Linking Meaning to Language: Verbs of Psychological State and the Linking Problem Joshua K. Hartshorne 1 Timothy J. O'Donnell 1 Yasutada Sudo 2 Miki Uruwashi 1 Jesse Snedeker 1 1 Department of Psychology, Harvard University, 33 Kirkland St., Cambridge, MA 02138 2 Department of Linguistics and Philosophy, Massachusetts Institute of Technology, 77 Massachusetts Avenue, 32-D808, Cambridge, MA 02139 Running Title: Linking meaning to language Send Correspondence to Joshua Hartshorne Massachusetts Institute of Technology 77 Massachusetts Avenue 46-4053H Cambridge, MA 02139 Email: [email protected]Tel: 617-253-3816 Acknowledgements: This work was presented at CogSci 2010 and benefited from comments there. The authors additionally thank David Pesetsky for comments and NDSEG (JH), the Allport Memorial Fund (JH), and the NSF (0921012; JS) for funding.
31
Embed
Linking Meaning to Language: Verbs of Psychological State and the Linking Problem
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Linking Meaning to Language: Verbs of Psychological State and the Linking Problem
Joshua K. Hartshorne1 Timothy J. O'Donnell1
Yasutada Sudo2
Miki Uruwashi1
Jesse Snedeker1
1Department of Psychology, Harvard University, 33 Kirkland St., Cambridge, MA 02138
2Department of Linguistics and Philosophy, Massachusetts Institute of Technology, 77 Massachusetts Avenue, 32-D808, Cambridge, MA 02139
Running Title: Linking meaning to language
Send Correspondence to
Joshua Hartshorne Massachusetts Institute of Technology 77 Massachusetts Avenue 46-4053H Cambridge, MA 02139
• We investigate intuitions about novel psych verbs in English and Japanese. • Speakers of both languages use semantics to infer syntactic structure. • Participants were more likely to map the experiencer of a long-lived state onto subject position
and a short-lived state onto object position. • However, we observed language-specific biases consistent with the distribution of verbs in each
language.
1. Introduction
1.1 The Linking Problem
To interpret the sentence Mary broke the vase, one must determine the kind of event described
(breaking), the participants in that event (Mary, the vase), and which participant played which role in the
event (Mary is the breaker, not breakee). A fundamental feature of natural languages is that the
mappings between the roles in an event and the positions in a syntactic structure are highly systematic
both within and between languages. For example, in English, the semantic agent of a causal event is
typically realized as the subject of sentence while the entity that undergoes a change of state is realized
as the object (Baker, 1988; Croft, 2012; Dowty, 1991; Levin & Rappaport Hovav, 2005; Tenny, 1994).1
Thus the breaker is the subject of break, the drier is the subject of dry and the liquidator is the subject of
liquidate. This generalization extends to novel words: Adults and children prefer an interpretation in
which The bear pilked the horse means the bear did something to the horse, not vice versa (Marantz,
1982; Pinker, 1989). This preference to map agents to subjects and patients to objects is present in
children as young as 20 months (Gertner et al., 2006). This is not an accidental fact about English; all
languages have systematic patterns for mapping the roles of causal events onto syntactic structures.
While many languages use the same mapping as English (agents are subjects and patients are objects),
other languages appear, on the surface at least, to reverse this pattern, with agents as objects and patients
as subjects (whether these “ergative” language truly reverse the linking patterns or only appear to do so
is a matter of some debate (Dixon, 1994; Levin, 1983, Levin & Rappaport Hovav, 2005; Marantz, 1984;
Pinker, 1989)). In both cases, the pattern is systematic and extends across a broad range of causal events.
These systematic mappings are central to theories of verb-argument structure. While there is
considerable debate about how the system is acquired and the inventory of semantic roles and syntactic
functions, theorists of all stripes acknowledge these systematic linkages and seek to explain them
1 For simplicity, here and throughout we focus on active verbs, though the theories cited do address passivized verbs as well. Discussions
of passives commonly involve syntactic movement or independent semantic and/or syntactic constructions.
Pesetsky, 1995; Pylkkanen, 1999), and are widely used within these languages.
