ANAPHORA RESOLUTION IN SLOT GRAMMAR 
Shalom Lappin and Michael McCord 
IBM T. J. Watson Research Center 
P.O.B. 704 
Yorktown Heights, NY 10598 
We present three algorithms for resolving anaphora in Slot Grammar: (1) an algorithm for interpreting 
elliptical VPs in antecedent-contained deletion structures, subdeletion constructions, and intersentential 
cases; (2) a syntactic filter on pronominal coreference; and (3) an algorithm for identifying the binder of an 
anaphor (reflexive pronoun or the reciprocal phrase "each other"). These algorithms operate on the output of 
a Slot Grammar parser, and, like the parser, they run in Prolog. The VP anaphora algorithm implements an 
S-structure analysis of VP ellipsis that we argue provides a more unified and empirically motivated treatment 
of VP anaphora resolution than analyses that attempt to interpret elliptical VPs at a level of logical form. 
Each algorithm can operate independently of the others, and we have incorporated each into an integrated 
anaphora resolution component. The interpreted elliptical VP structures that the VP anaphora algorithm 
produces provide the input to the two NP anaphora resolution procedures. The integrated anaphora 
resolution component provides a powerful syntactically driven module for generating partially interpreted 
representations that can serve as input to semantic and discourse interpretation systems. 
1 INTRODUCTION 
In this paper 1 we present algorithms for handling three 
different sorts of anaphora within Slot Grammar (McCord 
1980, 1989b, 1990). These algorithms are second-pass 
procedures that operate on the output of a Slot Grammar 
parser. The parser and the algorithms constituting the 
anaphora resolution component run in Prolog. In Section 2 
we present a brief overview of Slot Grammar and the parser 
that implements it. This section also includes a description 
of an alternative network representation of parser output 
on which the algorithms operate. In Section 3 we propose 
an analysis of VP anaphora that involves applying rules of 
interpretation directly to S-structure (parsed surface struc- 
ture) rather than to LF (logical form), as required by 
several recent accounts. We provide in Section 3.1 theoreti- 
cal motivation for preferring our analysis to an LF treat- 
ment. In Section 3.2 we present a schematic statement of 
the algorithm that implements this analysis in Slot Gram- 
mar, and illustrate the algorithm with examples of its 
output. Section 4 is devoted to a syntactic filter on pronom- 
inal anaphora that identifies noncoreferential NP-pronoun 
pairs within a sentence. A more detailed presentation of the 
filter algorithm is given in Lappin and McCord (1990). 
Section 5 contains a rule for locating possible NP anteced- 
ents for anaphors (reflexive pronouns and reciprocals). The 
conjunction of the latter two algorithms has roughly the 
same extension as Chomsky's (1981, 1986a) binding the- 
ory. However, while the conditions of the binding theory 
are stated in terms of the configurational relation of c-com- 
mand, the coreference filter and anaphor binding algorithm 
employ the head-complement structures defined by Slot 
Grammar. 2 
The three algorithms that make up the anaphora resolu- 
tion component of Slot Grammar are fully modular in that 
they apply independently of each other. Any two algo- 
rithms in this set can be conjoined. Moreover, both the 
pronominal noncoreference filter and anaphor binding algo- 
rithms have been combined with the VP anaphora algo- 
rithm to construct an integrated system of anaphora resolu- 
tion in which the two NP anaphora rules apply to the 
results of VP anaphora interpretation) In Section 6 we 
illustrate the operation of the integrated system with exam- 
ples of the representations it generates. 
2 SLOT GRAMMAR 
The original work on Slot Grammar was done around 
1976-1978 and appeared in McCord (1980). Recently, a 
new version (McCord 1989b, 1990) was developed in a 
logic programming framework, in connection with the ma- 
chine translation system LMT (McCord 1989a, 1989c, 
1989d). 
Slot Grammar is lexicalist and is dependency oriented. 
Every phrase has a head word (with a given word sense and 
morphosyntactic features). The constituents of a phrase 
besides the head word, also called the modifiers of the head, 
are obtained by "filling" slots associated with the head. 
Slots are symbols like subi, obi, and iobi representing 
grammatical relations, and are associated with a word 
Computational Linguistics Volume 16, Number 4, December 1990 197 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
(sense) in two ways. The lexical entry for the word specifies 
a set of complement slots, corresponding to logical argu- 
ments of the word sense, and the grammar specifies a set of 
adjunct slots for each part of speech. 4 A complement slot 
can be filled at most once, and an adjunct slot can by 
default be filled any number of times. 
The phenomena treated by augmented phrase structure 
rules in some grammatical systems are treated modularly 
by several different types of rules in Slot Grammar. The 
most important type of rule is the (slot) filler rule, which 
gives conditions (expressed largely through unification) on 
the filler phrase and its relations to the higher phrase. 
Filler rules are stated (normally) without reference to 
conditions on order among constituents. But there are 
separately stated ordering rules.5 Slot~head ordering rules 
state conditions on the position (left or right) of the slot 
(filler) relative to the head word. Slot~slot ordering rules 
place conditions on the relative left-to-right order of (the 
fillers of) two slots. 
A slot is obligatory (not optional) if it must be filled, 
either in the current phrase or in a raised position through 
left movement or coordination. Adjunct slots are always 
optional. Complement slots are optional by default, but 
they may be specified to be obligatory in a particular lexical 
entry, or they may be so specified in the grammar by 
obligatory slot rules. Such rules may be unconditional or 
be conditional on the characteristics of the higher phrase. 
They also may specify that a slot is obligatory relative to 
the filling of another slot. For example, the direct object 
slot in English may be declared obligatory on the condition 
that the indirect object slot is filled by a noun phrase. 
One aim of Slot Grammar is to develop a powerful 
language-independent module, a "shell," which can be 
used together with language-dependent modules, reducing 
the effort of writing grammars for new languages. The Slot 
Grammar shell module includes the parser, which is a 
bottom-up chart parser. It also includes most of the treat- 
ment of coordination, unbounded dependencies, controlled 
subjects, and punctuation. And the shell contains a system 
for evaluating parses, extending Heidorn's (1982) parse 
metric. The Slot Grammar evaluator is used not only for 
ranking final parses, as with Heidorn's, but also for pruning 
away unlikely partial analyses during parsing, thus reduc- 
ing the problem of parse space explosion. Parse evaluation 
expresses preferences for close attachment, for choice of 
complements over adjuncts, and for parallelism in coordina- 
tion. 
Although the shell contains most of the treatment of the 
above phenomena (coordination, etc.), a small part of their 
treatment is necessarily language dependent. A (language- 
specific) grammar can include for instance (1) rules for 
coordinating feature structures that override the defaults in 
the shell; (2) declarations of slots (called extraposer slots) 
that allow left extraposition of other slots out of their fillers; 
(3) language-specific rules for punctuation that override 
defaults; and (4) language-specific controls over parse eval- 
uation that override defaults. 
Currently, Slot Grammars are being developed for En- 
glish, (ESG) by McCord, for Danish (DSG) by Arendse 
Bernth, and for German (GSG) by Ulrike Schwall. ESG 
uses two lexicons: (1) a hand-coded lexicon of about 3,700 
common words, and (2) the UDICT lexicon (Byrd 1983; 
Klavans and Wacholder 1989) having over 60,000 lemmas, 
with a heuristic interface that produces Slot Grammar- 
style entries. 
Our anaphora algorithms apply in a second pass to the 
parse output; the remainder of this section describes Slot 
Grammar syntactic analysis structures. 
A syntactic structure is a tree; each node of the tree 
represents a phrase in the sentence and has a unique head 
word. Formally, a phrase is represented by a term 
phrase(X,H, Senso,Features,glotFramo,E~,Mods), 
where the components are as follows. (I) X is a logical 
variable called the marker of the phrase. Unifications of 
the marker play a crucial role in the anaphora algorithms. 
(2) H is an integer representing the position of the head 
word of the phrase. This integer identifies the phrase 
uniquely, and is used in the anaphora algorithms as the way 
of referring to phrases. (3) Sense is the word sense of the 
head word. (4) Features is the feature structure of the head 
word and of the phrase. It is a logic term (not an attribute- 
value list), which is generally rather sparse in information, 
showing mainly the part of speech and inflectional features 
of the head word. (5) SlotFrame is the list of complement 
slots, each slot being in the internal form slot(Slot,Ob,X), 
where Slot is the slot name, Ob shows whether it is an 
obligatory form of Slot, and X is the slot marker. The slot 
marker is unified (essentially) with the marker of the filler 
phrase when the slot is filled, even remotely, as in left 
movement or coordination. Such unifications are important 
for the anaphora algorithms. (6) Ext is the list of slots that 
have been extraposed or raised to the level of the current 
phrase. (7) The last component Mods represents the modi- 
fiers (daughters) of the phrase, and is of the form 
mods(LMods,RMods) where LMods and RMods are the 
lists of left modifiers and right modifiers, respectively. Each 
member of a modifier list is of the form Slot:Phrase where 
Slot is a slot and Phrase is a phrase that fills Slot. Modifier 
lists reflect surface order, and a given slot may appear more 
than once (if it is an adjunct). Thus modifier lists are not 
attribute-value lists. 
Figure I shows a sample parse produced by ESG for the 
sentence Who did John say wanted to try to find him? The 
tree is displayed by a procedure that uses only one line per 
node and exhibits tree structure lines on the left. In this 
display, each line (representing a node) shows (1) the tree 
connection lines, (2) the slot filled by the node, (3) the word 
sense predication, and (4) the feature structure. The fea- 
ture structure is abbreviated here by a display option, 
showing only the part of speech. The word sense predica- 
tion consists of the sense name of the head word with the 
198 Comimtational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
Who did John say wanted to try to find him? 
