ArticlePDF Available

It’s not Just the “Heavy NP”: Relative Phrase Length Modulates the Production of Heavy-NP Shift


Abstract and Figures

Heavy-NP shift is the tendency for speakers to place long direct object phrases at the end of a clause rather than next to the verb. Though some analyses have focused on length of the direct object phrase alone, results from two experiments demonstrate that the length of the direct object relative to that of other phrases, and not the length of the direct object alone, predicts production of the shifted structure. These data support an accessibility-based interpretation of length effects in word order emerging from incremental production processes, in which longer phrases tend to be less easily planned and therefore are delayed during utterance planning.
Content may be subject to copyright.
It’s not Just the “Heavy NP”: Relative Phrase Length Modulates
the Production of Heavy-NP Shift
Lynne M. Stallings and
Department of English, Ball State University, Muncie, IN 47306, USA
Maryellen C. MacDonald
Department of Psychology, University of Wisconsin-Madison, Madison, WI 53706, USA
Lynne M. Stallings:; Maryellen C. MacDonald:
Heavy-NP shift is the tendency for speakers to place long direct object phrases at the end of a
clause rather than next to the verb. Though some analyses have focused on length of the direct
object phrase alone, results from two experiments demonstrate that the length of the direct object
relative to that of other phrases, and not the length of the direct object alone, predicts production
of the shifted structure. These data support an accessibility-based interpretation of length effects in
word order emerging from incremental production processes, in which longer phrases tend to be
less easily planned and therefore are delayed during utterance planning.
Sentence production; Language production; Grammatical encoding; Heavy-NP shift; Constituent
Direct object noun phrases (NPs) typically are uttered just after their verbs in English, as in
Emma explained the regulations to Jim, where the direct object the regulations appears
immediately following explained. Other word orders (e.g., Emma explained to Jim the
regulations) seem ungrammatical or awkward to English speakers. When the direct object
becomes very long or “heavy,” however, it may optionally be uttered in clause-final
position, as in Emma explained to Jim all of the regulations regarding import and export
taxes for pottery. This additional phrase order option, often called “Heavy-NP Shift,” has
received extensive attention in linguistics (e.g., Givón 1988; Hawkins 1994; Ross 1967;
Wasow 1997; Zec and Inkelas 1990). In keeping with this literature, we will identify the late
appearance of a direct object NP as “shifting,” though our use of this term does not imply
that a “shifted” NP is literally moved during stages of production planning.
In contrast to its extensive treatment in linguistics, there has been relatively little attention to
heavy-NP shift in the psycholinguistics literature. As a result, comparatively little is known
about the comprehension of shifted structures (Staub et al. 2006), the utterance
environments that permit the option of Heavy-NP shift (Stallings et al. 1998; Yamashita and
Chang 2001) or the relationship of Heavy-NP shift to other length-based choices during
production (Arnold et al. 2000; McDonald et al. 1993; Wasow and Arnold 2003). Two
Correspondence to: Maryellen C. MacDonald,
NIH Public Access
Author Manuscript
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
Published in final edited form as:
J Psycholinguist Res
. 2011 June ; 40(3): 177–187. doi:10.1007/s10936-010-9163-x.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
important results have emerged, however. First, phrase length in number of words appears to
affect Heavy-NP Shift and other length-based syntactic structure choices, while word-
internal length, such as the number of syllables, does not (McDonald et al. 1993; Stallings et
al. 1998). The reason for this contrast appears to reside in the sequence of operations
executed during utterance planning, in which grammatical structure is assigned at a planning
stage that is largely completed before phonological encoding processes commence, so that
word length information is not available when syntactic decisions are made (Bock and
Levelt 1994; McDonald et al. 1993). The unavailability of word length information during
such stages is different from later planning stages, where word length does affect word
order. Thus, ordered lists or conjunctions within phrases, where shorter words tend to
precede longer ones, as in to love and to cherish, and planes, trains, and automobiles, utilize
a linearization process within phrases that appears to be at a later planning stage where word
length information can influence planning (McDonald et al. 1993).
