Вы находитесь на странице: 1из 10

On dummy reflexive pronouns following unaccusative verbs

Alex Nisnevich
December 20, 2013

Introduction

In this paper, I explore perceived acceptability of the dummy reflexive pronoun following
unaccusative verbs. While HW#8 treated all such constructions as unacceptable, many
speakers find the dummy reflexive acceptable following unaccusative verbs.
For example, consider (1) below:
(1)

a.
b.

The machinery ground to a halt.


The machinery ground itself to a halt.

To most speakers, (1)b appears just as acceptable as (1)a. (1)b works so well that it
almost seems that grind is acting as an unergative verb here, but the Direct Object
Restriction asserts that the grammaticality of (1)a means that grind is unaccusative, at
least in (1)a.
How acceptable the dummy reflexive pronoun is varies for dierent sentences. For
example, (2) below appears significantly worse than (1)b, despite syntactic similarities.
Thus, there appear to be semantic factors at play.
(2) ??The television exploded itself.
My goal in this paper is twofold: I am interested in the semantic factors that govern the
perceived acceptability of the dummy reflexive pronoun following unaccusative verbs, as
well as possible syntactic justifications for why the dummy reflexive pronoun is grammatical in this position for some speakers.
In order to find out what proportion of speakers judge the unaccusative reflexive
to be acceptable and what semantic factors aect acceptability, I conducted a survey
on Mechanical Turk asking respondents to rate the perceived acceptability of a set of
unaccusative sentences, half of which had a dummy reflexive pronoun and half of which
did not. In sections 2 and 3 of this paper, I discuss this experiment and its results.

Experiment Setup

I conducted a Mechanical Turk survey with 250 respondents who were paid $0.100.25
for survey completion. To ensure a U.S.-centric set of responses, most of the survey jobs
were geographically restricted to the United States.1
Respondents were given the following prompt:

Rate the grammaticality of sentences


In this survey, you will be given a set of sentences and be asked to rate how
grammatical each sentence is. For the purposes of this survey, a sentence is considered to be grammatical if it would be normal and not unusual to say it in a
conversation. There are four sets of sentences, and each set has between 6 and 10
sentences. Look at each sentence carefully - some are very similar to others, but
all of these sentences are dierent.
They were then given four sets of sentences, and asked to rate each sentence in response
to the following question:
Are the following sentences grammatically correct? (That is, would it be normal
to use them in a conversation?)
Each sentence could be rated Yes, Probably yes, Probably no, or No (I used
a four-way distinction in order to allow respondents a greater range of freedom than a
simple Yes/No question would, but I avoided adding a middle option in order to force
participants to choose between leaning toward Yes or leaning toward No).
The full list of sentences considered in the survey is included in Figure 1. Each
sentence appeared with and without the dummy reflexive pronoun. In addition, sentences
were chosen in order to examine several phenomena, including:
Varying levels of animacy (e.g. The ice cream melted versus The reactor melted )
The contribution of the resultative (e.g. The gate froze shut versus The gate froze
itself shut)
The contribution of prepositional adjuncts (e.g. The chair broke itself versus The
chair broke itself under the weight)

The first batch on 100 tasks was not geographically restricted and paid $0.10 for completion. The
remaining 150 tasks were limited to the U.S. and paid from $0.20 to $0.25.
1

Sentences without
Sentence
The gate froze.
The gate froze shut.
The television exploded.
The chair broke.
The chair broke under the weight.
The engine started.
The light turned on.
The computer booted.
The computer booted up.
The ice cream melted.
The reactor melted.
The cancer metastasized.
The machinery ground to a halt.
The sky cleared up.
The wind calmed.
The winds calmed.

dummy
Yes
57.5%
58.3%
80.4%
81.7%
65.4%
83.5%
81.9%
56.0%
75.7%
86.8%
78.6%
75.7%
70.4%
85.6%
65.4%
55.6%

reflexive
P. Yes P. No
26.7% 11.3%
22.5% 12.1%
12.5%
6.7%
12.9%
4.2%
15.4% 12.1%
14.4%
2.1%
14.4%
3.7%
21.0% 18.5%
16.0%
7.4%
12.3%
0.8%
14.4%
6.6%
14.0%
9.1%
17.7%
9.1%
11.5%
2.9%
21.0%
9.5%
21.4% 14.0%