I 
subj(n) to 
sugj(n) auxcmp(binf) 
obj(fln) prelnf 
comp(inf) prelnf 
obj(inf) obj(fin) 
who(X2) . dol(Xi,X3,X4) 
John(X3) say(X4,X3,X9,u) . 
want(Xg,X2sX2,Xl2) preinf(Xl2) . 
hry(Xl2,X2sXl3) prelnf(Xl3) 
find(Xl3,X2,Xl4,u,u) 
he(X14) 
noun verb 
noun verb 
verb preinf 
verb preinf 
verb noun 
Figure I 
following arguments. The first argument is the marker 
variable for the phrase (node) itself; it is like an event or 
state variable for verbs. The remaining arguments are the 
marker variables of the slots in the complement slot frame 
(u signifies "unbound"). As can be seen in the display, the 
complement arguments are unified with the marker vari- 
ables of the filler complement phrases. Note that in the 
example the marker X2 of the 'who' phrase is unified with 
the subject variables of "want," "try," and "find." (There 
are also some unifications created by adjunct slot filling, 
which will not be described here.) 
For the operation of our anaphora algorithms, there is a 
preliminary step in which pertinent information about the 
parse tree is represented in a more convenient way for the 
algorithms. As indicated above, nodes (phrases) themselves 
are represented by the word numbers of their head words. 
Properties of phrases and relations between them are repre- 
sented by unit clauses (predications) involving these inte- 
gers (and other data), which are asserted into the Prolog 
workspace. Because of this "dispersed" representation with 
a collection of unit clauses, the original phrase structure 
for the whole tree is first grounded (variables are bound to 
unique constants) before the unit clauses are created. 
As an example of this clausal representation, the clause 
hasarg(P,X) says that phrase P has X as one of its argu- 
ments; i.e., X is the slot marker variable for one of the 
complement slots of P. For the above sample parse, then, 
we would get clauses 
hasarg(S,'X2'), hasarg(5,'X12'). 
as information about the 'want' node (5). 
As another example, the clause phmarker(P,X) is added 
when phrase P has marker X. Thus for the above sample, 
we would get the unit clause 
phmarker(1,'X2'). 
An important predicate for our algorithms is pharg, 
defined by 
pharg(P, Q) ~ phmarker(P,X) & hasarg(Q,X). 
This says that phrase P is an argument of phrase Q. This 
includes remote arguments and controlled subjects, be- 
cause of the unifications of marker variables performed by 
the Slot Grammar parser. Thus for the above parse, we 
would get 
pharg(l,5), pharg(1,7), pharq(1,9). 
showing that "who" is an argument of "want," "try," and 
"find." 
3 VP ANAPHORA 
3.1. THE RESOLUTION OF VP ANAPHORA AT 
S-STRUCTURE 
Before presenting our algorithm for the interpretation of 
VP anaphora structures, we will provide motivation for the 
general view of VP anaphora that the algorithm imple- 
ments. We characterize VP anaphora as a relation between 
the head V and selected arguments and adjuncts of a 
structured empty, or partially empty, elliptical VP, and the 
head A and corresponding adjuncts of an antecedent VP. 
This relation is identified on S-structure parse representa- 
tions. The VP anaphora interpretation procedure copies the 
head A of the antecedent VP into the position of the head of 
the elliptical VP, and specifies which arguments and ad- 
juncts of the antecedent A are inherited by the elliptical V. 
In this way, it provides an interpretation of the elliptical 
VP. It is important to recognize that this procedure oper- 
ates on S-structure representations rather than on a more 
abstract level of LF. 
Let us briefly consider the case for an LF-based ap- 
proach to VP anaphora resolution. The elliptical VP in the 
relative clause of the object NP in 1 is contained in the 
matrix VP, which is its antecedent. 
1. Dulles suspected everyone who Angelton did. 
As May (1985) observes, if we copy the matrix VP into the 
position of the empty VP at S-structure, an interpretive 
regress results. The empty VP will reappear in the copied 
matrix VP. May proposes to solve this problem by applying 
the operation of quantifier raising (QR) to the object NP in 
1. QR adjoins the quantified NP to the matrix sentence to 
derive the LF representation 2. 6 
Computational Linguistics Volume 16, Number 4, December 1990 199 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
2. \[W'\[NP, everyone who 1 Angelton did 
\[vp\] \] \[ip Dulles \[vp suspected h\]\] \] 
The matrix VP of the IP in 2 is assigned to the empty VP of 
the adjoined NP to obtain 3, the desired interpretation of 1. 
3. \[IP'\[NP, everyone who Angelton \[vp suspected h\]\] 
\[w Dulles \[vp suspected t l\] \] \] 
May concludes that antecedent-contained deletion (ACD) 
structures can only be interpreted by a VP copying rule 
that applies at LF. 
There are at least two serious difficulties with May's 
analysis of ACD structures. 7 First, as Haik (1987) points 
out, the wh-phrase in the relative clause of an ACD sen- 
tence such as 1 is constrained by subjacency. 
4a. John read everything which Bill believes he did. 
b. *John read everything which Bill believes the claim 
that he did. 
On May's analysis, the VP in the relative clause in 1 and 
4a-b is empty at S-structure, and the wh-phrase binds a 
trace only after VP copying has applied to the LF produced 
by the movement of the object NP. But it is generally 
agreed that subjacency is a condition that constrains opera- 
tor-trace binding chains only at S-structure. 8 Given May's 
account, there is no trace at S-structure for the wh-operator 
to bind in 1 and 4a-b. Therefore, it is unclear how, on this 
analysis, subjacency can constrain wh-movement in ACD 
structures. 
May (in press) seeks to avoid this problem by suggesting 
that subjacency does, in fact, apply at LF. The examples in 
5 indicate that this is not the case. 
5a. At least one critic reviewed Mary's biography of each 
author. 
b. Who did at least one critic review 
\[Npa biography of t\] 
c. *Who did at least one critic review 
\[NeMary's biography of t\] 
5a permits two scope readings for "each author" relative to 
"at least one critic." On the narrow scope reading, a single 
critic reviewed all of Mary's biographies. When "each 
author" receives wide scope, there is at least one (possibly 
different) critic for each of Mary's biographies of an au- 
thor. If we accept May's view that the scope of a quantified 
NP is, in part, defined in terms of the constituent to which it 
is adjoined by QR, the fact that "each author" can take 
wide scope relative to "at least one critic" indicates that 
QR can move the former NP out of the NP "Mary's 
biography of each author." But 5b-c shows that the latter 
NP defines a syntactic island for wh-movement. It seems, 
then, that subjacency does constrain binding chains visible 
at S-structure, but not scope assignment. 
The fact that antecedent-contained VP anaphora exhib- 
its subjacency effects strongly suggests that the elliptical 
VP in these cases is not necessarily empty at S-structure, 
but may contain a trace bound by a wh-phase (or other 
ope:rator). 
Second, May's analysis does not extend to the subdele- 
tion variety of ACD, where arguments and adjuncts of an 
empty verb are realized within the partially elliptical VP 
that it heads, as in the sentences in 6. 9 
6a. John writes more books than Bill does articles. 
b. The university gives more money to the library for 
books than the city does to the orchestra for instru- 
ments. 
c. The university gives more money to the library for 
periodicals than it does for books. 
d. John wrote more articles for the journal about politics 
than he did about linguistics. 
e. John showed everything to Mary which he did to Bill. 
f. Mary argues about politics with everyone who she 
does about linguistics. 
g. Mary arrived in London before Sam did in New 
York. 
h. John reviewed the play for The New York Times 
shortly after Bill did for The Washington Post. 
As these examples illustrate, subdeletion occurs in a 
variel:y of syntactic environments, including comparative 
NPs (6a-d), relative clauses (6e-f), and adverbial phrases 
(6g-h). Given that May's analysis treats VP anaphora as a 
global relation between an empty VP and an antecedent 
VP, :it does not cover subdeletion, where an anaphoric 
relation holds between the head and selected constituents of 
an elliptical VP and its antecedent. But the full and subde- 
letion varieties of ACD are closely related phenomena, and 
an analysis that provides a unified explanation of both types 
of V\]? anaphora is clearly preferable to an account that 
handles only one type of antecedent-contained anaphora. 
I:~. is possible to capture the properties of ACD structures 
that create problems for May's LF analysis if we assume 
that the empty VP of a VP anaphora environment is 
structured, and may contain arguments or adjuncts of the 
head. l° The arguments appearing in a partially empty 
elliptical VP can be lexically realized, or they may be traces 
(or their counterparts in Slot Grammar). If we apply this 
analysis to 1, we obtain 7a as its S-structure. 
7a. Dulles suspected \[Nr\[N,everyone \[cpwhol Angelton 
did \[ve\[v\] \[Nph\]\]\]\]\] 
b. Dulles suspected \[NP\[N,everyone \[cpwhol Angelton 
\[vP\[vsuspected\] \[NPh\]\] \] \] \] 
In 7a, the object of the head of the elliptical VP is realized 
as a 'trace, and the VP is interpreted by copying only the 
head, "suspected," of the antecedent VP into the position of 
the empty head to yield 7b. Therefore, the interpretive 
regress is avoided without QR. Moreover, subjacency viola- 
tions can be identified at S-structure by computing the 
relation between the wh-phrase and the trace that it binds. 
Our proposal handles subdeletion in a natural and 
straightforward way. The only difference between the sub- 
deletion structures in 6 and the ACD structure in 7a is that 
200 Computational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
arguments and adjuncts of the elliptical verbs in 6 are 
lexically realized, while the object of the elliptical verb in 
7a is not. The interpretation procedure is the same for the 
two cases. The head of the antecedent VP is copied into the 
head of the elliptical VP. If any arguments or adjuncts are 
missing in the elliptical VP and corresponding arguments 
or adjuncts are realized in the antecedent VP, the latter are 
inherited by the head of the former. 
This approach can be extended to intersentential VP 
anaphora cases like 8.11 
8. John arrived yesterday, and Mary did too. 
We simply treat the anaphoric relation in these cases as 
holding between the head and constituents of a structured 
empty VP, and the head and counterpart constituents of a 
full VP in another conjunct or sentence. 
