Speaker's Intentions and Beliefs in Negative Imperatives 
Barbara Di Eugenio *t 
Department of Computer and Information Science 
University of Pennsylvania 
Philadelphia, PA, USA 
dieugeni@linc.cis.upenn.edu 
One of the questions asked of the participants to the workshop is What is the evidence for the 
existence of intentions? What types of intentions are useful to identify for communication? 
The former part of the question has already been positively answered by many researchers, 
such as \[GS86; GS90\], \[LC91\], \[LA90\]; the same answer emerges fl'om the abstracts submitted 
to the workshop, and I will therefore take it for granted. 
I will focus on the latter part of the question, and I will try to provide an answer by taking 
what seems to me a necessary preliminary step, namely, by identifying the factors underlying 
the speaker's communicative intentions. Some of these factors have been identified -- for 
example by \[LA90; Po186\] -- as the beliefs and intentions that the speaker (S for short, and 
referred to with feminine pronouns) believes the hearer (H for short, and referred to with 
masculine pronouns) has prior to the current utterance, and the new ones H will adopt as a 
consequence of the utterance. 
The analysis of such factors can be further finessed: one that is very pervasive in instructional 
text is the choices that S believes II has with respect to acting. While acting, an agellt 
is contimmlly faced with a.n infinite number of choices, of which the vast majority are not 
consciously considered, because they don't matter in the situation at hand. One task S is 
faced with, then, is to identify which choices matter, and to constrain the alternatives. I am 
going to show that the surface form of the utterance is affected by S's beliefs about the choices 
that H may either intentionally make or unintentionally overlook. In my previous work, I have 
identified how the choice among alternative courses of actions is constrained hy the goal that 
an action is meant to achieve, and how these constraints are expressed and computed in the 
case of purpose clauses \[Di 92; DW92\]. More evidence of the fact that S identifies and prunes 
H's possible choices is given by negative imperatives, and on those I want to concentrate in 
this contribution. 
Why is it appropriate to try to understand the factors underlying S's intentions in the work- 
shop? Consider for example Delin, Scott and Hartley's contribution \[DSH93\], which is par- 
ticularly relevant to my interests because they work on instructional text. They postulate 
six levels of representation that should be taken into account to generate instructions; among 
*Mailing address: IRCS - 3401, Wahmt St. - Suite 400C - Philadelphia, PA, 19104 - USA. 
tThis research was supported by the following grants: DARPA no. N00014-90-J-1863, ARO no. DAAL 
03-89-C-0031, NSF no. IRI 90-16592, and Ben Franklin no. 91S.3078C-1. Thanks to all the members of the 
AnimNL group, and in particular to Bonnie Webber, for very stimulating and helpful discussions. 
11 
them are the \[speaker's\] deep and shallow intentions, where the former is characterized as the 
,vpresentation of IS's\] intentions that \[H\] perform the sequence of actions that constitute a 
particular task, while the latter is a representation of the goals that the text has to achieve 
in order to motivate the required tasks. What is not specified in \[DSH93\] is how the shallow 
intcutions are formed on the basis of the deep intentions: it seems to me that factors such as 
the choices that S believes H has to face in acting play an important role in this process. 
Hearer's choices and negative imperatives 
One could take tim simple position that S's intention in uttering a negative imperative is 
to prevent a certain course of action on the part of H. However, we can finesse things by 
recognizing that, from S's point of view, a negative imperative is produced when 
• S believes II to be aware of a certain choice point, but expects him to choose the wrong 
altenJative among many .... possibly infinite -- ones. 
• S expects tl to overlook a choice point. The choice point is sometimes identified through 
a side effect that the wrong choice will cause. 
There appears to be a correlation between these two kinds of expectations and the two differ- 
(,t,l, classes of negative imperatives I ha.v(; identified, DONT and neg-TC. DON7' iml)eratives 
('<)ml)rise negative imperatives proper, characterized either by the negative auxiliary don'l or 
I)y negative polarity items; the other class is formed by verl)s such as take care, be sure and 
the like h)llowed by a negative infinitival complement, l 
As far as semantics goes, it is clear that a DONT imperative could be used when a neg-TC 
one is used: an expression like take care not to do a entails don't do a. In fact, in terse 
instructions only DONT imperatives are found. However, in the instructional texts I have 
examined, the two perform different functions. 
DONT imperatives are used when S thinks that H is likely to come to a choice point, and 
intentionally choose one course of action over another. Some situations in which this can 
hal)pen are: 
• When S provides H with general goals, or with rules of behavior to be always adopted in 
certain circumstances, as in 
(l) You can put parquet down over a variety of suTfaccs, whether old or new, if they 
ave firm, clean, .qmooth, and dry. Don't put parquet down on a surface that 
is below ground level because of the moisture problem. 
1My corpus consists of 36 instances of DONT and 27 of neg-TC imperatives. I collected the data from two 
"how-to-do" books plus a few from detergents and cosmetics containers. 
12 
• When S identifies an a.ction a which is an undesirable alternative to another action fl 
that she tells H to do, a.s in: 
(2) Caring for the floor. A good paste wax - not a water-based wax - will give added 
protection to the wood. Buff about twice a year; wax about once a year. Excessive 
waxing can cause wax to build up, detracting from the floor appearance. 
