COLEVG 82, J. Horeckf/ (ed.) 
North-Holland Publishing Company 
© Academia, 1982 
FRAME BASED RECOGNITION OF THEME CONTINUITY 
James T. Critz 
Hewlet t-Packard Company 
Palo Altow California 
U.S.A 
The paper describes a system which determines 
continuity and shifts in English texts on the basis 
of sentential themes. The theme object within the 
thematic component of a sentence is determined and 
a search is made to associate it with a frame 
shared with the previous theme or themes. If the 
theme object cannot be associated directly with one 
of these frames, interpretive rules are ~pplied to 
do so indirectly through one of the frames normally 
associated with the object but not yet with thetext. 
INTRODUCTION 
This paper describes a system which interprets the semantic theme of 
a text as a mapping of the semantic network of states associated in 
the text with one or more theme objects onto an established semantic 
frame. Theme objects are determined on the basis of the syntactic 
distinction of thematic and rhematic (topic - Comment) components of 
a sentence. The concept of "a frame" as used here is that of a 
semantic representation of experience~ whose functional roles are 
related, rather than of a psychological representation, where 
physical properties of an object which do not bear any functional 
relevance to the definition of the object and its relationship to 
other components of the frame are included. 
The problem of recognizing ccntinuity and shifts in text analysis is 
generally defined in terms of identifying the text's themes. It is 
apparent from examples like (i - 2) that a system which tries to 
discover continuity on the sole basis of searching for antecedent 
references for a pronoun or synonym is inadequate where a text is 
developed through the techniques of elaboration, analogy, comparison 
and other styles involving implied dependency. 
(i) 
(2) 
The store was crowded. The marketing campaign had been a 
Success. 
Harold notified hfs lawyer. A storm was brewing. 
In order to identify the implied theme object in text samples such 
as these, the system described in this paper makes use of the 
distinction ~f the syntactic categories "theme" and "rheme" 
developed by the Prague School of Linguistics. Once it has 
determined the sentential theme it further relates this th~me to a 
71 
72 J.T. CRITZ 
frame which can be reasonably associated with it in the given 
context and links the frame with those extablished in the preceding 
text. The objects which are the primary referents of a sentential 
theme are assigned the role of being major nodes through which one 
frame is linked directly or, where more abstract associations can be 
inferred, indirectly with others relevant to the continuity of the 
text. The concept of "frame" is that described in M. Minsky 
(1977). 
In the following sections the concepts of "theme" and "frame" are 
defined more explicitly and the manner in which the system relates 
them to each other is described. While the system is designed to 
operate on English text, comparisons will be made to Czech. 
THEMES 
Within the grammatical theory of the Prague School of Linguistics a 
sentence is composed of a minimum of two major constituents - a 
"theme" ("topic") and a "rheme" ("comment"). (A sentence may also 
contain components which function as a transition between these two, 
but such components will not be discussed here.) The theme - rheme 
division of a sentence typically conforms, in non-contrastive 
environments, to the NP subject and VP major constituents of a 
sentence, with the theme being the first of these to appear in the 
surface word order of the sentence. Because of its more restrictive 
word order, the theme of a sentence in English is normally its 
subject. In Czech, however, a sentential subject may more easily 
appear in the latter part of a sentence. 
In its attempt to associate the theme of a sentence with a frame, 
our system reduces the object of its operation to a "theme object", 
defining this for English as the referent of the highest noun within 
the theme constituent: normally, the subject head noun. Within VPs 
functioning in English sentences as theme the theme object would be 
the direct object's head noun, or, given other contrastive stress, 
the head noun of the phrase containing the highest peak of 
intonation were the sentence spoken. 
It is important to note that the theoretical basis for labelling a 
sentence constituent as theme does not necessarily mean that the 
theme object has been previously mentioned in the text. The 
immediate introduction of new theme objects in English is obvious 
from the existence of non-definite themes in examples like (3). 
(3) The old hotel burned down last year. A new restaurant 
is to take its place. 
However, unless some relation can be found between contextually new 
themes and the existing context, sentences with non-definite themes 
would be either make a text discontinuous or would maintain their 
continuity only indirectly through the relation of the theme to 
another noun in the same sentence which is anaphoric. In either 
case there would be little justification for assigning a thematic 
function to non-definite nouns, and the general definition of theme 
proposed by the Prague School would have to be greatly altered or 
dropped. 
FRAME BASED RECOGNITION OF THEME CONTINUITY 73 
The association of themes with frames provides continuity without 
the requirement for explicit previous mention. The claim made in 
associating themes and frames is that themes are not referentially 
simple: their meaning involves reference not only to some 
individual, act, event, etc., but also reference to some structured 
set of these with which the noun referent is associated. Continuity 
lies in the existence of an assumed shared relation holding between 
a theme and other objects previously mentioned. 
FRAMES 
A frame may be generally defined as a limited complex of objects, 
acts, events, and/or states, whose association is determined by 
convention or experience and is assumed and expected by a speaker or 
author to be known by his or her ~udience. Simple frames include: 
the association of the objects "the bathroom" ("koupelna"), "the 
bedrooms" ("lo~nice"), "the kitchen" ("kuchyn~"), "the roof" 