Several attempts have been made to find systematic semantic distinctions between the two types,
thus explaining away the apparent difference in linking patterns, as was done for dative verbs and
locative verbs, as described in section 1.2 (Croft, 2012; Dowty, 1991; Grimshaw, 1990; Landau, 2010;
Pesetsky, 1995; Pylkkanen, 1999). For instance, Pesetsky (1995) argues that experiencer-object verbs
(frighten/anger/delight) construe events as causal, whereas experiencer-subject verbs (fear/hate/love) do
not encode cause and all. In contrast, Croft (2012) argues that both types of verbs are causal, with
experiencer-object verbs focusing on the causal role of the stimulus and experiencer-subject verbs
focusing on the causal role of the experiencer. Pylkkanen (1999) argues that the distinction is and how
verbs grammatically encode states and events (see below). Yet other researchers provide yet other
hypotheses.
The difficulty in finding a persuasive semantic distinction upon which all researchers can agree
has led some to express skepticism about the project itself. Culicover and Jackendoff (2005) write,
“There have been attempts to demonstrate a consistent semantic difference associated with these
configurations (e.g., Grimshaw 1990; Pesetsky 1995), but we find them unpersuasive when one
considers the full range of predicates in each configuration” (p. 184), and “Linking has to be specially
stipulated by either experiencer-subject verbs (regard, enjoy, like) or experiencer-object verbs (strike,
please, appeal to) or both” (p. 215) Moreover, two lines of psycholinguistic research have been used to
argue that experiencer-object verbs (frightened/angered/delighted) are exceptions to a general rule that
links experiencers to the subject position: Second-language learners appear to have particular difficulty
acquiring experiencer-object verbs relative to experiencer-subject verbs (Sato, 2003; White, et al., 1998;
Montrul, 2001), and agrammatic aphasics show worse performance on experiencer-object verbs
(Pinango, 2000; Thompson & Lee, 2009). (However, these findings are open to other interpretations; see
General Discussion). Critically, there is no existing experimental evidence that speakers perceive any
semantic distinction between these two kinds of events, or use this distinction to determine the correct
mapping for new verbs.
1.4 Overview of the Experiments
The present study reconsiders the psychological reality of one possible semantic difference
between experiencer-subject (fear-type) and experiencer-object (frighten-type) verbs. Specifically, we
teach naïve participants new psych-verbs that describe events that systematically differ in the likely
duration of the psychological state that they describe, and then we ask participants to decide how those
verbs would be used in a sentence. This manipulation is motivated by Pylkkanen’s (1999) observation
that in Finnish, situations described by experiencer-object verbs can be bound to a particular time and
place (John frightened/angered/delighted Mary yesterday in the kitchen), whereas those described by
experiencer-subject verbs cannot be (*John liked/loved/hated Mary yesterday in the kitchen).2 Rather,
the latter describe generic properties that are true over a long period of time. Interestingly, psychologists
have also made a distinction between emotions (fright/anger/delight), which are defined as brief
2 Specifically, she argues that the two classes can be distinguished by whether they encode stage-level or individual-level predicates in
the sense of Carlson (1977).
physiological states, and dispositions (love/liking/hatred), which are stable tendencies to feel a particular
way (Ekman, 1999). If participants perceive this distinction and use it to guide verb learning, then we
would expect them to systematically prefer the experiencer-subject linking pattern for long-lived states
and the experiencer-object pattern for short-lived states. In Experiments 1 and 4, this prediction is tested
in English.
Given that many linking patterns are argued to be consistent across languages (Baker, 1988;
Croft, 2012; Dowty, 1991; Marantz, 1984; Pinker, 1989) one may additionally wonder whether the
results we find in English will extend to other languages. As a first step in this direction, in Experiments
2 and 3 we investigate Japanese, which is historically unrelated to English and has very different
grammatical structure (Tsujimura, 1996). Importantly, its psych-verb system differs from that of English
in critical ways. Whereas transitive English psych-verbs are mostly experiencer-object (220 experiencer-
object vs. 44 experiencer-subject; Levin, 1993), most monomorphemic Japanese psych-verbs are
experiencer-subject (74 vs. 5 by our count). Additional morphologically-complex experiencer-object
verbs can be formed in Japanese by adding the causative –(s)ase- affix to an experiencer-subject verb:
(1) a. Taro-wa koomori-o kowagat-ta.