No
4.6%
7.1%
0.4%
1.3%
7.1%
0.0%
0.0%
4.5%
0.8%
0.0%
0.4%
1.2%
2.9%
0.0%
4.1%
9.1%

Sentences with dummy reflexive


Sentence
Yes
P. Yes
The gate froze itself.
9.6%
20.0%
The gate froze itself shut.
20.8% 20.8%
The television exploded itself.
8.8%
14.2%
The chair broke itself.
11.3% 12.9%
The chair broke itself under the weight. 15.4% 18.3%
The engine started itself.
18.9% 22.6%
The light turned itself on.
24.7% 21.0%
The computer booted itself.
16.5% 19.3%
The computer booted itself up.
27.2% 27.6%
The ice cream melted itself.
10.7% 11.9%
The reactor melted itself.
13.6% 20.2%
The cancer metastasized itself.
17.3% 21.4%
The machinery ground itself to a halt.
22.6% 28.0%
The sky cleared itself up.
16.9% 28.4%
The wind calmed itself.
15.2% 28.4%
The winds calmed themselves.
15.2% 22.6%

P. No
39.6%
34.6%
35.8%
40.0%
34.2%
36.2%
33.3%
43.2%
30.5%
38.3%
39.1%
38.3%
29.2%
35.4%
34.6%
31.7%

No
30.8%
23.8%
41.3%
35.8%
32.1%
22.2%
21.0%
21.0%
14.8%
39.1%
27.2%
23.0%
20.2%
19.3%
21.8%
30.5%

Figure 1: Complete survey results. The question asked was Are the following sentences grammatically correct? (That is, would it be
normal to use them in a conversation?), and the four options that respondents had for each sentence were Yes, Probably yes,
Probably no, and No.

4
Sentence
The computer booted itself up.
The machinery ground itself to a halt.
The light turned itself on.
The sky cleared itself up.
The wind calmed itself.
The gate froze itself shut.
The engine started itself.
The cancer metastasized itself.
The winds calmed themselves.
The computer booted itself.
The reactor melted itself.
The chair broke itself under the weight.
The gate froze itself.
The chair broke itself.
The television exploded itself.
The ice cream melted itself.

% Favorable
54.8%
50.6%
45.7%
45.3%
43.6%
41.6%
41.5%
38.7%
37.9%
35.8%
33.8%
33.7%
29.6%
24.2%
23.0%
22.6%

Figure 2: Sentences with dummy reflexives ordered by what % of respondents answered


Yes or Probably yes in regards to their grammaticality.

3
3.1

Results
Overall Results

The complete survey results are shown on Figure 1 on the preceding page. However, for
the remainder of this paper, I am interested only in the sentences with dummy reflexives,
and I will add the Yes and Probably yes results together in order to get a rough
sense of what % of respondents answered favorably in regards to each sentence. These
numbers are compiled in Figure 2 above.
Overall, the % of respondents who found these sentences acceptable ranged from
22.6% for The ice cream melted itself to 54.8% for The computer booted itself up, so even
on the most widely accepted sentence, almost half of respondents found it ungrammatical.
Unfortunately, I dont have access to cross-tabulated data, but I would assume that
there is high overlap between people who find all of these sentences favorable and/or
unfavorable.
Given this, the data seems to suggest that about half of the respondents found dummy
reflexive pronouns after unaccusative verbs grammatical overall and half found them
ungrammatical overall. Of course, even for the former group, certain semantic properties
need to be satisfied in order for a dummy reflexive pronouns to be acceptable after an

5
unaccusative verb.
Lets now take a look at the factors that seem to influence which sentences are deemed
more acceptable than others.