("st{echa"), etc., with the frame "h)use", or the sequence "the 
morning" ("r~no" a "dopoledne"), "noon" ("poledne"), etc., as parts 
of the frame "day". Reference to one member of a frame suffices to 
make other members of the frame available as possible themes. For 
example, mention of "the kitchen and bathroom" implies the frame 
"house", and this mention, if themes are assumed to be bound to 
frames, suffices to allow the occurrence of "the house", "the 
garden" or other associated object as a theme in a later sentence. 
The primary characteristics of a frame, as defined in our natural 
language system, and distinguishing frames from other relational 
structures, are that each member of the frame must be uniquely 
identifiable (This uniqueness also specifies the member's role in 
the frame.) and that the members share a common relation or 
property which uniquely characterizes the frame to which they 
belong. Recognition that a given object is a unique member of a 
frame is the attribution of the property of "definiteness" to that 
object. A non-definite theme is interpreted as a description of a 
frame object, but the reference itself as definite. 
DETERMINING CONTINUITY 
Themes and frames are related to each other through the following 
process. The main components of the system include: 
- a parser 
- a lexicon, providing not only the possible syntactic and 
semantic representations of each lexical item and 
conditions on their use, but also an index to the frames 
normally associated with the entry 
- a frame dictionary, stating the set types of the possible 
members for each frame and their relationship to each other 
within the frame 
- a set of logical inference and interpretive rules, which can 
be used to relate frames in the construction of a frame 
system representing changes in the continuity and development 
of a text 
- a list of frames currently activated by theme reference, 
interpretation or inference 
74 \]J.T. CRITZ 
During the parse of a sentence the head noun of the sentence is 
selected as the most likely candidate for being its theme and, 
provided that tests of the sentence and the onepreceding it do not 
indicate a contrastive environment requiring the selection of 
another noun as theme, declares it to be the theme. The frames 
indexed in the lexical entry for the theme noun are then searched 
and compared for identity with or inclusion within those frames 
currently active. Those frames most recently used and those which 
a~e associated with anaphoric noun phrases within the sentence are 
examined first. If the sentence being processed is the first 
sentence in the text or if a search of all active frames fails, the 
theme is associated with a "base" frame. The base frame for a text 
includes as its members the set types {AUTHOR}, {AUTHOR'S GOAL~, 
READER and other nodes which may be filled by objects or persons 
these as part of their immediate world. 
The frame system for the text is built of the relationships holding 
between all the activated frames and the base frame. 
The process is illustrated for (4). For reasons of space, the 
hierarchical relations which hold between members of each frame is 
not shown. An AGENT frame is automatically assigned to persons 
mentioned in a text. The LOCATION, BUILDING and LOCK frames in (4) 
are interpreted. 
(4) John turned the key. The door still would not move. 
BASE {... \[John\] \[author\] \[reader\] ...~ 
AGENT (... \[John\] \[LOCATION > BUILDING\] \[GOAL\] 
BUILDING {... \[door\] ..-7 
LOCK (... \[key\] \[bolt\] ...~ 
DOOR (... \[lock\] \[handle\] ...) 
} 
COMPARING ENGLISH AND CZECH TEXTS 
Comparing the processing of English and Czech texts, it is important 
to note that while most frames can be expected to be the same for 
both languages, frame membership and relations between frame members 
which are dependent on cultural, political, linguistic and similar 
differences will cause frames to differ at least at the lower node 
levels. The "day" frame menti6ned earlier, in which English 
"morning" is paralleled by "r~no" and "dopoledne" in Czech, is one 
example. 
An even more significant difference is found in the effect of 
varying restrictions on word order in the two languages. Czech 
seems to indicate shifts to new frames by moving subject nouns which 
would b e thematic in the English equivalent of the sentence to the end 
of the sentence. Where the same frame or same samll set of frames 
can be associated with consequetive subjects in the text, the 
subject is thematic. An example of this can be seen in (5), 
where "cesta" ("trip") is associated with the general frame of a 
person's daily life and with Karel's in particular. In (6), 
FRAME BASED RECOGNITION OF THEME CONTINUITY 75 
however, where a shift is made from an AGENT frame to introduce an 
object which is not, on the basis of previous-themes, expected, 
although an association could be found, Czech makes the subject 
thematic, and uses as its theme an object more easily associated 
with existing frames. English may optionally keep "envelope" as the 
theme of its equivalent for (6) and allows interpretive rules to 
determine the frame it is associated with. 
(5) Karel se vcera prestehoval do nov~ho bytu. Cesta 
Karel himself yesterday moved to new apartment, trip 
do prace u{ bude krat{{. 
to work now will-be shorter. 
Karel moved to a new apartment yesterday. His trip 
to work now will be shorter. 
(6) Marie otev{ela dve{e. Na zemi le{ela velk~ obalka. 
Marie opened doors, on ground lay large envelope. 
Marie opened the door. I A large envelope lay on the ground.~ 
There was a large envelope laying \[ 
on the ground. J 
References

\[i\] Critz, J., Definiteness and Knowledge Frames, Paper presented 
to the Annual Meeting of the Linguistic Society of America, 
(December 1981). 

\[2\] Danes, F., Papers on Functional Sentence Perspective (Cesko- 
slovensko akademie red, Prague, 1974). 

\[3\] Hirst, G., Discourse-Oriented Anaphora Resolution, American 
Journal of Computational Linguistics 7.2 (1981) 85-98. 

\[4\] Johnson-Laird, P. and Wason, P. (eds.), Thinking: readings 
in cognitive science (Cambridge University, New York, 1977) 

\[5\] Minsky, M., Frame System Theory, in: Johnson-Laird, P. and 
Wason, P. (eds.), Thinking: readings in cognitive science 
(Cambridge University, New York, 1977) 