Taro-TOP bat-ACC fear-PAST
Taro feared bats.
b. Koomori-wa Taro-o kowagar-ase-ta.
bat-TOP Taro-ACC fear-CAUS-PAST
Bats frightened Taro.
Thus, in Japanese morphology is a highly reliable predictor of argument realization in psych-
verbs. If state duration nevertheless influences argument realization, this would provide striking
evidence that this aspect of the syntax-semantics interface is cross-linguistically robust.
2 Experiment 1: English
2.1 Method
We selected sixteen Japanese nouns that described psychological states for which there is no
verb in English. These nouns were turned into verbs, applying any necessary phonological
accommodations. Based on a description of the psychological state, participants were asked to choose
between an experiencer-subject (fear-type) or experiencer-object (frighten-type) usage of the verb:
(2) douyo: uneasiness.
a. Ken douyos the unexpected exam.
b. The unexpected exam doyous Ken.
The experiencer of the state was unambiguous: Only one argument of the verb was animate (e.g.,
Ken). Eight verbs described short-lived states (uneasiness) and were paired with short-lived STIMULI (the
unexpected exam). Eight described long-lived states (the feeling of rivalry) and were paired with long-
lived STIMULI (Harvard’s basketball team). Four additional filler sentences not involving psych-verbs
were included.
To test generality, verbs were presented in both present tense (Experiment 1a) and past tense
(Experiment 1b). As experiencer-subject verbs cannot naturally used in the progressive form in English
(*John was fearing bats), we used simple tenses only.3 The order of verbs was pseudo-randomized such
that the same condition (short-lived/long-lived) did not occur more than twice in a row. Four test forms
were created for each experiment by counterbalancing the order of stimuli (forwards/backwards) and the
order of the sentence pairs. Five English-speaking adults participated in each test form, for a total of 20
in Experiment 1a and 20 in Experiment 1b.
2.2 Results and Discussion
As Figure 2 indicates, participants were more likely to choose the experiencer-object form for 3 There appears to be a trend in colloquial English to allow such forms (cf McDonald’s “I’m loving it” campaign). This usage is marked
and appears to invoke a different meaning (compare John loves his job with John is loving his job) and thus was avoided.
short-lived events than for long-lived events, in both Experiment 1a (M=68%, SE=9% vs. M=38%,
SE=7%, d=1.4) and 1b (M=67%, SE=9% vs. M=41%, SE=9%, d=1.0).4 The main effect of state
duration was significant (F1(1,38)=60.8, p<.001; F2(1,14)=6.2, p=.03), but the main effect of tense was
not (Fs<1) nor was the interaction of tense and duration (Fs<1). Thus, English-speakers use the
semantics of psych-verbs to guide learning of new psych-verbs.
Fig. 2. Percentage of participants choosing the experiencer-object form for each verb for English verbs
(Exp. 1), monomorphemic Japanese verbs (Exp. 2), and causally-affixed Japanese verbs (Exp. 3).
3 Experiment 2: Monomorphemic Japanese Psych-Verbs
3.1 Overview
Having found evidence that the duration of the psychological state guides expectations about
syntactic form in English, we tested the same contrast in Japanese using monomorphemic verbs. As
discussed in the Introduction, nearly all monomorphemic psych-verbs in Japanese are experiencer-
4 Because generality of the effect across items is of core interest, means, standard errors and Cohen’s d are here and elsewhere calculated
by items. ANOVAs are calculated by both subjects and items.
0%
20%
40%
60%
80%
100%
% c
hosi
ng e
xper
ienc
er-o
bjec
t for
m
short long short long short long short long short longEnglish Present English Past Japanese Present Japanese Past Japanese Causal
Experiment 1 Experiment 2 Experiment 3
* *
*
*
*
subject. Thus, if Japanese participants are influenced by the distribution of forms in their language, they
should expect novel monomorphemic psych-verbs to be experiencer-object.