Subdeletion is generally marginal with intersentential 
VP anaphora when arguments are left in the elliptical VP. 
9a. ??John writes articles, and Bill does books. 
b. ?Mary spoke to Max, but Sam won't to Lucy. 
However, subdeletion with adjuncts in these structures is 
considerably better. 
10a. John arrived today, and Bill did yesterday. 
b. Max spoke after Mary, and Sam will before Lucy. 
The fact that adjuncts can remain in partially empty VPs in 
these cases provides motivation for applying the structured 
(partially) empty VP analysis to intersentential VP ana- 
phora. 
3.2. AN ALGORITHM FOR VP ANAPHORA 
INTERPRETATION 
We define the predicate P is contained in Q recursively as 
follows. A phrase P is immediately contained in a head Q 
iff (i) P is an argument of Q, or (ii) P is an adjunct ofQ. P is 
contained in Q iff (i) P is immediately contained in Q, or 
(ii) P is immediately contained in a head R, and (the phrase 
with head) R is contained in Q. The following is a sche- 
matic description of our algorithm for VP anaphora resolu- 
tion. 
VP ANAPHORA ALGORITHM 
A. Identify an elliptical verb-antecedent verb pair (V,A) 
as follows. 
1. An elliptical verb V is identified by the presence of 
an auxiliary verb or the infinitival complementizer 
"to," where the auxiliary verb or the complemen- 
tizer does not have a realized verb complement. 
2. A candidate A for an antecedent of V is a verb that is 
not elliptical and not an auxiliary verb with a real- 
ized complement. 
3. Check that A and V stand in at least one of the 
following relations: 
a. V is contained in the clausal complement of a 
subordinate conjunction SC, and the SC-phrase 
is either (i) an adjunct of A, or (ii) an adjunct of a 
noun N and N heads an NP argument of A, or N 
heads the NP argument of an adjunct of A. 
b. V is contained in a relative clause that modifies a 
head noun N, N is contained in A, and, if a verb 
A' is contained in A and N is contained in A', then 
A' is an infinitival complement of A or of a verb 
contained in A. 
c. V is contained in the right conjunct of a sentential 
conjunction S, and A is contained in the left 
conjunct of S. 
B. Generate a new tree in which A is substituted for V as 
the head of the elliptical verb phrase VP' that V heads, 
and A is assigned the agreement features required by 
the head of VP'. (We will refer to this new occurrence of 
A as A'). 
C. Consider in sequence each argument slot Sloti in the 
argument frame of A. 
1. If Slot i is filled by a phrase C, then 
a. If there is a phrase C' in VP' that is of the 
appropriate type for filling Slot i, then fill Slot i in 
the argument frame of A' with the marker vari- 
able of C'. Else, 
b. Fill Slot i in A' with the marker variable of C, and 
list C as a new argument of A'. 
2. If Slot i is empty in the frame of A, it remains empty 
in the frame of A'. 
D. For each adjunct Adj of A, if there is no adjunct of the 
same type as Adj in VP', then list Adj as a new adjunct 
of A'. 
Part A of the algorithm specifies the procedures for 
identifying pairs whose first element is the head of an 
antecedent VP and whose second element is the head of an 
elliptical VP. Elliptical VPs are identified by the presence 
of a bare auxiliary verb or the bare complementizer "to." 
In fact, for reasons of convenience, we take bare auxiliaries 
and bare complementizers as standing for the head of an 
elliptical VP, and so the algorithm treats them as surrogate 
VP heads. A.3 defines the structural relations that hold 
between the head of an elliptical VP and the head of a 
possible antecedent VP. 12 
Part B describes the operation of generating a new 
interpreted VP anaphora tree in which the head of the 
antecedent VP is substituted for the head of the elliptical 
VP, and the features of the new head of the interpreted 
elliptical VP are adjusted in accordance with the require- 
ments of this VP. 
Part C characterizes a procedure for filtering the argu- 
ments of the antecedent verb to determine which of them 
are inherited by the head of the interpreted elliptical VP. 
Similarly, Part D describes the filtering process that gives 
the adjuncts of the antecedent verb that are inherited by 
the interpreted elliptical verb. The combination of the new 
verb heading the elliptical VP and the lists of arguments 
and adjuncts it inherits from the antecedent verb provide 
the interpretation of the elliptical VP. 
To illustrate the Prolog implementation of this algorithm 
Computational Linguistics Volume 16, Number 4, December 1990 201 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
on the basis of the network representation, we will give the 
clauses pertinent to A.3.b, which identifies the case in 
which the elliptical verb V is contained in a relative clause. 
The top-level predicate for testing that A is an anteced- 
ent of an elliptical verb V (used for implementing A.3) is 
anaph(A,V). The clauses for this predicate relevant to 
A.3.b are 
anaph(A,V) ~-- arel(A,V). 
arel(A,V) 
A =/V & relcont(V,N) & 
(pharg(N,A) I pharg(N,T) & phadjunct(T,A)). 
(Here A =/B means "A is not equal to B" in IBM Prolog.) 
The relation relcont(V,N) holds if V is contained in a 
relative clause adjunct of noun N. The predicate pharg 
(P,Q), which says that P is an argument of Q, was defined 
in Section 2 in terms of the network representation. The 
relation phadjunet(P,Q) says that P is an adjunct of Q, and 
is also straightforwardly defined in terms of the network. 
Let us consider several examples of the VP algorithm's 
results. The system produces the following output. For each 
ESG analysis of the input sentence, the parse tree is dis- 
played, and then all pairs (antecedent verb, elliptical verb) 
found by the algorithm are displayed. Then, for each such 
pair, the following three things are displayed: (1) the new 
arguments inherited by the interpreted elliptical verb from 
its antecedent; (2) the new adjuncts inherited by the inter- 
preted elliptical verb from its antecedent; and (3) the 
interpreted VP anaphora tree, as a modification of the 
original parse tree. 
1 \] in Figure 2 shows the output for May's ACD example 
1. In the parse tree, variable X5 in the object slot of the 
complement frame for the auxiliary "did" unifies with the 
phrase marker of the head of the relative "everyone" (and 
that of the wh-phrase "who"). Moreover, this variable is 
marked as a trace in the internal representation of the 
phrase structure from which the tree is projected. (Such a 
trace is marked in the feature structure of the verb--- 
although it is not shown in the following abbreviated dis- 
play.) Hence, the parse tree corresponds to the S-structure 
given in 7a. The VP anaphora algorithm identifies 
"suspected" as the antecedent of the elliptical verb (repre- 
sented by the auxiliary), and substitutes it for the auxiliary 
in the interpreted VP anaphora tree. No arguments or 
adjuncts are inherited from the antecedent verb. The inter- 
pretecl VP anaphora tree in 11 is the SG counterpart of the 
interpreted S-structure 7b. 
A similar ACD case involving an elliptical VP that 
follows a bare occurrence of the complementizer "to" is 
given in 12 in Figure 3. Here, the interpreted verb "write" 
inherits the object argument "notes," while its indirect 
object argument is a trace bound by the wh-phrase corre- 
sponding to the head of the relative clause. 
In the following examples, we give the output in an 
abbreviated way in order to save space. The uninterpreted 
tree is not shown, and the analysis resulting from the 
ii. Dulles suspected everyone who Angelton did. 
Syntactic analysis time = 87 msec. 
I sub3 Dulles(X4) noun(prop,nom.X2:,nwh) : top suspect(XI,X4,X5) verb(Ein~pers3,X2,past,X3)) 
u-r--- ob\] everyone(X5) noun(pron(all),,acc.sg,X6) l V- ob). who(X5) . noun(pron(wh),Xl0.sg~wh) 
I \[- SUD~ Anqelton(X9) noun(prop,nom.XS,nwh) ~-~-nre± do(X7,X9,X5) verb(~in~pers3,~\[8,past,dep:dcl:wh)) 
Antecedent verb-elliptical verb pairs: suspect.2 - do.6 
Elliptical verb-new argument pairs: none 
Elliptical verb-new adjunct pairs: none 
Interpreted VP anaphora tree: 
subj Dulles(X4) noun(prop,nom.X2,nwh) top suspect(Xl,X4,X5) verb(~in~pers3,X2,past,X3)) ~ 
OD~ everyone(X5) noun(pron~all),acc.sg,X6) ob who(~5) . _ noun(pron(wh),Xl0.sg/wh) sug  
Angelton(Xg) . noun(prop,nom.XS,nwh) nrel suspect(X7,X9,X5) verb(~in~pers3,~i8,past,dep:dcl:wh)) 
Anaphora analysis time = 69 msec. 
Figure 2 
202 Computational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
12. John wrote notes to everyone who asked him to. 
subj top 
ob" , ip~j ~ 
oD)prep SUD~ 
nrel obj 
L_ comp 
John(X3 ) nounl write(Xi,X3,X4,X5,u) verb~ 
note(X4) nounl to(X7,X5) prepq 
everyone(X5) nounq who(X5) nounl 
ask(Xg,XS,XII,XI2) verbq he(Xll~ nounq 
preinf(Xl2) preiz 
prop,nom.sg,nwh) ~in(pers3,sg,past,X2)) 
cn,acc.pl,nwh) to,XS,e(X5)) 
pron(all),acc.sg,XS) pron(wh),Xl3.sg,wh) 
~in(Xl0,sg,past,dep:dcl:wh)) t~ron(defprn),acc.sg,nwh) 
Antecedent verb-elliptical verb pairs: write.2 - preinf.9 
Elliptical verb-new argument pairs: preinf.9 - note.3 
Elliptical verb-new adjunct pairs: none 
Interpreted 
I 
VP anaphora tree: 
subj John(X3) noun(prop,nom.sg,nwh) top write(Xi,X3,X4,X5,u) verb(~in(pers3,sg,past,X2)) 
ob'3 no%e~X4) noun(cn,acc.pl,nwh) iob\] to(X7,X5~ prep~to,X8,e(X5)) 
ob~prep everyone(X5) noun(pron(all),acc.sg,X8) ~i 
who(X5) noun(Dron(wh),Xl3.sg,wh) ask(X9,X5,XII,XI2) verb(~in(Xl0,sg,pas£,dep:dcl:wh)) 
obj he(Xll} L noun(pron(defprn),acc.sg,nwh) comp preinf(Xl2) preinf 
auxcmp writel(Xl5,Xll,X4,X5) verb(inf(bare)) 
Figure 3 
algorithm is shown in an abbreviated linear form, consist- 
ing simply of head words and their arguments. 