3.2

Animacy

First of all, the top three sentences in Figure 2 all have subjects that are machines or
mechanical/electrical devices in some sense (the computer, the machinery, and the light,
respectively). The next two sentences have the sky and the wind as subjects. On the
other hand, the bottom of the table has subjects like the ice cream, the chair, and the
gate. There appears to be a rough animacy hierarchy:
devices (e.g. machinery, computer )
i nature (e.g. sky, wind )
i other objects (e.g. chair, gate, ice cream)
That is, barring other factors, dummy reflexives after an unaccusative verb are acceptable
when the subject is a device more often than if the subject is a natural force, which in
turn is acceptable more often than when the subject is something else.
This rule is certainly not without exceptions. For example, The television exploded
itself has television certainly has a device as a subject, but is still the second-worstrated sentence. Whats going on here? It seems that this sentence is ranked lower than,
say, The computer booted itself up because, while booting up is standard behavior for a
computer, exploding is very unexpected behavior coming from a television. Even The
machinery ground to a halt, while describing non-ideal behavior, is still describing an
action that is reasonable given the subject, as opposed to The television exploded itself.
Thus, I would expect something like The bomb exploded itself to be rated much higher,
though I did not include that sentence in the survey.

3.3

The Role of Adjuncts

Consider (3) below:


(3)

a.
b.

(24.2%) The chair broke itself.


(33.7%) The chair broke itself under the weight.

It seems counterintuitive that adding a prepositional adjunct would significantly aect


how grammatical the sentence is perceived to be, but perhaps this phenomenon could be
explained by considering the semantic reasonableness of each sentence.
(3)a is semantically unusual: chairs are low on the animacy hierarchy in 3.2, and
so it is strange to say that a chair could break itself. On the other hand, in (3)b, the
addition of the adjunct under the weight clarifies the event that occurred: the chair broke

6
as a result of weight piled on it. In this case, it could be argued that the sentence is
semantically reasonable: the chair may have collapsed into itself due to the weight, thus
breaking itself.
In other words, by clarifying the situation at hand, the prepositional adjunct can help
suggest a reasonable semantic interpretation of the sentence, thus making the sentence
more acceptable than it would be without an adjunct.

3.4

Resultative Secondary Predication Again

As expected from Assignment #8, constructions without the resultative secondary predicate tend to be rated much lower than constructions with the resultative secondary
predicate. For example, consider (4) and (5) below.
(4)

a.
b.

(35.8%) The computer booted itself.


(54.8%) The computer booted itself up.

(5)

a.
b.

(29.6%) The gate froze itself.


(41.6%) The gate froze itself shut.

(4)b and (5)b perform much better than (4)a and (4)b, respectively. This is to be expected, because under our analysis, it is the resultative secondary predicate that licenses
the dummy reflexive pronoun in the first place.
However, this raises the question of why there are relatively highly-rated sentences
with dummy reflexive pronouns but no resultative secondary predicate. For example,
The wind calmed itself has 43.6% acceptance and The engine started itself has 41.5%
acceptance. One potentially plausible explanation is that, while calm and start are
usually unaccusative, they are behaving as reflexive verbs in this context, and so the
reflexive pronoun is licensed by the verb itself.
I am primarily concerned with dummy reflexive pronouns rather than actual reflexives, so in my syntactic analysis below, I will only take a look at constructions with
resultative secondary predicates.

Syntactic Analysis

The majority of respondents found both of the following sentences to be acceptable:


(6)

a.
b.

(77.0%) The computer booted up.


(54.8%) The computer booted itself up.

The analysis that we developed in Assignment #8 only accounts for the grammaticality
of (3)a. My goal now is to come up with a plausible analysis that also accounts for (3)b.

4.1

The Story So Far

First, let me briefly review my proposed mechanism for the Direct Object Constraint for
resultative secondary predication in Assignment #8. I proposed the following:
1. A new functional category Res with a single element:
Res[uTheme , uPred]
2. A new feature Pred that all Adjs and Ps have, as well as Cnf (the complementizer
that triggers non-finite verb phrases).
3. A new element in the v category:
vresult [uRes]
4. A new element in the D category, called the dummy reflexive pronoun:
D[refl, uacc, uRes]
5. A new rule:
The Reflexive Generalization. A reflexive (D[refl]) must be coreferential
with a c-commanding DP. When two expressions are coreferential, they must
share the same -features.
6. An addendum to UTAH:
d. ResP daughter of vresult P ! interpreted as Result
Under my analysis, The computer booted up has the following structure:

8
TP
T0

DP
the computer
T[past]

vP

v
boot[V,acc]

vresult P
v[uInfl:past]
0
vresult

vresult

ResP
VP

hbooti

hthe computeri

Res0

hthe computeri
Res

PP[Pred]
up

Note that, since boot up is an unaccusative verb, it assigns only a Theme, the computer, which moves first to the Spec-ResP position to satisfy Ress uTheme feature and
then to the spec-TP position to satisfy the EPP.
However, my analysis asserted that constructions such as The computer booted itself
up would be unacceptable, because there is no place for the dummy reflexive D[refl, uacc, uRes]:
it can occur only in the Theme position, but the Theme position appears to be occupied
by the television.