A second finding is that, despite its name, the production of heavy-NP shift does not stem
from properties of the NP alone. Both experimental studies (Stallings et al. 1998) and corpus
analyses (Wasow 1997) have shown that the rate of shifting varies with the type of verb
uttered, and verb properties also affect the ease of comprehending these structures (Staub et
al. 2006). These results suggest that during the production process, several factors may
conspire to yield greater or lesser rates of shifting in different circumstances, even with
object NP length controlled. If so, the study of heavy-NP shift could shed light on the
processes of grammatical encoding, the stage of language production in which producers
translate a message into a syntactic structure (Bock and Levelt 1994). As Stallings et al.
(1998) noted, shifted versus unshifted structures provide an interesting contrast from several
other syntactic alternations that have been investigated in the production literature, because
in heavy-NP shift, NPs retain the same grammatical roles independent of their phrase order
—the direct object retains the same grammatical role no matter its location in the sentence.
By contrast, the passive/active alternation affects both word order and grammatical role; for
example, the NP the regulations has the role of direct object in the active sentence Emma
explained the regulations and the role of subject in the passive sentence The regulations
were explained by Emma. Contrasts between role-preserving and role-changing structures
have been important in the development of theories of grammatical encoding (McDonald et
al. 1993), and additional study of heavy-NP shift may similarly be informative.
Hawkins (1994) reported findings from a small corpus analysis that suggested that the rate
of heavy-NP shifting was better predicted by “relative weight,” the length (in words) of the
direct object NP relative to that of the other material in the verb phrase (e.g., a prepositional
phrase such as to Jim), than by the length of the object NP alone. Several researchers have
suggested that if speakers could weigh the relative length of the object NP versus other
material during utterance planning, it would reduce processing load for the speaker. There is
extensive evidence that production planning is incremental, such that speakers plan and
begin to utter initial parts of an utterance while planning later portions of the utterance
(Bock and Levelt 1994; Bock and Warren 1985; De Smedt 1994). Production fluency is
maximized by placing more easily-planned or accessible sentence elements earlier in the
sentence, so that these can be uttered while planning of more difficult material is underway.
If shorter phrases tend to be more accessible, by virtue of having fewer words to plan, then
they should tend to be placed earlier in the utterance plan, yielding more time to continue
planning the longer, less accessible phrases (Bock and Levelt 1994; Chang 2009; De Smedt
1994; Wasow 1997). On this view, in which length is one factor modulating accessibility,
and accessibility in turn shapes phrase ordering, it makes sense that the relative length of
phrases, rather than the absolute length of only one element, would modulate the relative
accessibility and thus phrase ordering in English. The available evidence on relative length
stems from corpus analyses (Hawkins 1994; Wasow 1997) and judgments of acceptability
Stallings and MacDonald Page 2
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
(Wasow and Arnold 2003), and is generally consistent with this relative length-accessibility
What is lacking, however, is an experimental test of the role of relative phrase length in
production. Production experiments provide an important complement to other assessments
of production processes such as acceptability judgment studies, which may reflect
comprehension mechanisms more than production processes. Laboratory production studies
also complement corpus analyses: studies of spoken or written corpora provide a naturalistic
sample of utterances, but they also necessarily afford less control over the production
environment than in a laboratory experiment, in that the element of interest (here, syntactic
choices) may be confounded with other factors such as the topic being discussed, the
linguistic genre(s) sampled in the corpus, and other factors. The interrelationships between
factors such as these are ultimately critical to a full account of language use, but
psycholinguistic theories of production mechanisms greatly benefit from experimental
manipulations of a small number of factors and holding others constant. We therefore
conducted two well-controlled production experiments investigating the relative length of
the direct object NP and a verb-modifying prepositional phrase (PP) in order to investigate
whether relative length of these constituents affects speakers’ structure choices during
language production.
