What if Chinese had Linguistic Markers for Counterfactual Conditionals?

Language and Thought Revisited

Gary Feng (garyfeng@duke.edu)

Department of Psychology: Social and Health Sciences, Box 90085,
Duke University, Durham, NC 27708, USA

Li Yi (ly3@duke.edu)
Department of Psychology: Social and Health Sciences, Box 90085,
Duke University, Durham, NC 27708, USA

Abstract investment of cognitive effort and hence less naturally

We report a series of study critically examines the widely
accepted claim that the Chinese language provides no
consistent linguistic forms for expressing counterfactual ideas
(Bloom, 1981). In Study 1, native speakers of Chinese
identified counterfactual sentences from a large corpus of
Chinese texts. A number of syntactic and/or lexical forms are
found to be highly predictive of the counterfactuality of the
sentence, as high as over 95% in some cases. Aspect
modification and other markers emerge as potentially
universal linguistic mechanisms for counterfactual marking.
Study 2 demonstrated that these linguistic markers
significantly influence readers' interpretation of whether a
sentence is factual or counterfactual, even after semantic and
pragmatic information was controlled. Study 3 showed that
the counterfactual markers influence on-line sentence
comprehension in a self-paced reading paradigm. Taken
together, our data suggest that most previous studies on this
issue are based on a premise that is not backed by linguistic
data. We also offer an alternative framework for thinking
about language and thought that emphasize on universal
constraints and cognitive processing.
Introduction
Research on the relation between language and thought has
largely focused on the idea that cognition is constrained or
shaped by linguistic structures, often referred to as linguistic
determinism or the Sapir-Whorf Hypothesis (@@). The
logic is perhaps best illustrated by the controversial work on
counterfactual reasoning that began some 25 years ago
(Bloom, 1981, 1984). Alfred Bloom observed that:
… the Chinese language has no distinct lexical,
grammatical, or intonational device to signal entry into
the counterfactual realm, to indicate explicitly that the
events referred to have definitely not occurred and are
being discussed for the purpose only of exploring the
might-have-been or the might-be. [Bloom, 1981, p.16]
In comparison, counterfactual conditions in English are
thought to be consistently marked with the subjunctive
mood. Bloom reasoned that because
“Chinese speakers have not been led by their language
to construct schemas specific to counterfactual speech
and thought … they would typically do so
[counterfactual reasoning] less directly, with a greater
than their English-speaking counterparts. [p. 22]”
Empirical support for Bloom’s prediction is weak at the
best. With the noticeable exception of Bloom’s original
studies (1981; 1984), most studies reported remarkably
similar performances among Chinese- and English-speakers
in counterfactual reasoning tasks (Au, 1983; Liu, 1985; Hsu
et al., 2004; Wu, 1994; Yeh & Gentner, 2005). It is tempting
to conclude that the lack of a consistent linguistic form for
marking counterfactuality in Chinese has little negative
effect on counterfactual thinking, thus refuting the strong
version of linguistic determinism (Au, 1983). But there is
another logical possibility that has so far eluded critical
examination – the empirical data would make perfect sense
if Chinese, like English, has consistent linguistic cues for
counterfactual propositions.
Linguistic marking of counterfactuals
Bloom’s assertion that Chinese does not have any linguistic
device to signal counterfactuals was based more on
intuitions than serious linguistic research. In fact, despite the
popularity of the topic, we were only able to find a handful
of linguistic inquires on this topic (e.g., Chao, 1968; Jiang,
2000; Wu, 1999). The conjecture may “feel right” to naïve
Chinese speakers because such discussions almost certainly
arise in the context of comparing Chinese and English.
Being an isolate language, Chinese basically has no verb
inflections that correspond to the “subjunctive mood” that is
associated with counterfactuality in English. This, however,
by no means preclude systematic marking of counterfactuals
using other linguistic forms.
Yuan-ren Chao, a world-renown Chinese linguist,
speculated that the many “if-words” in Chinese may serve to
signal different degrees of certainty, and some could be used
exclusively for counterfactual propositions (Chao, 1968).
His claim, however, appeared to be based on word
etymology rather than linguistic data. Recently, Jiang (2000)
conducted an insightful linguistic analysis of potential
markers of counterfactuality in Chinese but denied all of
them on the logical ground that each form could potentially
be used in non-counterfactual contexts. In doing so,
however, Jiang (2000) may have raised the standard too
high. By the same criterion, English would not have any
linguistic markers for counterfactuals because the usual
marking device, the subjunctive mood, is also used in other
situations, too (Lycan, 2001).
In this study we adopt an information theoretic view of
linguistic marking. A linguistic form is a marker of
counterfactuality if its presence strongly predicts a
counterfactual interpretation of the sentence. This
probabilistic definition of linguistic markers is consistent
with the semantics of previous research.
Are there consistent and informative markers of
counterfactuality in Chinese? The research closest to
answering this question is Wu (1994), in which she asked