In 13, both "promise" and its infinitival complement 
"read" satisfy the condition given in A.3.b on the anteced- 
ent of an elliptical verb in a relative clause. Therefore, the 
algorithm correctly generates two possible interpreted VP 
anaphora trees. Tree 1 gives the reading on which "promise" 
is taken as the antecedent of the head of the empty VP in 
the relative clause, and the infinitival clause headed by 
"read" is inherited as a new argument of the interpreted 
verb. Tree 2 specifies the interpretation where "read" is 
substituted for the head of the empty verb, and the trace of 
the relative operator "which" is its (noninherited) argu- 
ment. The algorithm correctly excludes "said" as a possible 
antecedent of the empty verb in the relative clause in 14. 
This is because it has a tensed rather than an infinitival 
complement that contains the verb that contains the noun 
modified by the relative clause. Therefore, the algorithm 
produces only two possible interpretations for 14, which are 
represented by the two interpreted VP anaphora trees that 
it generates. 
13. John promised to read everything which Mary did. 
Antecedent verb-elliptical verb pairs: 
promise.2 - do.8, read.4 - do.8 
Elliptical verb-new argument pairs: 
promise.8 - preinf.3, promise.8 - read.4 
Elliptical verb-new adjunct pairs: none 
Interpreted VP anaphora treel: 
John(X3) promise(X 1,X3,X4,u) preinf(X4) 
read(X4,X3,X7,u) everything(X7) 
which(X7) Mary(Xg) promise(X8,X9,X4) 
Interpreted VP anaphora tree2: 
John(X3) promise(X 1,X3,X4,u) preinf(X4) 
read(X4,X3,X7,u) everything(X7) 
which(X7) Mary(X9) read(X8,X9,X7) 
14. John said that Mary promised to read everything 
which Max has. 
Antecedent verb-elliptical verb pairs: 
promise.5 - have. 11, read.7 - have. 11 
Elliptical verb-new argument pairs: 
have.11 - preinf.6, have.11 - read.7 
Elliptical verb-new adjunct pairs: none 
Interpreted VP anaphora treel. 
John(X3) say(X 1, X3, X4,u, u) thatconj(X4, X9) 
Mary(X 10) promise(X9, X 10, X 11,u) 
preinf(X11) read(X11,X10,X12,u) 
everything(X 12) which(X 12) Max(X 14) 
have(X13,X14,X12) promise(X18,X14,X11) 
Interpreted VP anaphora tree2: 
John(X3) say(X1,X3,X4,u,u) thatconj(X4,Xg) 
Computational Linguistics Volume 16, Number 4, December 1990 203 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
Mary(X 10) promise 1 (X9,X 10, X 11,u) 
preinf(X 11) read(X11,X10,X12,u) 
everything(X 12) which(X 12) Max(X14) 
have(X13,X14,X12) read(X18,X14,X12) 
15 is a subdeletion case in which all of the arguments of 
the elliptical verb are filled locally within the elliptical VP. 
Hence, the algorithm substitutes the antecedent verb 
"write" for the auxiliary, and fills the direct and indirect 
argument slots in its frames with "notes" and "to Bill," 
respectively. It should be pointed out that the algorithm 
corrects ESG's parse of "to Bill" as a PP adjunct of "notes" 
in the original tree. This misparse is due to the fact that 
"do" does not allow an indirect object in its frame. The 
algorithm recognizes this adjunct as a possible filler for the 
indirect object slot in the frame of "write," and uses it to fill 
the slot when "write" is substituted for "do" in the inter- 
preted tree. 
15. Max writes more letters to Sam than Mary does notes 
to Bill. 
Antecedent verb-elliptical verb pairs: 
write.2 - do.9 
Elliptical verb-new argument pairs: none 
Elliptical verb-new adjunct pairs: none 
Interpreted VP anaphora tree: 
Max(X3) write(X1,X3,X4,X5,u) 
more(X14) letter (X4) to l(X12,X5) Sam(XS) 
than(X 1,XT) Mary(X8) write(XT, X8, X9,X 10) 
note(Xg) to(Xg,x 10) Bill(X 10) 
In 16, the indirect object slot of the interpreted verb 
"write" is filled locally by "to Bill," but "letters" is inher- 
ited from the antecedent. The algorithm also corrects the 
misparse of "to Bill" as a PP zdjunct of the elliptical verb in 
the original tree by a strategy similar to the one used to 
correct the parse of the PP in 15 
16. Max writes more letters to Sam than Mary does to 
Bill. 
Antecedent verb-elliptical verb pairs: 
write.2 - do.9 
Elliptical verb-new argument pairs: 
write.9 - letter.4 
Elliptical verb-new adjunct pairs: none 
Interpreted VP anaphora tree: 
Max(X3) write(X 1, X3, X4, X 5,u) more~X 13) 
letter(X4) to 1 (X 11,X5) Sam(X5) 
than(X1,X7) Mary(X8) write(X7,X8,X4,X9) 
to(X7,Xg) BIll(X9) 
Both the direct object "letters" and the indirect object 
"Sam" are inherited by the interpreted verb "write" in the 
fully empty ACD structure in 17. 
17. Max writes more letters to Sam than Mary does. 
Antecedent verb-elliptical verb pairs: 
write.2 - do.9 
Elliptical verb-new argument pairs: 
wTite.9 - letter.4, write.9 - Sam.6 
Elliptical verb-new adjunct pairs: 
n one 
Interpreted VP anaphora tree: 
Max(X3) write(X1,X3,X4,X5,u) more(X11) 
letter(X4) to(X9,X5) Sam(X5) 
than(X1,X7) Mary(X8) write(X7,X8,X4,X5) 
18 and 19 show the operation of argument filtering in an 
ACID passive case. 13 As in 13 and 14, the algorithm corrects 
the misparse, in the original tree, of the second "by" phrase 
as a PP adjunct. It raises it to the status of the agent (deep 
subject) argument of the head of the new verb in the 
interpreted tree. 
18. John was interviewed by Bill before Mary could 
have been by Max. 
Antecedent verb-elliptical verb pairs: 
interview.3 - be. 10 
Elliptical verb-new argument pairs: none 
E\]~tiptical verb-new adjunct pairs: none 
Inte:rpreted VP anaphora tree: 
John(X3) be(X 1, X3, X4) interview(X4,X 13,X3) 
by(X 14,X 13) Bill(X 13) before(X 1, X 5) 
Mary(X6) can(X5,X6,X7) have perf(XT,X6,X8) 
be(X8,X6,u) interview(X 12,X 10,X6) 
b:z(X8,X 10) Max(X 10) 
19. John was interviewed by Bill before Mary could 
have been. 
Antecedent verb-elliptical verb pairs: 
interview.3 - be. 10 
E\]2iptical verb-new argument pairs: 
be. 10 - Bill.5 
El'tiptical verb-new adjunct pairs: 
Inte:rpreted VP anaphora tree: 
John(X3) be(X1,X3,X4) interview(X4,X11,X3) 
bT(X12,X11) Bill(X 11) before(X1,X5) Mary(X6) 
can(X5,X6,X7) have perf(X7,X6,X8) be(X8,X6,u) 
interview(X 10, X 11,X6) 
20 and 21 illustrate the adjunct filtering procedure of the 
algorithm in an intersentential case of VP anaphora. The 
adverbial "yesterday" is inherited in 20, but not in 21. 
20. John arrived yesterday, and Mary did too. 
Antecedent verb-elliptical verb pairs: 
arrive.2 - do.7 
Elliptical verb-new argument pairs: 
none 
Elliptical verb-new adjunct pairs: 
arrive.7 - yesterday.3 
Interpreted VP anaphora tree: 
Jchn(X9) arrive(X8, Xg,u) yesterday(X I 1 ) 
and(XI,X8,X13) 
Mary(X14) arrive(Xl3,Xl4,u) too(Xl3) 
204 Com'p~ttational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
21. John arrived yesterday, and Mary will tomorrow. 
Antecedent verb-elliptical verb pairs: 
arrive.2 - will.7 
Elliptical verb-new argument pairs: 
none 
Elliptical verb-new adjunct pairs: 
none 
Interpreted VP anaphora tree: 
John(X9) arrive(X8,Xg,u) yesterday(X 11) 
and(X1,X8,X13) 
Mary(X14) will(X13,X14,u) arrive(X19,X14,u) 
tomorrow(X17) 
22 and 23 exhibit the effects of adjunct filtering of a more 
complex variety in an ACD structure. Both the adverb 
"briefly" and the PP adjunct "with Bill" are inherited by 
the interpreted verb "walked" in 22, but only "briefly" is 
inherited in 23. 
22. John walked briefly with Bill before Mary did. 
Antecedent verb-elliptical verb pairs: 
walk.2 - dol.8 
Elliptical verb-new argument pairs: 
none 
Elliptical verb-new adjunct pairs: 
walk.8 - briefly.3, walk.8 - with.4 
Interpreted VP anaphora tree. 
John(X3) walk(X l,X3,u, u) briefly(X I ) 
with(XI,X I O) Bill(X I0) 
before(X l,X6) Ma~/(X7) walk(X6,X7,u) 
23. John walked briefly with Bill before Mary did with 
Max. 