3.2 Method
Experiment 2 followed the same procedure as Experiment 1. We selected sixteen English nouns
describing psychological states for which there is no verb in Japanese. To turn these nouns into verbs,
we created loanwords using the semantically neutral, semi-productive verbalizer -r- (e.g., gugu-r-u: ‘to
google’) and made any necessary phonological accommodations. Again, eight states were long-lived
(greed) and eight were short-lived (jolt). An example trial is given below:
(4) guriifu (jolt): a surprise or shock, esp. of an unpleasant kind and often manifested physically
a. Sono keeji-wa sono koroshi-no genba-o joruto-t-ei-ta
That detective-TOP that murder-GEN scene-ACC jolt-V-PROG-PAST
The detective jolted the scene of the murder.
b. Sono koroshi-no genba-wa sono keeji-o joruto-t-ei-ta
That murder-GEN scene-TOP that detective-ACC jolt-V-PROG-PASS
The scene of the murder jolted the detective.
The four filler verbs (2 experiencer-subject) were existing English-derived psych-verbs formed with the
light verb -suru. All verbs were presented with progressive morphology, which was judged by native
speakers (YS & MU) as the most natural-sounding form. Experiment 2 was identical to Experiment 1 in
all other respects. Japanese-speaking adults were recruited in public spaces around Tokyo. Twenty
participated in the present-tense version (2a), and twenty in the past-tense version (2b). Two of the filler
verbs in 2a were problematic and were replaced in 2b.
3.3 Results and Discussion
Like English speakers, Japanese participants (Figure 2) were more likely to select the
experiencer-object form for the short-lived verbs than the long-lived verbs in both Experiments 2a
(M=29%, SE=3% vs. 9%, M=3%, d=2.3) and 2b (M=44%, SE=6% vs. M=27%, SE=4%, d=1.2). There
were significant main effects of state duration (F1(1,38)=28.6, p<.001; F2(1,14)=16.8, p=.002) and
tense (F1(1,38)=6.3, p=.02; F2(1,14)=21.5, p<.001), but the interaction was not significant (Fs<1).
Thus, Japanese-speakers, like English-speakers, utilize state duration when predicting the linking of
semantics to syntax for new psych-verbs. However, comparison of Experiments 1 and 2 reveals an
interesting cross-linguistic difference. As depicted in Figure 1, Japanese-speakers were across-the-board
less likely than English-speakers to choose the experiencer-object form for novel psych-verbs (M=28%,
SE=4%; M=53%, SE=7% vs.; t1(78)=5.8, p<.001; t2(30)=3.5, p=.001; d=1.2), suggesting a role for a
language-specific biases (note that Japanese has fewer monomorphemic experiencer-object verbs than
English does).
4 Experiment 3: Causative-Affixed Japanese Psych-Verbs
4.1 Overview
As noted above, while both English and Japanese speakers were significantly more likely to
choose experience-object (frighten-type) uses of novel psych-verbs when the duration of the described
mental state was short, they showed different baseline preference for the two forms. This may be due to
differences in kind of monomorphemic psych-verbs that already exist in each language and the
availability of morphological processes for building new psych-verbs. Monomorphemic psych-verbs in
Japanese are overwhelmingly experiencer-subject (fear-type); however, Japanese also has a highly
productive causative marker which, when used to form psych predicates, produces uniformly
experiencer-object (frighten-type) verbs. If the differences we observed between Experiment 1 and 2 are
due to language-specific knowledge about how morphological form relates to linking patterns, then we
should expect Japanese-speakers to prefer the experiencer-object form for affixed psych-verbs. If the
differences are due to inadvertent differences in the stimuli or a general preference for experiencer-
subject verbs on the part of Japanese-speakers, morphological structure should have no effect.
4.2 Method
The materials and procedure were identical to Experiment 2a, except that all verbs were
causativized by the addition of the -(s)ase- affix and that the (better) filler verbs from Experiment 2b
were used. Twenty Japanese-speaking adults recruited in public spaces around Tokyo participated.
4.3 Results and Discussion
Japanese-speakers showed a strong bias to choose the experiencer-object forms (M=73%,
SE=3% vs. M=19%, SE=3%), resulting in a reliable difference between Experiment 3 and Experiment
2a (t1(38)=10.3, p<.001; t2(15)=20.5, p<.001; d=4.5). Nevertheless, participants preferred the
experiencer-object form even more for the short-lived states than the long-lived states (Figure 2;
M=79%, SE=2% vs. M=67%, SE=3%; t1(19)=2.41, p=.03; t2(14)=2.83, p=.01; d=1.4) suggesting that
even in this strongly constrained morphological context, meaning played a role in preferred argument
realization patterns.