native Chinese speakers to identify counterfactual sentences
from newspapers and looked for recurring linguistic forms
in the sample. Although she was able to identify some
interesting patterns, these are not markers by our definition
because Wu’s approach yields the probability of using a
linguistic form if the sentence is counterfactual, whereas we
want the conditional probability in the opposite direction.
To this end, we constructed a large corpus of Chinese
sentences that follow certain linguistic forms, and calculated
the percentage of them judged to be counterfactual by naïve
native speakers of Chinese. We show that there are distinct
lexical and syntactic forms in the Chinese language that
strongly predict a counterfactual interpretation of the
sentence. Study 2 further demonstrates that these markers
are informative even after semantic information is
controlled. Together, Studies 1 and 2 strongly refute
Bloom’s (1981) basic premise that Chinese does not provide
linguistic signals of counterfactuality. The implication to the
Chinese counterfactual debate will be discussed.
Many signs of counterfactuals
Rather than focusing on accidental hits and misses of
language features, it is perhaps useful to reframe the issue of
language and thought, and ask the following question: if one
wants to communicate a counterfactual thought, is it
necessary to mark the counterfactualness linguistically? And
if yes, what linguistic forms are available to serve the
purpose? These language-universal questions may help to
advance the field of research from the current debate.
We argue that the answer to the first question is Yes,
because something has to be in the message to prevent a
literal reading of the message by the listener. Granted, the
signal does not have to be linguistic – a wink from the
speaker is sometime enough to trigger a non-literal
interpretation – but they are exceptions rather than the rule.
Linguistic marking of counterfactual readings may occur
at any level of linguistic analysis. Satires, for instance, are
often expressed using a marked prosody or an unusual
choice of words. Counterfactuality may also be marked
using special lexical entries, as Chao (1968) suggested. Two
additional possibilities require more discussion.
The first is syntax. The notion that counterfactuals are
expressed in the subjunctive mood in English is inaccurate.
Cross-linguistically, counterfactualness appears to be
associated changes in the aspect of the sentence as much as
the mood of the verb. If some sort of cognitive universals
are driving the apparent linguistic universals, one might
expect the Chinese language – which is aspectual – to
follow the same pattern.
Several authors pointed out that the marking of
counterfactuality may occur at the discourse/pragmatics
level. Bloom (1981) suggested that Chinese speakers could
reason counterfactually, but only be engaging in an odd
style of argument (“Everyone knows X is not Y. But let’s
suppose X is Y, then …”) that clearly violates basic
discourse principles. Yeh and Gentner (2005) distinguished
semantically “transparent” and “non-transparent”
counterfactuals. The former contains obvious contradictions
with the listener’s world knowledge (“if pigs can fly…”),
whereas the truth value of the non-transparent ones is
ambiguous from the sentence itself. Both groups of authors
imply that counterfactuals can be reliably marked by
exploiting discourse rules, common grounding, and world
knowledge alone. Without going into details, we would
argue that these channels are less reliable than lexical or
syntactic marking. In fact, counterfactual propositions in
natural language uses often contain multiple cues at
different levels.
Two predictions follow from the above analysis: (a)
linguistic marking of counterfactuals is a linguistic universal
and Chinese should be no exception; and (b) the actual
markers in Chinese may be diverse and numerous, and
markers at different linguistic levels may interact. Some of
these predictions are tested in the present study.
Linguistic marking and cognitive performance
Counterfactual markings would not be very interesting had
it not have any impact on cognition. The focus of the
analysis, however, is no longer direct comparisons between
Chinese and English speakers, if both languages
consistently mark counterfactual sentences linguistically.
We are particularly interested in two issues. First, when we
control the semantic and pragmatic information in a
sentence, do lexical and syntactic markers affect readers’
interpretation of the sentence? Yeh and Gentner (2005)
suggest that Chinese readers have trouble with non-
transparent counterfactuals, presumably because the lack of
lexical/syntactic marking. Without seeing the Chinese
sentences, it is unclear whether lexical/syntactic markers
were present in their Chinese material. Study 2 will examine
the strength of markers when the context is completely
uninformative.
Study 3 asks a different question – if linguistic markers
are strongly predictive of a counterfactual interpretation of
the sentence, do readers engage in a different process as
soon as they see the markers? A self-paced reading task was
used to investigate the time course of marker processing.
Study 1: Corpus Study
The purpose of this study is to identify linguistic markers of
Chinese counterfactuals. We define a marker as a consistent
linguistic form that strongly predicts a counterfactual universal hypothesis, they modify the aspect of the
interpretation of the sentence. counterfactual event and effectively turn the event into past
perfective.