Antecedent verb-elliptical verb pairs: 
walk.2 - doi.8 
Elliptical verb-new argument pairs: 
none 
Elliptical verb-new adjunct pairs: 
walk.8 - briefly.3 
Interpreted VP anaphora tree. 
John(X3) walk(XI,X3,u,u) briefly(X1) 
with(XI,Xl 2) Bill(X 12) 
before(X 1,X6) Mary(X7) walk(X6,X7,u) 
with(X6,X 10) Max(X 10) 
4 A SYNTACTIC FILTER ON PRONOMINAL 
ANAPHORA 
The filter consists of six conditions for NP-pronoun non- 
coreference within a sentence. To state these conditions, we 
use the following terminology. The agreement features of 
an NP are its number, person, and gender features. We will 
say that a phrase P is in the argument domain of a phrase 
N iff P and N are both arguments of the same head. We 
will say that P is in the adjunct domain of N iff N is an 
argument of a head H, P is the object of a preposition 
PREP, and PREP is an adjunct of H. P is the NP domain of 
N iff N is the determiner of a noun Q and (i) P is an 
argument of Q, or (ii) P is the object of a preposition PREP 
and PREP is an adjunct of Q. 
4.1 FILTER ON PRONOMINAL ANAPHORA 
A pronoun P is noncoreferential with a (nonreflexive or 
nonreciprocal) noun phrase N if any of the following 
conditions hold. 
I. P and N have incompatible agreement features. 
II. P is in the argument domain of N. 
III. P is in the adjunct domain of N. 
IV. P is an argument of a head H, N is not a pronoun, and 
N is contained in H. 
V. P is in the NP domain of N. 
VI. P is a determiner of a noun Q, and N is contained in Q. 
Condition I rules out coreference between a pronoun and 
an NP with incompatible agreement features. It will iden- 
tify the co-indexed expressions in 24a-c as noncoreferen- 
tial. 
24a. *He i said that they i came. 
b. *The woman i said that he i is funny. 
c. *I i believe that she i is competent. 
The filter treats ("he," "they") as a noncoreferring pair, 
which entails only that the intended denotation of "he" 
cannot be taken as identical to that of "they." The referent 
of "he" can, of course, be a part of the referent of "they," 
and, in appropriate contexts, a discourse interpretation 
system, like the LODUS system of Bernth (1988, 1989), 
should be able to recognize this possibility. 
Condition II covers cases in which a pronoun and an NP 
are arguments of the same head, and so it rules out corefer- 
ence between the coindexed expressions in 25a-d. 
25a. *Mary i likes her i. 
b. *She i likes her i. 
c. *John i seems to want to see him i. 
d. *This is the girl i Mary said she i saw. 
It is important to note that the conditions of the filter 
apply to pronouns and NPs regardless of whether they are 
lexically realized in argument position (25a-b), or bind the 
argument slots which they fill in their heads at a distance 
through control (25c) and unbounded dependency relations 
(25d). This is due to the fact that the variable that fills an 
argument slot is unified with the phrase marker of the head 
of the phrase to which it corresponds. Therefore, it is not 
necessary to incorporate empty categories such as traces 
and PRO into the parse output, and compute appropriate 
binding chains for these categories in order for the algo- 
rithm to handle noncoreference in cases involving control 
and wh-movement. Mechanisms of this kind are required 
for implementations of Chomsky's binding theory in Gov- 
ernment Binding-based parsers, such as those described in 
Correa (1988) and Ingria and Stallard (1989). 
Computational Linguistics Volume 16, Number 4, December 1990 205 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
Condition III rules out coreference between an argument 
of a verb V and the object of a prepositional adjunct of V, as 
in 26a-bJ 4 
26a. *Mary i arrived with her i. 
b. *Who i did John say wants to sit near himi? 
Condition IV prevents coreference between a pronoun 
that is an argument of a head H, and a nonpronominal NP 
contained in H, as in 27a-c. 
27a. *Who i did she i say Mary i kissed? 
b. *This is the man i he i said Max i wrote about. 
c. *He i likes Johni's mother. 
The filter does permit coreference in 28a-b. "His" in 28a is 
not an argument of "likes," and so ("his," "John") do not 
satisfy Condition IV (or any other condition of the filter). 
An ordering constraint attached to Condition IV requires 
that a possessive adjunct of a noun contained in a head H 
follow a pronominal argument of H for this condition to 
apply to the pair. 
28a. His i mother likes John i. 
b. Johni's mother likes him i. 
Finally, V and VI in effect apply conditions II and III, 
respectively, to NP internal cases. They prevent corefer- 
ence in 29a-c, while allowing it in 29d. 
29a. *His i portrait of John i is interesting. 
b. *Johni's portrait of him i is interesting. 
c. *His i description of the portrait by John i is interest- 
ing. 
d. Johni's description of the portrait by him i is interest- 
ing. 
The filter on pronominal anaphora restricts the search 
space that a discourse system of anaphora resolution must 
consider. Bernth has integrated the filter into LODUS 
(Bernth 1988, 1989), which resolves pronominal anaphora 
and NP denotation through semantic and pragmatic rules 
of inference. The anaphora resolution component of 
LOI)US applies only to the pronoun-NP pairs that the 
syntactic filter has not identified as noncoreferential. 
An example of the filter algorithm's output is given in 30 
in Figure 4. The list of noncoreferential pronoun-NP pairs 
appears after the parse tree. Ulrike Schwall has success- 
fully implemented the algorithm in German Slot Grammar 
30. This is the girl who she wanted Mary to talk to. 
I subj this(X3) top be(XI,X3,X4) 
ndet the(X7) predcmp girl(X4) 
ob~prep who(X4) 
want(X8,X9,XlO,Xll) 
• Mary(XlO) . nf prelnf(Xll) 
halk(Xll,Xl0,X4,u) pob~ to(Xl6,X4) 
noun Dron(defDrn),nom.sq,nwh) verb ~in(pers3,sg,pres,~2)) 
det(g,de~) noun cn,X5.sg,X6) 
noun pron(wh),Xl3.sg,wh) noun pron(defDrn),nom.sq,nwh) 
verb ~in(pers3,sg,past,dep:dcl:wh)) nount~rop,acc.sg,nwh) 
preinz verb(inf(full)) 
prep(to,Xl7,e(X4)) 
Non-coreferential pronoun-NP pairs: 
she.6 - girl.4, she.6 - who.5, she.6 - Mary.8 
31. paul gab Peter das Buch, um ihn zu beeindrucken. (Paul gave Peter the book to impress him.) 
subj Paul(X2) .-- to geb(Xl,X2,X3,X4,u) 
io~j Peter{X3) ndet d(Xl3) 
F obj buch(X4) 
F-- sep preumzu prfumzu(X7) 
F- r- obj er(X8). 
\[- preinf preinf(X7) umzu beeindruck(X7,X2,X8,u) 
noun(prop,nom.sg.m.nda.X6,nwh) verb(Ein(pers3,sg,past~ind:dcl:nwh)) 
noun(prop,dat.sg.Xl4.nda.Xl5,nwh) det(naNS,def) 
noun(cn,acc.sg.nt.na.nrflx,nwh) separator 
preumzu noun(~ron(pers3),acc.sg.m.kda.Xl2,nwh) 
preinr verb(inf(ufull,a)) 
Non-coreferential pronoun-NP pairs: er.8 Paul.l 
Figure 4 
206 Computational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
(GSG). An example of its output for GSG is given in 31 in 
Figure 4. 
5 AN ANAPHOR-BINDING ALGORITHM 
We take the set of anaphors to include reflexive pronouns 
and the reciprocal NP "each other." The notion higher 
argument slot used in the formulation of the algorithm is 
defined by the hierarchy of argument slots given in 32. 
32. subj > agent > obj > (iobj I pobj) 
subj is the surface subject slot, agent is the deep subject slot 
of a verb heading a passive VP, obj is the direct object slot, 
iobj is the indirect object slot, and pobj is the object of a PP 
complement of a verb, as in "put NP on NP." We assume 
the definitions of argument domain, adjunct domain, and 
NP domain given in Section 4. 
5.1 ANAPHOR-BINDING ALGORITHM 
A noun phrase N is a possible antecedent binder for an 
anaphor A iff N and A do not have incompatible agreement 
features, and one of the following five conditions holds. 
I. A is in the argument domain of N, and N fills a higher 
argument slot than A. 
II. A is in the adjunct domain of N. 
III. A is in the NP domain of N. 
IV. N is an argument of a verb V, there is an NP Q in the 
argument domain or the adjunct domain of N such 
that Q has no noun determiner, and A is (i) an 
argument of Q, or (ii) A is an argument of a preposi- 
tion PREP and PREP is an adjunct of Q. 
V. A is a determiner of a noun Q, and (i) Q is in the 
argument domain of N and N fills a higher argument 
slot than Q, or (ii) Q is in the adjunct domain of N. 
Conditions I and II cover anaphoric binding in cases like 
33-34, respectively. 
33a. They i wanted to see themselvesi. 
b. Mary knows the people i who John introduced to each 
other i. 
34a. He i worked by himself i. 
b. Which friends i plan to travel with each otheri? 
Condition III handles binding of an anaphor inside an 
NP by the determiner of the NP, as in 35, and IV deals with 
NP internal anaphors which are bound from outside of the 
NP, as in 36. 
35. John liked Billi's portrait of himself i. 
36. They i told stories about themselves i. 
Condition V applies to cases in which a reciprocal deter- 
miner is bound by an argument in the same clause as the 
NP containing the reciprocal. 37 is an example of this 
binding relation, and 38 illustrates the combined effect of 
IV and V. 
37. \[John and Mary\]i like each otheri's portraits. 
38. \[John and Mary\]i like each otheri's portraits of them- 
selves i. 