Experiment 1: Manipulating PP Length
In this study, experimental participants constructed sentences from phrases appearing on a
computer screen, with the dependent variable being the rate of sentences with heavy-NP
shift structures (Verb PP object-NP) versus the more common Verb NP PP order. A
modified recall-based production method was used (McDonald et al. 1993; Race and
MacDonald 2003; Stallings et al. 1998; Yamashita and Chang 2001). Recall-based
production tasks have become an important method in studies of sentence production,
following Lombardi and Potter’s (1998) findings that sentence recall is driven by speakers’
regeneration of the sentence from the remembered conceptual message, so that structures
produced in recall show similar patterns to those generated in other production
environments. In particular, Stallings et al. (1998) showed that the recall methodology used
here yielded patterns of heavy-NP shift that closely corresponded to people’s judgments of
the acceptability of shifted and unshifted structures.
To investigate the effect of relative length, we examined how shifting rates changed for NPs
of a fixed length, as a function of PP length. All experimental items contained direct object
NPs of 10 words in length, and PP length was manipulated at three levels (2, 5, and 7
words). If NP length alone affects shifting, then shifting rates should be constant across all
PP length conditions. However, if relative length guides behavior, then shifting should be
more common with shorter PP phrases than with longer ones.
Participants—Thirty-six undergraduates were either paid or received extra credit in
psychology courses for their participation. All were native speakers of English.
Materials—Thirty experimental items were constructed, each consisting of three
components: a subject-verb phrase, a 10-word noun phrase (NP), and short (2-word), mid
(5-word), and long (7-word) prepositional phrases (PP) (see Table 1 for examples). The
subject-verb phrases were 3–4 words in length (e.g., The dancer realized, The radio listeners
accepted) and did not include any alternating dative verbs such as give, because these verbs
can participate in phrase orders other than those that were under investigation here. The
nouns in the 10-word NPs were modified by only adjectives and prepositional phrases and
Stallings and MacDonald Page 3
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
excluded other structures such as relative clauses. At all three lengths, the PPs expressed
Time, Location or Manner roles modifying the verb, so that the PPs could not be interpreted
as modifying the direct object NP. Ten practice and thirty filler items were similarly
constructed with subject-verb and object NP phrases (of varying lengths) and PPs or other
non-object phrases.
Design and Procedure—The experiment used a constrained production paradigm that
Stallings et al. (1998) had previously used to elicit heavy-NP shift structures, in which
participants planned sentences using phrases presented on a computer screen and recalled
them after a brief delay. The PP length variable was manipulated within subjects with 10
trials each at the short, mid, and long PP conditions.
In each trial, the subject-verb phrase appeared in the center of the screen and the direct
object NP and PP appeared above and below this phrase. Six lists were prepared to
counterbalance the top versus bottom screen position of the NP and PP phrases and to
counterbalance PP length across items.
Participants were instructed to read the phrases, arrange them in a sensible order, and
prepare to produce the resulting sentence from memory when cued to speak. They were
informed that sentences should always start with the words in the middle of the screen.
All 10 practice and 20 of the 30 filler items were constructed so that only one ordering of the
top and bottom phrases yielded a grammatical sentence, to discourage participants from
perseverating on a single phrase order pattern. The other 10 filler items consisted of a two-
word PP and a two-word NP to provide variety of NP lengths. To counteract a tendency,
observed in pilot work, to favor the phrase order MIDDLE (of screen)-BOTTOM-TOP,
thirteen (65%) of the 20 fixed-order filler items were grammatical only in the phrase order
At the start of each trial, three left-justified phrases appeared at their assigned screen
locations. Participants pressed a key to indicate when they were ready to begin speaking.
This keypress was followed by a 1 s blank screen interval, which was in turn followed by
the reappearance of the subject-verb phrase on screen, serving as a prompt to begin uttering
the entire sentence as accurately as possible. As soon as participants began speaking, the
prompt disappeared, and the screen remained blank through the rest of the trial.