Methods Study 2: Sentence Comprehension

The study was based on a list of potential counterfactual In real language use, semantic and linguistic cues work
markers in Chinese from Wu (1994) and other prior together. Study 2 examines whether those counterfactual
research. With those markers as keywords, a Chinese native markers identified in Study 1 are informative above and
speaker was asked to find 200 sentences containing those beyond contextual effects.
markers, either from an open-access Chinese text database
or from the internet search engines. Two other native Methods
Chinese speakers were asked to judge whether each
Participants
sentence is a counterfactual or an open conditional one.
The English part of the study was conducted in the United
Extensive training was provided to the coders so that they
were confident and consistent in making the categorization. States. Thirty undergraduate students at Duke University
participated in the English part of the study, where they
received course credit for participating. The Chinese part of
Results the study was conducted in Beijing, China. Thirty
undergraduate students at Peking University participated
Inter-coder reliability and they received payment for their participation.
The inter-coder reliability was calculated for the two raters.
They agreed with each other 86% of the time (percentage of
Materials
agreement=86%). After controlling the agreements that
would occur by chance, they agreed with each other 73% of 48 sentence frames were made so that the conditional clause
the time (Krippendorff’s alpha=73%). is not predictive of the interpretation of the main clause.
There were 4 conditions for English material: open
Informativeness of the markers conditional in present tense, counterfactual in present tense,
The percentage of sentences judged as counterfactual (out of open conditional in past tense, and counterfactual in past
200) was calculated for each marker. Table 1 shows the tense. Sentences are the same across conditions except for
average percentages of the two rates. the markers.
Similarly, sentence frames were created in Chinese and
Table 1: Percentages of counterfactual responses for each potential markers are inserted in the appropriate places in
marker the sentence. There were 6 conditions for Chinese material:
open conditional, open conditional with marker (“ba”) at the
category marker % CF end, counterfactual with aspect marker, counterfactual with
Temporal temporal reference, counterfactual with negator 1 (“yao4
早(early) 83% bu4 shi4”), and counterfactual with negator 2 (“mei2”).
reference
Sentences were the same across conditions except for the
Aspect marker 了 21%
markers.
要不是 (had it not been the
Negators: 91%
case) Procedure
没 (did not) 14% Participants were asked to read a sentence, and then judge
要不然 (had it not been the whether the next statement was true or false according to the
43%
case) meaning of that sentence they read before. They identified
Predicates 就好了(if only) 55% the likelihood of the statements being true on a scale of 0-
100%..
还以为… (thought) 91%
There was an additional task in the Chinese test. In this
原来应该 (should have done) 92% forced choice task, participants were asked to read a dialog
Others …的话 9% with a blank in it, and then to choose one sentence from a
真的 (really) 10% pair of sentences to fill in the blank. There were 2 kinds of
settings of the dialog: open setting where the correct answer
A number of lexical and syntactic markers of is an open conditional, and counterfactual setting where the
counterfactuals strongly predict (over 90%) a counterfactual correct answer is a counterfactual sentence. The pair of
interpretation of the sentence. These tend to be lexicalized sentences always consisted of an open conditional and a
phrases that serve as lexical signal of counterfactuality. counterfactual conditional.
They are also used in different contexts, as the gloss English
translations suggest. The “temporal reference” and “aspect
marker” categories (Wu, 1994) are the only syntactic
markers we were able to identify. Consistent with the
 Results Chinese: Forced-choice
English: Counterfactual rating As a measure of the strength of the marker, the Phi
coefficients were calculated for the data from this task. Phi
Figure 1 shows the percentage of English sentences judged is derived from Chi-square and is a measure of the
to be counterfactual for each condition. Having past tense associations between settings and sentence types. Table 2
subjunctive is a strong signal that the sentence should be shows phi values of each pair of comparison. Consistent
read as a counterfactual. However, the data on present tense with other tasks, having a lexical marker (the negators) or a
subjunctive is ambiguous. One possibility is that the lexical marker (Aspect marker or Temporal reference) in the
syntactic form for present tense subjunctive is similar to that sentence increase the readers’ interpretation of the sentence.
of the simple past tense and readers might have confused the
two. Alternatively, like Chinese speakers, English readers Table 2: Phi values of each pair of comparisons
may need both the past tense cue and the subjunctive mood
cue to trigger a counterfactual reading. comparisons phi values
OP2-AM 0.89**
OP1-TR 0.74**
80%
OP1-NG1 0.53**
70%
OP1-NG2 0.31*
60% Note: *p< .001, *p< .0001.
% of CF responses