An example of the anaphor binding algorithm's output is 
presented in 39 in Figure 5. Notice that the sentence in this 
example is ambiguous concerning antecedents for "himself," 
and the algorithm correctly identifies both "who" and 
"John" as possible binders of the reflexive. When a dis- 
course interpretation system makes use of this algorithm, it 
must, of course, constrain the interpetation of anaphors by 
requiring that exactly one binding pair be selected from the 
list of pairs that the algorithm provides for any given 
anaphor, relative to the clause in which it appears. 
Ulrike Schwall has implemented the algorithm in GSG, 
and 40 in Figure 5 illustrates its output for a German 
sentence. 
6 AN INTEGRATED SYSTEM FOR ANAPHORA 
RESOLUTION 
Any two of the algorithms described in Sections 4-6 can 
operate in conjunction with each other. Examples of the 
results provided by such combinations are given in 41-44 
(see Figure 6 for 41). In 45, both the filter and anaphor 
binding algorithms have been integrated into the VP ana- 
phora algorithm, and operate on the interpreted VP ana- 
phora tree it generates. 
6.1 VP ANAPHORA ALGORITHM WITH 
PRONOMINAL ANAPHORA FILTER 
42. John talked to him, and Bill did too. 
Antecedent verb-elliptical verb pairs: 
talk.2 - dol.8 
Elliptical verb-new argument pairs: 
talk.8 - he.4 
Elliptical verb-new adjunct pairs: 
none 
Interpreted VP anaphora tree: 
John(X9 ) talk(X8,X9, X I 0, u) to(X 12, X 10) 
he(XlO) and(XI,X8,Xl4) 
Bill(X I S) talk(X 14, X 15,u, X 10) too2(X 14) 
Noncoreferential pronoun-NP pairs: 
he.4 - John. I, he.4 - Bill.7 
43. Mary sent John to everyone who he did. 
Antecedent verb--elliptical verb pairs: 
send.2 - do2.8 
Elliptical verb-new argument pairs: 
send.8 - John.3 
Elliptical verb-new adjunct pairs: 
none 
Interpreted VP anaphora tree: 
Mary(X3) send(X I,X3,X4,X5) John(X4) 
to(X6,XS) everyone(X5) who(XS) 
he(X9) send(X8,X9,X4,X5) 
Computational Linguistics Volume 16, Number 4, December 1990 207 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
39. Who did John talk to about himself? 
l objprep who(X2) noun(pron(wh),X7.sg,wh) ~j do(XI,X3,X4) verb(~in(pers3,sg,past,ind\]:wh)) 
John(X3) noun(prop,nom.sg,nwh) ~ auxcmp talk(X4,X3,X2,u) verb(lnf~bare)) 
pobj to(Xll,X2) • prep(to,Xl2,e(X2)) L-~vprep about(X4,Xl0) prep(about,X9.e(Xl0)) 
oDjprep himself(Xl0) noun(pron(ref!prn),acc.sg,X9) 
~ntecedent NP-Anaphor Binding Pairs: who.l - himself.7, John.3 - himself.7 
40. Dies ist der Mann, der ueber sich sprechen soll. (This is the man who should speak about himself.) 
subj die§(X2) top sei(Xl,X2,X3,u) 
ndet d(X8~ predcmp mann(X3) 
pob3 ueber(Xl6,XI5) objprep reflex(Xl5) 
auxcmp sprech(Xll,X3,u,Xl5) nrel soll(Xg,X3,Xll) 
noun(Dron(defDrn),nom.sg.X5.na.X6,nwh) verb(~in(pers3,sg,pres,lnd:dcl nwh)) 
det(nMgdFSgP,def) .... noun(cn,nom.sg.m.w.A/,nwn) 
separator relpro 
prep(uber.acc,nwh,e(Xl5)) noun(pron(reflprn),acc.sg.Xl7.XlS.xlg,n wh 
verb(inf(bare,a)) verb(fin(pers3,sg,pres,dep:dcl:wh)) 
Antecedent NP-reflexive pairs: mann.4 - reflex.8 
Figure 5 
Noncoreferential pronoun-NP pairs: 
he.7 - Mary. 1, he.7 - John.3, 
he.7 - everyone.5, he.7 - who.6 
6.2 VP ANAPHORA ALGORITHM WITH ANAPHOR 
BINDING ALGORITHM 
44. The girl will write a book about herself, and Mary 
might too. 
Antecedent verb-elliptical verb pairs: 
write.4 - may. 12 
Elliptical verb-new argument pairs: 
may. 12 - book.6, may. 12 - about.7 
Elliptical verb-new adjunct pairs: 
none 
Interpreted VP anaphora tree: 
the(X 11) girl(Xg) will(X8,Xg,X 10) 
write(X10,X9,X12,u,u) a(X15) book(X12) 
about(X 12,X 16) hersell(X 16) 
and(X1,X8,X18) Maryl(X19) mayl(X18,X19,u) 
write(X24,X 19, X 12 ) too(X 18) 
Antecedent NP-reflexive pairs: 
girl. 2 - herself.8, Mary. 11 - herself.8 
6.3 VP ANAPHORA ALGORITHM WITH THE 
PRONOMINAL ANAPHORA FILTER AND ANAPHOR 
BINDING ALGORITHM 
45. They discussed each other's portraits of themselves 
before John and Mary did. 
Antecedent verb-elliptical verb pairs: 
discuss.2 - dol. 12 
Elliptical verb-new argument pairs: 
discuss. 12 - portrait.5 
El\].iptical verb-new adjunct pairs: 
none 
Interpreted VP anaphora tree: 
they(X3) discuss(X 1,X3,X4) 
each.other(X10) 's portrait(X4,Xg) 
o!(X 11,X9) themselves(X9) 
before(X1,X5) John(X7) and(X6,X7,X8) 
Mary(X8) discuss(X5,X6,X4) 
Noncoreferential pronoun-NP pairs: 
th.ey. 1 - portrait. 5, they. 1 - John.9, 
they. 1 - coord(and, John, Mary). 10, 
they. 1 - Mary. 11 
Antecedent NP-anaphor pairs: 
they. 1 - (each.other).3, 
(each.other).3 - themselves.7, 
coord(and, John, Mary). 10 - (each.other).3 
Our integrated system for anaphora resolution is syntac- 
tically based, and it must be supplemented by additional 
semantic procedures to yield fully adequate interpretations 
of e!iliptical VP structures. This can be seen quite clearly by 
considering the interpreted VP anaphora tree of 42. Here 
"him,," the indirect object of "talk," is inherited by the 
interpreted verb in the second conjunct, and its marker 
208 Computational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
Pronominal Anaphora Filter with Anaphor Binding Algorithm 
41. This is the man who he asked to talk about himself. 
. subj this( • top be(Xl 
l ndet the(X predcmp man(X 
w o(x he(X9 
nrel ask(X preinf Dreln 
comp halk(i pobj about 
objprep himse 
l~,x4) 
I,X9,X4,XIO) ~(xl0) 
II0,X4,usXI5) XI6,X15) 
f(x15) 
Non-coreferential pronoun-NP pairs: he.6 - man.4, he.6 - 
Antecedent NP-reflexive pairs: man.4 - himself.ll, who.5 
noun(pron(defprn),nom.sa,nwh) yerb(~in(pers3,sg,pres,X2)) 
det(sg,de~) noun(cn,X5.sq,X6) 
noun(pron(wh~,Xll.Xl2,wh) noun(pron(defDrn),nom.sg,nwh) 
verb(~in(pers3,sg,past,dep:dcl:wh) prelnr 
verb(inf(full)) prep(about,Xl7,e(Xl5)) 
noun(pron(ref!prn),acc.XlS,Xl7 ) 
who.5 
- himself.ll 
Figure 6 
variable X10 is unified with the marker of the indirect 
object slot in the argument frame of this verb. Therefore, 
the interpreted VP anaphora tree correctly represents the 
fact that the second conjunct in 42 must be understood as 
asserting that Bill spoke to the same person as John did. 
The list of noncoreferential pronoun-NP pairs specifies that 
"him" is distinct in reference from both "John" and "Bill." 
However, in its present form, the VP anaphora algorithm 
unifies the marker variables of all inherited arguments with 
the appropriate slots in the frame of an interpreted verb. 
This will yield incorrect results for a sentence like 46, where 
"a book" is the inherited argument. 
46. John read a book, and Mary did too. 
On at least one possible reading of the sentence, John and 
Mary read distinct books. To complete the interpretation of 
elliptical VPs, it will be necessary to add procedures for 
substituting new marker variables for the occurrence of 
inherited arguments and adjuncts in the interpreted VP, 
when these expressions need not be taken as having the 
same denotations that they receive as arguments and ad- 
juncts of the antecedent verb. 
A related problem concerns scope assignment in sen- 
tences like 47. 
47. Mary spoke to everyone after Max did. 
Dalrymple, Shieber, and Pereira (1990) point out that 47 is 
ambiguous between a narrow scope reading on which Mary 
spoke to everyone after Max spoke to everyone, and a wide 
scope reading according to which everyone is such that 
Mary spoke to him/her after Max spoke to him/her. At 
this point, the VP anaphora algorithm generates only the 
former reading, as "everyone" is inherited as an argument 
by the interpreted head of the ellided VP. 
We could capture the wide scope reading by modifying 
our S-structure copying analysis of VP anaphora to allow 
copying to apply to more abstract semantic representations. 