5 Experiment 4: States and Stimuli
5.1 Overview
Before discussing the above results, we address one final question about the source of the short-
lived/long-lived distinction presented above. In Experiments 1-3, we biased participants to interpret
novel verbs and short-lived or long-lived both with the definition of the psychological state (uneasiness
vs. the feeling of rivalry) and the longevity of the inanimate stimulus (the unexpected exam vs.
Harvard’s basketball team). There are two ways in which these stimuli could affect judgments. First,
short-lived and long-lived stimuli may reinforce the interpretation of the psychological states as short- or
long-lived. This was our original intention and would strengthen the above conclusions. However, it is
also possible that participants used linking-patterns that mapped particular kinds of noun-phrases to
subject or object position, entirely ignoring the verb’s meaning. In Experiment 4, conducted in English,
we test whether the psychological state definitions by themselves are sufficient to guide psych-verb
learning by using a slight variant of the task in which both the experiencer and stimulus of the
psychological state were humans.
5.2 Method
The 16 definitions of psychological states from Experiment 1 were used. A separate group of 16
English-speaking participants recruited through Amazon Mechanical Turk rated each state (in lists
counterbalanced by order) according to how long it would likely last: seconds, minutes, hours, days,
weeks, months or years. Based on these ratings, the stimuli were divided into eight short-lived and eight
long-lived states, resulting in the re-classification of two of the items.
Participants were introduced to a novel character “Susan” who has many emotional relationships
with friends. For each friend, participants were told Susan that experienced one of the 16 psychological
states. Participants were asked to produce a three-word sentence using the novel verb that described this
state and used both character’s names. All verbs were presented in the past tense. The two counter-
balanced orders from Experiment 1 were used; fillers were not included.
Forty English-speaking US residents were recruited through Amazon Mechanical Turk. An
additional 12 were excluded for failing to follow directions or for reporting dyslexia.
5.3 Results and Discussion
Once again participants were more likely to link the experiencer (Susan) with object position for
short-lived relative to long-lived states (Figure 3; M=52%, SE=6% vs. M=24%, SE=6%; t1(39)=7.20,
p<.001; t2(14)=3.25, p=.01; d=1.6). These results confirm that the perceived length of the psychological
state played a role in the choice of linking rule. While the inanimate arguments used in Experiments 1-3
may have made additional contributions, the results of Experiment 4 show that the expected duration of
the psychological state alone is sufficient to influence argument realization.
Fig. 3. Percentage of participants choosing the experiencer-object form for each verb in English (Exp.
4).
6 General Discussion
6.1 Summary of Results
To communicate about events and states, speakers must map participants onto syntactic roles
(such subject or object). To the extent that these linking patterns are generalizable, they can facilitate the
acquisition (and creation) of new verbs, and, to the extent that these generalizations are available to
infants, they can guide the acquisition of syntax (Pinker, 1984) and verb meaning (Gleitman, 1990). In
contrast, to the extent that the mappings from semantics to syntax are unpredictable or riddled with
exceptions, learners will have to acquire knowledge of these linking patterns through their encounters
with individual verbs in contexts where the relevant aspects of meaning and form are themselves
transparent (Tomasello, 1992). The primary evidence against broad linking rules comes from cases
where a single event can be described by sentences in which a given participant is mapped onto distinct
syntactic positions. In this paper, we examined one of the most widely-discussed cases of this kind:
psych-verbs, which exhibit two distinct linking-patterns, either linking the EXPERIENCER of the
psychological state to subject position and the STIMULUS of the state to object position (fear/hate/love) or
vice versa (frighten/anger/delight). Following Pylkkanen (1999), we propose that this distinction in
linking-patterns mirrors a distinction in verb semantics: experiencer-subject verbs describe long-lived
states; experiencer-object verbs describe short-lived states.