50%
40% Study 3: Self-paced Reading
30% Study 3 investigates the time course of the effect of
counterfactual markers on on-line sentence comprehension.
20%
The counterfactual reading of a sentence should take more
10% processing resources and more time not only because the
0% reader has to process the linguistic markers so as to switch
OP_present CF_present OP_past CF_past to the non-default interpretation, but also because the reader
Figure 1: Percentage of judging English sentences as may need to inhibit the literal reading of the sentence. In the
counterfactual in each condition in Study 2 context of self-paced reading, it is predicted that
counterfactual readings of a sentence will take more time
then the default, literal reading.
Chinese: Counterfactual rating The processing load increase may begin as soon as the
reader identifies the marker for counterfactual conditionals.
Figure 2 shows the percentages of counterfactual responses The immediacy of the effect of the marker should be a
for each condition. The data are based on identical sentence function of the predictive strength of each marker,
frames that only differ by the markers. Even after semantic particularly in the Chinese study. On the other hand, a
information was controlled, lexical (negator 1) and syntactic reader may wait until the end of the sentence to integrate
(Aspect Marker and Temporal Reference) markers information.
significantly increased readers’ chance to read them as A caveat with the self-paced reading paradigm is that it
counterfactuals. may not be sensitive enough to identify a strategy change in
reading, and participants’ normal reading processes may be
altered by the one-word-at-a-time reading method.
100
Methods
80
Participants
60 Thirty undergraduate students at Duke University
participated in the English part of the study. The Chinese
%

40 study was conducted in Beijing, China. Thirty

undergraduate students at Peking University participated in
20 the Chinese part of the study. They received payment for
participating.
0
Materials
OP1 OP2 AM TR NG1 NG2
The same English and Chinese material in Study 2 was used
in Study 3.
Figure 2: Percentage of judging Chinese sentences as
counterfactual in each condition in Study 2
Procedure Chinese: Mean word reading time
Using a self-paced reading paradigm, Chinese and English
speakers were asked to read sentences word-by-word on a Figure 5 shows the mean reading time of each word in
computer screen. Reading time of each word was recorded. Chinese sentences in each condition. ANOVA F1 analysis
Participants were asked to read a sentence with self-paced showed there is no difference with mean word reading time
reading paradigm, and then judge whether a following across conditions in Chinese.
statement was true or false according to the meaning of that
sentence.
300
Results
English: Mean word reading time 250

reading time (ms)

Figure 3 shows the mean reading time of each word in 200
English sentences in each condition. ANOVA showed that
English speakers’ mean word reading time was significantly 150
longer for sentences in counterfactual condition in past tense
than in open condition in present tense ( F(3) =6.10, p< 100
0.05).
50

0
500
OP1 TR NG1 NG2
Mean word RT (ms)