This approach involves adopting an interpolated copying 
theory of VP ellipses on which copying is permitted not only 
at the level of S-structure, but also after the antecedent 
clause has been assigned a partial or full semantic interpre- 
tation. In the case of 47, copying could apply after 
"everyone" has been assigned scope through the operation 
of NP storage and a semantic variable appears in its 
place) 5 The result of such copying would be an interpreta- 
tion on which "everyone" would have wide scope by virtue 
of the fact that it binds variables in both the antecedent and 
interpreted VPs. The interpolated copying analysis could 
be implemented within Slot Grammar by permitting either 
expressions or simply their marker variables to be inher- 
ited. The former case corresponds to S-structure copying of 
a constituent, the latter to copying at a level of representa- 
tion to which interpretation has already applied. If "every- 
one" is inherited in 47, it is, in effect, copied, and the nar- 
row scope reading of the sentence results. When only its 
marker variable is inherited, the semantic variable within its 
scope is copied, which yields the wide scope interpretation. 16 
43 is particularly interesting. The sentence is a variant of 
an example that May (in press) claims provides evidence 
for this QR treatment of ACD structures. He maintains 
that only after QR has been applied to "everyone who he 
did" and the matrix VP "sent John to t" copied into the 
empty VP in the relative clause, can Principle C of Chom- 
sky's binding theory rule out coreference between "he" and 
"John." In fact, the application of our filter to the inter- 
preted VP anaphora tree provides the correct results for 
this case. This is due to the fact that the VP algorithm 
identifies "John" as the inherited object of the verb that it 
Computational Linguistics Volume 16, Number 4, December 1990 209 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
substitutes for the elliptical verb in the new tree. Condition 
II of the filter algorithm is then satisfied by ("he," 
"John"). z7 This example provides strong support for our 
treatment of VP anaphora. 
In the interpreted VP anaphora tree of 44, the substi- 
tuted verb "writes" inherits "herself" as a new argument, 
and so we capture the "sloppy" reading of this sentence, on 
which each occurrence of the reflexive is bound by the 
subject of the clause in which it occurs. To obtain the 
"strict" interpretation, according to which "herself" is 
bound only by "the girl," it will be necessary to allow 
copying of the marker variable associated with "herself," in 
the manner required for the wide scope reading of 
"everyone" in 47 (see footnote 16). 
The fully integrated algorithm provides the desired re- 
sults for the sentence in 45. "They" is identified as nonco- 
referential with any of the NPs contained in the head of 
which it is the subject. "They" binds "each other" in the 
matrix clause, and "John and Mary" binds "each other" as 
the determiner of the inherited argument "portrait" in the 
adverbial phrase. "Each other" binds "themselves," and so, 
by transitivity of binding, "they" binds "themselves" in its 
occurrence in the object NP headed by "portrait" in the 
matrix clause, and "John and Mary" binds "themselves" in 
its occurrence in this NP as the inherited object of the 
substituted verb in the adverbial phrase. 
7 CONCLUSION 
We have presented three implemented algorithms for ana- 
phora resolution in Slot Grammar. The conjunction of the 
two NP anaphora rules covers approximately the same 
phenomena as Chomsky's binding theory, but these rules 
do not require empty categories and the definition of bind- 
ing chains in parse output. The VP anaphora algorithm 
implements an S-structure analysis of VP ellipsis that we 
argue offers a more unified and principled explanation of 
different sorts of VP anaphora than recent LF-based ac- 
counts. The success of the algorithm in providing appropri- 
ate representations for the VP anaphora cases that we used 
to motivate our theoretical approach supports this view. 
The combination of the three algorithms constitutes a 
powerful syntactically driven system for anaphora resolu- 
tion in Slot Grammar. This system reduces the burden on 
modules of semantic and discourse interpretation by supply- 
ing partially interpreted representations to which the rules 
of the latter can apply. 
NOTES 
1. Earlier versions of the paper were presented to the SRI natural 
language group at Menlo Park, CA in June, 1990, and to the AT&T 
Bell Laboratories natural language and speech generation group at 
Murray Hill, NJ in July, 1990. We are grateful to the participants of 
these two forums for their comments. We thank Mori Rimon for 
detailed and useful comments on an earlier version of the paper. We 
also very much appreciate the careful reading of the paper and the 
suggestions of three anonymous referees. We would particularly like 
to express our thanks to Fernando Pereira and Mary Dalrymple for 
extended discussion of this paper and the problems involved in VP 
anaphora resolution. Their own work in this area has provided us 
with considerable stimulation and insight. 
2. See Reinhart (1976, 1981, 1984), and Chomsky (1981, 1986b) for 
alternative definitions of c-command, and discussions of the role of 
c-command in determining the possibilities for anaphora. See Lap- 
pin and McCord (1990) for comparisons between the pronominal 
anaphora filter in Slot Grammar and recent implementations of 
Chomsky's binding theory in GB-based parsers. 
3. Shortly after we designed and implemented these three algorithms 
in Slot Grammar, Karen Jensen constructed three alternative proce- 
dures for anaphora resolution in the PEG grammar (see Jensen 1986 
for a general description of PEG). Moreover, George Heidorn has 
implemented a version of our filter on pronominal anaphora in PEG. 
Jensen's procedures and Heidorn's implementation of our filter 
algorithm rely on and apply after a set of socond-pass operations that 
comprise a module referred to as PEGASUS. This module computes 
deep grammatical roles from the surface configurational structures 
constituting the PEG parse. (See Jensen and Heidorn 1990 for a 
brief description of PEGASUS and an outline of Jensen's anaphora 
resolution procedures.) By contrast, in Slot Grammar deep grammat- 
ical roles are obtained directly in the course of parsing, through the 
unification of complement "marker variables" with variables in the 
argument frames of their heads. While PEGASUS reconstructs 
deep grammatical role information (primarily) from surface config- 
urational relations, the representation of these roles in Slot Gram- 
mar is lexically driven and is an integral part of the parsing process. 
Therefore, where Jensen's anaphora resolution procedures operate 
on the output of a second-pass module (they are, in effect, third-pass 
rules), our algorithms are formulated in terms of the head- 
complement structures provided directly by the Slot Grammar 
parser. See McCord (1984) and Lappin et al. (1989) for earlier 
systems that compute deep grammatical roles from PEG's surface 
parse structures. 
4. The list of complement slots in the argument frame of a verb 
includes its subject. Therefore, SG represents argument structure in 
a manner analogous to that of LFG in that it makes no structural 
distinction between the subject as an external argument of a VP and 
the internal arguments of the verb, as does Government Binding 
theory. 
5. The distinction between slot filler rules and ordering constraints 
parallels the difference between immediate dominance rules and 
linear precedence rules in GPSG. See Gazdar et al. (1985) for a 
characterization of ID and LP rules in GPSG. See McCord (1989b) 
for more discussion of the relation of Slot Grammar to other 
systems. 
6. IP is an inflectional phrase, the category to which sentences corre- 
spond in current versions of X' theory. See Chomsky (1986b) for 
details of the IP analysis of sentences. 
7. May's QR-based analysis of VP anaphora extends several of the 
ideas concerning the interaction of quantified NPs and VP anaphora 
originally proposed in Sag (1976). Webber (1978) adopts and 
modifies Sag's approach to VP anaphora within a computationally 
oriented framework. See Lappin (1984) for discussion of some of the 
difficulties that arise with Sag's original analysis. Lappin (in press) 
presents more detailed criticism of May's account, and of a variant 
of this analysis proposed in Haik (1987). This paper also deals with 
Larson's (1987, 1988) extension of May's account in ACD struc- 
tures in adverbial phrases. Other treatments of VP anaphora are 
discussed, and motivation is given for the S-structure interpretation 
view adopted here. In the following, we limit ourselves to a brief 
presentation of two main arguments against the LF approach to VP 
anaphora resolution, and a summary of the S-structure alternative 
that we propose. 
8. See Chomsky 1981 and 1986b for formulations of subjacency and 
arguments to the effect that it is an S-structure constraint. Haik 
presents an LF analysis of VP anaphora that classifies an empty 
antecedent-contained VP as a variable bound by a wh-(or empty) 
operator at S-structure. While Haik's account permits subjacency to 
constrain ACD structures, it requires that the VP variable be 
reanalyzed as an NP trace at LF in order to obtain a structure like 3. 
This is an ad hoc and otherwise unmotivated device. See Lappin (in 
press) for more detailed discussion of Haik's proposal. 
9. See Bresnan (1975) and Chomsky (1977) for the classical discussion 
of subdeletion. 
10. The idea that empty VPs are structured was initially proposed in 
Wasow (1972) and adopted in Williams (1977). 
11. See Lappin (1984) and the references cited there for discussions of 
intersentential VP anaphora. 
12. The VP anaphora algorithm identifies an elided VP by the presence 
of a bare auxiliary or infinitival complementizer. Therefore, it will 
not deal with elided VPs that are not introduced by auxiliaries or the 
complementizer "to," as in (i)a-b. 
(i)a. John wrote more papers than Mary. 
b. Bill arrived before Lucy. 
Complex syntactic and semantic factors must be invoked to distin- 
Computational Linguistics Volume 16, Number 4, December 1990 211 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
13. 
14. 
guish "bare" VP ellipsis cases of this sort from structurally similar 
sentences that do not contain elided VPs, such as (ii)a-b. 
(ii)a. John gave more papers than books to Mary. 
b. Bill arrived before the beginning of the concert. 
Extending the algorithm to cover "bare" VP ellipsis is clearly a 
nontrivial task, which is beyond the scope of our current work. We 
hope to take up this matter in future research. 
Unlike the examples where the elliptical verb is identified by the 
auxiliary "do," the elliptical verb in 19 is represented by the 
auxiliary "be" rather than the antecedent verb in the list of elliptical 
verb-new argument pairs. This is due to the fact that with auxiliaries 
other than "do" the algorithm copies the elliptical verb immediately 
after the auxiliary in the interpreted tree, while in the case of "do" 
the antecedent is substituted for the auxiliary. The filtering compo- 
nent of the algorithm applies after the antecedent has been inserted 
into the new tree, and so it identifies the antecedent with the 
elliptical verb for purposes of listing new arguments and adjuncts 
when the antecedent replaces the auxiliary, but not otherwise. 
Unfortunately, Condition III also incorrectly blocks coreference in 
cases like (i)a-b, (discussed in, for example, Reinhart (1981)) when 
"near him" is taken as an adverb modifying the head verb "saw." 