4.2

A Second Look

Lets examine the thematic roles in (6) more closely:


(6)

a.
b.

The computer booted up.


The computer booted itself up.

In (6)a, it is clear that the computer is the Theme: it is being booted up by an unseen
Agent. The semantic equivalence of the transitive construction in (7)a to the causative
construction in (6)b illustrates that the computer is behaving as the Theme in both (7)a
and (6)b, and thus also in (6)a:

9
(7)

a.
b.

I booted the computer up.


I caused the computer to boot up.

However, assuming that (6)b is valid, the computer is clearly behaving as an Agent there.
Thus, while boot appears to be unaccusative in (6)a, it appears to be unergative in (6)b.
There are two possible explanations for this behavior: either there are two lexical
entries for boot, one unaccusative and one unergative (and the unergative entry is only in
the lexicon of about half the sampled population), or there is some mechanism by which
a verb that is normally unaccusative can become unergative in this situation.
As an example of the latter, there could be a vagent that selects for a DP but has
complicated semantic requirements (for example, the DP should have some degree of animacy, though what degree is required may vary between speakers). These dierences in
s-selectional requirements of vagent can explain why dierent speakers can have dierent
grammaticality judgments when it comes to the dummy reflexive in this situation. (6)b
could then be interpreted something like this:
TP
T0

DP
the computer T[past]

vagent P
0
vagent

hthe computeri
vagent

vP

v
boot[V,acc]

vresult P
v[uInfl:past]
0
vresult

vresult

ResP
VP

hbooti

hitselfi

Res0

D[refl, uacc, uRes]


itself
Res

PP[Pred]
up

10
In this case the presence of the dummy reflexive pronoun is licensed, because boot
assigns accusative case to it, and Res satisfies its uRes feature. By the Reflexive Generalization, the dummy reflexive pronoun must be coreferential with a c-commanding
DP, the only candidate for which is the computer. By sharing its -features with its
coreferent, the dummy reflexive takes the form itself.

Conclusion

In summary, I found that more than 50% of respondents found at least some instance
of a dummy reflexive following an unaccusative verb to be grammatical. Animacy is
a major semantic factor governing how many respondents found a given sentence to
be acceptable, though I only have a vague sketch of a possible animacy hierarchy in
this context, and more detailed experiments would be needed to get a better sense
of it. Important syntactic factors include the presence of resultatives, and, somewhat
surprisingly, the presence of adjuncts.
Finally, I proposed that the unexpectedly high rates of acceptability for (at least
some of) these sentences could be explained by unaccusative verbs possibly exhibiting
an unergative alternation in certain contexts, conditioned by semantic viability.
An alternative explanation could be that some speakers have a secondary unergative
lexical entry for certain unaccusative verbs, though this explanation is somewhat unsatisfactory because the subject seems to be at least as important than the verb, if not
more, in determining acceptability, as shown in (8) below.
(8)

a.
b.

(22.6%) The ice cream melted itself.


(33.8%) The reactor melted itself.

There are still more directions to look at in this topic. In particular, consider (9) below:
(9)

a.
b.

(97.9%) The engine started.


(41.5%) The engine started itself.

As discussed in 3.4, while start behaves as an unaccusative verb in (9)a, it behaves as a


reflexive verb in (9)b. This is a phenomenon that was outside of the scope of this paper,
but it would be interesting to see how the unaccusative/reflexive alternation fits in with
the unaccusative/unergative alternation that I discussed above, as well as whether there
are unaccusative verbs that can exhibit both alternations with high rates of acceptance.
On a more general note, I found it interesting and enlightening to deal with phenomena that vary so much between English speakers. In a field that generally strives
toward a grammar that explains all possible constructions in a single framework, its certainly a compelling change of pace to deal with grammatical properties that are highly
speaker-dependent and work with probabilities rather than hard-and-fast rules.

Вам также может понравиться