Across four experiments, we find that this distinction is psychologically real and guides the use
of novel verbs. These effects were robust, exhibiting large effect sizes (Cohen’s ds ≥ 1) across different
verb tenses and morphological forms. The effect was found in both English (Exps. 1 & 4) and Japanese
(Exps. 2-3) and was suggested by an analysis of existing Finnish verbs (Pylkkanen, 1999).
Critically, the presence of this effect in Japanese strongly suggests that it does not arise by a
simple analogy between the novel verb and known verbs. In Japanese, there are no causative
experiencer-subject verbs and few monomorphemic experiencer-object verbs. Nevertheless, when
confronted with a causative verb that described a long-lived psychological state or monomorphemic
verb describing a short-lived state, about 30% of the time our Japanese participants overrode the
language-specific morphological cues and followed the semantic generalization. The presence of this
pattern in two typologically unrelated languages raises the possibility that the pattern is cross-
linguistically robust, and perhaps even universal. Further research on the lexicalization of novel psych-
verbs in a variety is needed to explore this possibility.
These findings may seem in conflict with previous psycholinguistic studies of second-language
learners and agrammatic aphasics, which appeared to suggest that experiencer-object verbs are treated as
have argued that experiencer-object verbs explicitly mark the STIMULUS argument as the cause of the
psychological state. Similarly, at least several languages that employ causal morphemes (like Finnish
and Japanese) often use them to construct experiencer-object verbs (cf Pesetsky, 1995). Causes are
robustly linked with subject position across languages (Baker, 1988; Pinker, 1984; Levin & Rappaport
Hovav, 2005), which would account for linking-pattern of these verbs. In contrast, on this theory
experiencer-subject verbs do not express the cause but simply assert that the experiencer is having an
emotional state with a particular target or content, thus a default linking rule linking animate entities to
subject position may apply. This proposal is supported by the fact that in languages with causally-
affixed psych-verbs (like Japanese), these verbs are experiencer-object.
Thus, people may prefer to encode short-lived psychological states as experiencer-object verbs
because they wish to talk about the change of state (and thus link experiencer to object position), they
wish to talk about the cause of the state (and thus map the stimulus onto subject position), or both. In
either case the semantic distinction underlying these effects is one relevant to a broad range of
predicates.
6.3 Variation within and between languages
Although speakers of both languages were sensitive to psychological state duration, English-
speakers were far more likely than Japanese-speakers to choose the experiencer-object form for
monomorphemic verbs. This effect is large and consistent across items (d=1.2), suggesting that small,
accidental variations in items are not to blame. All but one of the short-lived English verbs in
Experiment 1a had more experiencer-object attributions than any of the short-lived Japanese verbs in
Experiment 2a and all but one of the long-lived English verbs in 1a had more experiencer-object
attributions than any of the long-lived Japanese verbs in 2a.
We suspect that this pattern reflects the difference in the morphological structure of psych-verbs
in these languages. As noted earlier, in Japanese most monomorphemic psych-verbs have experiencer
subjects, while experiencer-object verbs are created from experiencer-subject verbs by applying the –
(s)ase affix. Our participants clearly knew this and applied this knowledge in our task: verbs with the
causative affix were expressed with experiencer-objects twice as often as monomorphemic verbs.
It may be that when Japanese speakers encounter a causally-affixed verb, they infer that it
encodes a caused change-of-state, even if the event is fairly long in duration, resulting in semantics
appropriate for an experiencer-object verb. In contrast, when they encounter a monomorphemic verb,
they infer that it does not describe a caused change-of-state (or else it would have the causative
morpheme), resulting in semantics appropriate for an experiencer-subject verb. Because English does
not explicitly mark verbs as causal or not, the English-speaking participants were more open to both
possible interpretations of the novel verb. This hypothesis would account for cross-lingustic variation
while remaining consistent with a cross-linguistically universal linking bias (for a parallel account of
cross-linguistic differences in object construal, see Barner et al., 2010; Li et al., 2009).