(i)a. Dani saw a snake near him i. 
b. Near him~, Dan i saw a snake. 
The problem with the cases of this kind is that the possibilities for 
pronominal coreference are notoriously susceptible to lexical varia- 
tion, as indicated by (ii)-(iii). 
(ii)a. John~ took a book with him~. 
b. *John i took a walk with him~. 
(iii)a. Maryi heard music near herl. 
b. ??Mary~ played music near her~. 
A variety of syntactic and lexical semantic factors seem to be 
involved in determining the possibility of coreference in these cases. 
However, when no complement intervenes between the subject of a 
verb and the pronominal object of its PP adjunct, coreference is 
always excluded. In light of this fact and the lexically governed 
complexity of the coreference patterns in structures like (i)-(iii), we 
h~tve decided to retain Condition III in its present general form. 
Clearly, it would be desirable to refine it to allow for the distinctions 
illustrated in the examples given here. 
15. Dalrymple, Shieber, and Pereira (1990) obtain both scope readings 
fi~r 47 by permitting interaction between storage and an equational 
procedure for ellipsis resolution. This procedure involves solving an 
equation in which the interpretation of the source clause appears on 
one side and a predication containing a higher order property 
wxiable corresponding to the ellided VP is on the other. The 
representation of the source clause in the equation can contain either 
a released or an unreleased stored NP interpretation. Lappin (1984) 
employs an analogous free interaction between NP storage and 
interpretation to derive both the globally wide and locally (conjunct) 
wide readings of "many windows" in (i). Sentences of this kind were 
originally discussed in Hirshbuhler (1982). 
(i) A Canadian flag was hanging in front of many windows, and an 
American flag was too. 
16. It will be necessary to constrain copying at all levels by the filter on 
pronominal anaphora, as the possibilities for pronominal coreference 
in ellided VPs are restricted by the conditions that the filter imple- 
ments. Thus, even if only the marker variable of a pronoun is 
inherited by an interpreted verb, it will be identified as the (denota- 
tion) marker variable of a pronoun, and subject to the pronominal 
coreference filter within the VP that the interpreted verb heads. 
Similarly, copying of reciprocal NPs must be restricted by the 
anaphor binding algorithm at every level of copying. The situation 
with respect to reflexives is less clear, given the possibility of strict as 
well as sloppy readings for reflexives in the interpretation of ana- 
phoric VPs. 
17. The variable of the wh-phrase (which unifies with that of the head 
noun of the relative clause) fills the object slot in the frame of the 
elliptical (auxiliary) verb in the original tree. It is correctly reana- 
lyzed as filling the indirect object slot of the substituted verb. 

REFERENCES 
Bernth, A. (1989). "Discourse understanding in logic." In Proceedings of 
the North American Conference on Logic Programming. MIT Press. 
755-771. 
Bernth, A. (1988 ). Computational discourse semantics. Doctoral disserta- 
tion, University of Copenhagen and IBM Research. 
Bresnan, J. (1975). "Comparative deletion and constraints on 
transformations." Linguistic Analysis, 1: 25-74. 
Byrd, 17,. J. (1983). "Word formation in natural language processing 
systems." Proceedings of the 8th International Joint Conference on 
Ar,ffficial Intelligence, 704-706. 
Corrca. N. (1988). "A binding rule for government-binding parsing." 
COLING '88, Budapest: 123-129. 
Chomsky, N. (1986a). Knowledge of Language: Its Nature, Origin, and 
Use. Praeger. 
Choms3cy, N. (1986b). Barriers. MIT Press. 
Chomsky, N. (1981). Lectures on Government and Binding. Foris. 
Chomsky, N. (1977). "On wh-movement." In Formal Syntax, P. Culi- 
cover, T. Wasow, and A. Akmajian, edited by Academic Press. 
Dalryrnple, M.; Shieber, S; and Pereira, F. (1990). "Ellipsis and higher- 
order unification." SRI, Harvard University, and AT&T Bell Laborato- 
ries, unpublished ms. 
Gazdar, G.; Klein, E.; Pullum, G.; and Sag, I. (1985). Generalized Phrase 
Structure Grammar. Blackwell. 
Haik, I. (1987). "Bound VP's that need to be." Linguistics and Philoso- 
phy 11: 503-530. 
Heidorn, G. E. (1982). "Experience with an easily computed metric for 
ranking alternative parses." Proceedings of the 20th Annual Meeting of 
the Association of Computational Linguistics, 1982: 82-84. 
Hirshbuhler. (1982). "VP deletion and across-the-board quantifier scope." 
Proceedings of NELS 12, edited by J. Pustejovsky and P. Sells. GLAS. 
Klawtns, J. L. and Wacholder, N. (1989). "Documentation of features 
and attributes in UDICT." Research Report RC14251, IBM T. J. 
Wat:~on Research Center. 
Ingria, R. and Stallard, D. (1989). "A computational mechanism for 
pronominal reference." In Proceedings of the 27th Annual Meeting of 
the Association of Computational Linguistics, Vancouver: 262-271. 
Jensen, K. (1986). "PEG: A broad-coverage computational syntax of 
English." Technical Report, IBM T. J. Watson Research Center. 
Jensen,. K. and Heidorn, G. (1990). "Post-syntactic processing of argu- 
ments and anaphora." Technical Report, IBM. 
Lappin, S. (1984). "VP anaphora, quantifier scope, and logical form." 
Linguistic Analysis 13:273-315. 
Lappi~., S. In press. "Concepts of logical form in linguistics and 
philosophy." In The Chomskyian Turn, edited by A. Kasher. Black- 
well. 
Lappin, S. and McCord, M. (1990). "A syntactic filter on pronominal 
anaphora in slot grammar." In Proceedings, 28th Annual Meeting of 
the Association for Computational Linguistics: 135-142. 
Lappin, S.; Golan, I.; and Rimon, M. (1989). "Computing Grammatical 
Functions from Configurational Parse Trees." Technical report 88.268, 
IBM Scientific Center, Haifa. 
Larson, R. (1988). "Scope and comparatives." Linguistics and Philoso- 
phy, 11: 1-26. 
Larson, R. (1987). "Missing prepositions and the analysis of English free 
relatives." Linguistic Inquiry, 13:273-315. 
McCord, M. C. (1990). "SLOT GRAMMAR: A system for simpler 
construction of practical natural language grammars." To appear in 
Natural Language and Logic, International Scientific Symposium, 
edited by R. Studer, Lecture Notes in Computer Science, Springer 
Verlag, 118-145. 
McC.ord, M. C. (1989a). "Design of LMT: A prolog-based machine 
tran:dation system." Computational Linguistics, 15: 33-52. 
McC, ord, M. C. (1989b). "A new version of slot grammar." Research 
report RC 14506, IBM Research Division, Yorktown Heights, NY 
10598. 
McCord, M. C. (1989c). "A new version of the machine translation 
system LMT." Journal of Literary and Linguistic Computing, 4: 218- 
229. 
McC.ord, M. C. (1989d). "LMT." In Proceedings of the MT Summit II, 
Deutsche Gesellschaft fiir Dokumentation, Frankfurt, 94-99. 
McCord, M. C. (1988). "A multi-target machine translation system." In 
Proceedings, International Conference on Fifth Generation Computer 
Systems, 1988, Institute for New Generation Computer Technology, 
Tokyo, Japan, 1141-1149. 
McCord, M. C. (1987). "Natural language processing in prolog." In 
210 Comp~ltational Linguistics Volume 16, Number 4, December 1990 
Shalom Lappin and Michael McCord Anaphora Resolution in Slot Grammar 
Knowledge Systems and Prolog: .4 Logical Approach to Expert Sys- 
tems and Natural Language Processing, edited by A. Walker, M. C. 
McCord, J. F. Sowa, and W. G. Wilson, Addison-Wesley. 
McCord, M. C. (1986). "Design of a prolog-based machine translation 
system." In Proceedings of the Third International Logic Program- 
ming Conference, Springer-Verlag, Berlin, 350-374. 
McCord, M. C. (1985). "Modular logic grammars." In Proceedings, 23rd 
Annual Meeting of the Association for Computational Linguistics, 
104-117. 
McCord, M. C. (1984). "Semantic interpretation for the EPISTLE 
system." In Proceedings, Second International Logic Programming 
Conference, Uppsala, Sweden, 65-76. 
McCord, M. C. (1982). "Using slots and modifiers in logic grammars for 
natural language." Artificial Intelligence, 18: 327-367. 
McCord, M. C. (1980). "Slot grammars." Computational Linguistics, 
6: 31-43. 
May, R. In press. "Syntax, semantics, and logical form." In The Chomsky- 
ian Turn, edited by A. Kasher. Blackwell. 
May, R. (1985). Logical Form: Its Structure and Derivation. The MIT 
Press, Cambridge, MA. 
Reinhart, T. (1984). Anaphora. Croom Helm. 
Reinhart, T. (1981). "Definite NP anaphora and c-command domains." 
Linguistic Inquiry, 12: 605~35. 
Reinhart, T. (1976). The syntactic domain ofanaphora. Doctoral disser- 
tation, MIT, Cambridge, MA. 
Sag, I. (1976). Deletion and logical form. Doctoral dissertation, MIT, 
Cambridge, MA. 
Walker, A.; McCord, M. C.; Sowa, J. F.; and Wilson, W. G. (1987). 
Knowledge Systems and Prolog: ,4 Logical Approach to Expert Sys- 
tems and Natural Language Processing. Addison-Wesley. 
Wasow, T. (1972). Anaphora in generative grammar. Doctoral disserta- 
tion, MIT, Cambridge, MA. 
Webber, B. (1978). A formal approach to discourse anaphora. Doctoral 
dissertation, Harvard University, Cambridge, MA. 
Williams, E. (1977). "Discourse and logical form." Linguistic Inquiry, 
8" 107-139. 