Dust-mop or vacuum your parquet floor as you would carpeting. Do not scrub or 
wet-mop the parquet. 
Clearly, S thinks that H, after adopting the intention of cleaning the parquet, may choose 
to do so in a wrong way. Notice that using a neg-TC verb in this case would be infelicitous, 
as it would seem to imply that H could unintentionally choose to perform either scrub 
or wet-mop. 
Neg-TC imperatives are used when S expects H to overlook a certain choice point; such 
choice point may be identified through a possible side effect. Moreover, a neg-TC seems 
to be used only when S relates the negated action a to another action fl in the discourse 
-- in contrast, DONT imperatives can be used independently of other actions mentioned in 
discourse, as in warnings like In case of fire don't use the elevator. A form like "Do/3. 7hke 
care not to do a" appears to be used when 
• a. is a.n undesirable way of performing ft. The description of fl is always underspecilied, 
and therefore H has many degrees of freedom in executing it. Consider 
(3) To make a piercing cut, first drill a hole in the waste stock on the interior of the 
pattern. The diameter of the hole must be larger than the width of the blade. If you 
want to save the waste stock for later use, drill the hole near a corner in the pattern. 
Be careful not to drill through the pattern line. 
fl is drill a hole near a corner in the pattern. The interpretation of near still leaves H 
some choices as regards the exact position where to drill: S constrains them by ruling 
out a "drill \[a hole near a corner in the pattern\] through the pattern line". 
Notice the difference in the relations between the actions involved in Exs. 2 and 3: ill tile 
former, there exists a third action 7 "clean parquet" such that a "scrub" and fl "vacuum" 
are alternative ways of achieving it; in the latter, there is no such 7, and a is directly 
related to fl as one of its possible specializations. 
• ra, is an undesirable effect of/3, which may be under H's control, or under the control of 
external laws, as in 
(4) 7b hang the border, begin at the least conspicuous corner. The work will go much 
faster if you have someone hold the folded section while you apply the border to the 
wall. Take care not to drip paste onto the wall. 
S expects H not to realize that there are different ways of performing the action hanging 
the border, some of which result in the side effect of dripping paste on the wall; and 
therefore S alerts H to take steps in order to prevent that from happening. 
13 
Some further evidence that S's expectations on H's possible choices affect the use of a DONT 
versus a ncg-TC imperative is provided by the infelicitousness of a--- really occurring! -- 
DONT imperative: 
(5) # If you must replace a tile, first cut a~vund the edges with a circular saw. Set the blade 
to the depth of the tile and don't damage adjoining tiles. 
The previous example is not felicitous because, in the context of an assembly task, damage is 
ltot a choice that I\[ has at his disposal, but rather, it is a side effect that may derive from 
certain choices in executing cut edges with a saw: therefore neg-TCwould be more appropriate 
than DONT. 
3'0 conclude, I have presented evidence that the possible choices that S believes H has with 
respect to acting affect S's communicative intentions. Further work is needed to identify other 
factors underlying S's communicative intentions; as far as negative imperatives are concerned, I 
have started doing work on other components that come into play, such as the lexical semantics 
of the negated action \[Di 93\]. 

References 
\[Di 92\] Barbara Di Eugenio. Understanding Natural Language Instructions: the Case of Pur- 
pose Clauses. In Proceedings of the 30th Meeting of the ACL, pages 120-127, 1992. 
\[Di 93\] Barbara Di Eugenio. A Study of Negation in Instructions. In The Penn Review of 
Linguistics, Volume 17, 1993. 
\[DSII93\] Judy Delin, Donia Scott, and Tony Hartley. Knowledge, Intention, Rhetoric: Levels 
of Variation in Multilingual Instructions. In ACL Workshop on Intentionality and 
Structure in Discourse Relations, 1993. 
\[DW92\] Barbara Di Eugenio and Michael White. On the Interpretation of Natural Language 
Instructions. In Proceedings of the Fourteenth International Conference on Computa- 
tional Linguistics, COLING 92, pages 1147-1151, 1992. 
\[GS86\] Barbara Grosz and Candace Sidner. Attention, Intentions, and the Structure of Dis- 
course. Computational Linguistics, 12:175-204, 1986. 
\[GS90\] Barbara Grosz and Candace Sidner. Plans for Discourse. In P. Cohen, J. Morgan, and 
M. Pollack, editors, Intentions in Communication. MIT Press, 1990. 
\[LA90\] Diane Litman and James Allen. Discourse Processing and Commonsense Plans. In 
P. Cohen, J. Morgan, and M. Pollack, editors, Intentions in Communication. MIT 
Press, 1990. 
\[LC91\] Lynn Lambert and Sandra Carberry. A tripartite plan-based model of dialogue. In 
Proceedings of the 29th Annual Meeting of the ACL, pages 47-54, 1991. 
\[Po186\] Martha Pollack. Inferring domain plans in question-answering. PhD thesis, University 
of Pennsylvania, 1986. 