Alternatively, it is possible that English speakers and Japanese speakers conceived of the verbs
in the preceding experiments in the same way, and shared a bias to encode long-lived states as
experiencer-subject verbs, but these cross-linguistic linking patterns had been modified by differences in
their linguistic experience. Japanese-speakers may have learned that monomorphemic psych-verbs are
likely to have EXPERIENCER subjects (regardless of whether they encode a change-of-state), while psych-
verbs with the causative morpheme typically have EXPERIENCER objects. English-speakers have a less
pronounced bias because existing English psych-verbs – all of which are monomorphemic5 – contain
substantial numbers of both experiencer-subject and experiencer-object verbs (see recent work on
implementing overhypotheses – hypotheses about hypotheses – in hierarchical Bayesian models; e.g.,
Perfors et al., 2010).
While this final proposal gets the direction of results across experiments correct, one would have
predicted that English participants would show an overall bias towards the experiencer-object form,
given that that form comprises the majority of English psych-verbs. Instead, English-speaking
participants chose the experiencer-object form 54% of the time in Experiment 1 and 38% of the time in
Experiment 4. This could be evidence against the above proposal. Alternatively, it may be (a) due to the
choice of stimuli (the long-lived stimuli made better exemplars of experiencer-subject verbs than the 5 Historically, English had several causally-affixed psych-verbs. Frighten is historically derived from the adjective frighten and a semi-
productive causal morpheme –en. In contemporary English, such verbs are exceedingly rare, and –en is at best minimally productive, suggesting that this plays little role in modern linguistic intuitions.
short-lived stimuli made experiencer-object verbs), or; (b) due to some additional bias, such as a bias to
put necessarily-animate arguments in subject position (see Paczynski & Kuperberg, 2011, for a review).
6.4 Cross-Linguistic Consistency and Acquisition of the Linking Rules
Because our participants were linguistically competent adults, our experiments do not provide
direct evidence about the origins of these syntactic expectations. Below we consider how this bias might
arise during acquisition.
First, this bias may reflect innate linking patterns. Second, it may be that children come to the
problem of verb-learning with broad semantic categories of verbs/events (e.g., short-lived, caused
changes of emotional state) already in place, and only the linking patterns must be learned (for related
discussion, see Levin, 1993; Pinker, 1984, 1989). Finally, it may be that both semantic categories and
linking patterns arise as generalizations during language acquisition (Bowerman, 1990; Goldberg, 1995;
Tomasello, 1992, 2003).
These three accounts interact with the partially orthogonal issue of cross-linguistic stability. The
pattern we describe above was seen in English and Japanese and inspired by previous work in Finnish
(Pylkkanen, 1999). This is predicted – and, in fact, required – by the first account (innate linking
patterns); neither of the latter two accounts directly predict this result, at least not by themselves.
What additional considerations could lead the pattern to be consistent across English, Japanese,
and Finnish? One possibility is simple chance, though given the historical distance between the three
languages, this seems unlikely. Nonetheless, it should be tested by investigating additional languages. A
second possibility is that that certain cognitive constructs are more salient than others (e.g., causes), and
certain syntactic positions are more prominent than others (e.g., subjects), and there is a preference to
map salient cognitive constructs onto prominent syntactic positions. This could lead to the pattern
observed without requiring that either semantic roles or syntactic positions are innate (i.e., consistent
with accounts 2-3, above). Finally, it may be that many or all languages share the same structure because
of communicative constraints. That is, languages that map short-lived psychological states onto
experiencer-object verbs are more efficient at communication (cf Chater & Christiansen, 2010). The
challenge for this proposal is to show that such languages truly are superior under some defensible
definition of “efficiency”.
7 Conclusions
Some verbs can be used to describe the same event but undergo different linking patterns
(fear/frighten, send/receive, chase/flee). These cases present problems for theories on which linking
patterns are highly systematic. Previous work on the dative alternation (John gave Mary a book vs. John
gave a book to Mary) and the locative alternation (Mary loaded the wagon with hay vs. Mary loaded hay
onto the wagon) has suggested that variation in syntactic encoding reflects fine-grained differences in
semantics: The different forms encode subtly different semantic construals of the event, each of which is
mapped to syntax using a consistent set of linking rules (Ambridge, Pine & Rowland, in press;
Ambridge, Pine, Rowland & Chang, in press; Gropen et al., 1991; Gropen et al., 1989).6
In the present study, we show that the finer-grained semantics approach can also account for psych-
verbs, a particularly intractable test case which some have argued would require lexically encoded