400

300 Figure 5: Mean reading time of each word in Chinese

sentences by condition
200

100
Chinese: Reading time by region
0
OP_Present CF_present OP_past CF_past
Chinese sentences were divided into several regions, and the
reading time was averaged by region for each condition.
Figure 3. Mean reading time of each word in English Figure 6 shows the mean reading time by region in Chinese
sentences by condition sentences in each condition. ANOVA F1 showed that the
time spent at the end of the sentence was significantly
English: Reading time by region longer for sentences in the counterfactual condition with
English sentences were divided into several regions, and the temporal reference than in the open condition (F(1,
reading time per word was averaged by region for each 29)=10.05 p<0.005).
condition. Figure 4 shows the mean reading time by region
in English sentences in each condition. ANOVA did not
640
show any significant differences.
590 OP1
540 TR
1200 OP_present
CF_present 490 NG1
rt (ms)

1000
OP_past 440 NG2
CF_past
800
390
rt (ms)

600 340
400 290

200 240

0
190
IF subject1 verb1 object1 THEN subject2 verb2 object2 PERIOD subject m123 verb1 m45 verb2 period

Figure 4. Mean reading time by region for English sentences Figure 6. Mean reading time by region for Chinese
in each condition sentences in each condition
 Discussion References
Several key findings emerge form the three studies reported
here. First, it is possible to identify linguistic forms (lexical Au, T. K (1983) Chinese and English counterfactuals: The
or syntactic) that reliably predict a counterfactual reading of Sapir-Whorf hypothesis revisited. Cognition, 15, 155-
a Chinese sentence. These forms are shown to be 187.
informative in naturally occurring sentences in which the Au, T. K. (1984) Counterfactuals: In reply to Alfred
work with contextual cues to highlight counterfactuality. Bloom,Cognition, 17, 289-302.
They also independently signal a counterfactual reading Bloom, A. H. (1981) The Linguistic Shaping of Thought: A
when contextual information is controlled. Contrary to Study in the Impact of Language on Thinking in China
Bloom’s (1981) assertion, the Chinese language provides and the West. Hillsdale. Erlbaum Associates.
lexical and syntactical devices to mark counterfactuals. Chao, Yuen Ren. (1968). A Grammar of Spoken Chinese.
This finding nicely explains the puzzling finding in the
Berkeley & Los Angeles: University of California Press.
past two decades that Chinese speakers’ counterfactual
Jiang, Y. (2000). Counterfactual interpretations of Chinese
reasoning ability is on par with that of English speakers. No
substantial differences should be expected if both languages conditionals. [Studies and Explorations on Syntax
mark counterfactuality in the language itself. For Bloom (Chinese)], 10, 257-279.
(1981), our finding suggests that the logic of the study was Liu, L. G (1985). Reasoning counterfactually in Chinese:
false, and the cross-language differences he reported were Are there any obstacles? Cognition, 21, 239-270.
most likely due to translation and other technical problems, Yeh, D. & Gentner, D. (2005). Reasoning counterfactually
as pointed out by Au (1983). The criticism applies equally in Chinese: Picking up the pieces. Proceedings of the
to any study that accepted the false premise by Bloom. Twenty-seventh Annual Meeting of the Cognitive Science
Secondly, the study began to uncover some potential Society, pp. 2410-2415.
language-universals. For example, few people would have Wu (1994) If Triangle Were Circles,…” – A Study of
guessed before this study that Chinese uses the temporal Counterfactuals in Chinese and English. The Crane
(tense) and aspect markers to signify a counterfactual Publishing Co., Ltd, TW.
interpretation. The analogy with the English subjunctive
mood – which modifies the tense and aspect of the main
verbs – is obvious. More research, particularly cross-
linguistic, is needed to identify linguistic and cognitive
universals.
Last but not least, it is clear that counterfactual markers
affect how people understand counterfactual conditionals.
When there is no other useful information, readers of
Chinese can rely solely on the linguistic markers to solve
the problem. Message from the self-paced reading stud is
less clear-cut, but overall reading time was longer, and for
the Chinese a significant sentence-end wrap up effect is
observed. We are now in the process of conducting eye
movement experiments, where reader can move their eyes at
will and re-read sentences if necessary. The eye movement
technology is expect to yield rich information about the time
course of counterfactual processing, particularly the
immediacy of the marker effect.

Acknowledgments
This study was partially supported by a summer research
grant from the Asian/Pacific Studies Institute (APSI) at
Duke University. The authors also wish to acknowledge Dr.
Xiaolin Zhou and Dr. Xiangzhi Meng for assistance with
conducting the experiments in China.