
then the budget". 

References 

Fillmore, C. (1982). Frame Semantics. In Lin- 
guistic Society of Korea, ed., Linguistics in 
the Morning Calm. Hanshin, Seoul. 

Hankamer, J. (1973). Unacceptable Ambiguity. 
Linguistic Inquiry, 4:17-68. 

Kuno, S. (1976). Gapping: A Functional Anal- 
ysis. Linguistic fnquiry, 7:300-318. 

Lakoff, G. (1986). Women, Fire and Danger- 
ous Things. University of Chicago Press, 
Chicago. 

Langacker, R. (1985). Foundations of Cogni- 
tive Grammar, Vol. 1. Stanford University 
Press, Stanford. 

Sgall, P., Haji£ov£, E., & Panevov£, J. (1986). 
The Meaning of the Sentence in Its Se- 
mantic and Pragmatic Aspects. Academia, 
Prague. 

van Oirsouw, R. R. (1987): The Syntax o/Coor- 
dination. Croom Hehn, London, New York. 