Following the 10 practice items, the experimental and filler trials were presented in random
order. The sessions were tape-recorded for later transcription and lasted between 30 and 45
Results and Discussion
Participants’ utterances were transcribed and scored for both the structure uttered and the
number of words produced in the NP and PP components. Trials were excluded when
participants’ “ready to speak” response times were less than 500 ms (1.9%), and when
participants failed to utter both an NP and PP phrase (6.7% of the trials).
Participants’ responses were not scored for accuracy of recall of the exact words presented
onscreen but were scored for the number of words uttered and the syntactic structure of the
sentence. Participants often made slight changes to the words in their utterances, similar to
the patterns in Stallings et al. (1998) and consistent with previous research showing that
sentence recall is not just verbatim repetition from phonological memory but stems from
actual production planning from the remembered meaning (Lombardi and Potter 1998).
Since the length of participants’ utterances is critical for analyses of relative length effects, it
Stallings and MacDonald Page 4
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
is important to establish the length of the phrases actually uttered when considering the role
of relative length on structure choice. Coding of participants’ productions showed that even
though participants were not completely accurate, their utterances mirrored the relative
length of the three conditions quite well. Participants produced an average of 8.18 words
(SD = 1.45) for the 10-word NPs. For the three PP conditions, they produced on average
2.08 words (SD = 0.23) for the short (2-word) PPs, 4.33 words (SD = 0.76) for the mid (5-
word) PPs, and 5.14 words (SD = 1.34) in the long (7-word) PP condition. Thus, on average,
trials in the Short PP condition had an NP-PP length difference of 6.10 words, trials in the
Mid PP condition had an average NP-PP difference of 3.85 words, and Long PP trials had an
average of 3.04 words. These mean length differences did not change as a function of
whether speakers produced the shifted or basic word order, Fs < 1. These results suggest that
though speakers were not perfect in their recall of the phrases, their utterances can be used to
investigate the effects of relative length on shifting.
An analysis of shifting and screen position revealed a somewhat greater tendency for
participants to produce shifted utterances when the PP appeared at the top of the screen
(36.99%) than at the bottom (29.68%); F 1(1, 35) = 4.756, p < .05; F 2(1, 29) = 6.479, p < .
05. However, screen position did not interact with PP length, Fs < 1, and will not be
discussed further.
Phrase Order Choice—As shown in Fig. 1, there were effects of PP length on the rate of
shifted sentences uttered, such that as the PP grew longer (and relative length between the
PP and NP decreased), the rate of shifted utterances declined, F 1(2, 70) = 3.11, p = .05; F
2(2, 58) = 3.31, p < .05. Participants shifted reliably more in the short (2-word) PP condition
(38.7% shifted utterances) than in the long (7-word) PP condition (29.7% shifting) F 1(1,
35) = 4.22, p < .05; F 2(1, 29) = 4.56, p < .05. The mid PP condition (5-word, 37% shifting)
was numerically in between these two conditions in shifting rate and did not differ reliably
from either the short or the long PP condition.
These results show that the same 10-word NPs were produced in shifted structures as a
function of the length of the PP. This result means that production of the heavy-NP shift
syntactic structure does not stem from properties of the heavy NP alone but appears to vary
with the relative length difference between the NP and the PP. Hawkins (1994) suggested
that this relative length factor might be the only length-based factor that affects shifting
rates, such that the length of the NP alone has no effect on shifting once relative length is
considered. Given the strict control on the 10-word NPs in this study, the results of
Experiment 1 cannot address this claim. Moreover, the Experiment 1 results might not
generalize to NPs of different lengths; relative length effects might be different for longer or
shorter NPs. Experiment 2 begins to investigate both of these issues by manipulating both
NP length and the relative length of NPs and PPs.
Experiment 2: NP Length and Relative Length Manipulations
Experiment 2 used the same production task to investigate whether absolute NP length, in
addition to relative length, affects shifting behavior. Three conditions were compared. Two
conditions were matched for relative length but varied in NP length. Both had a 5-word
length difference between the object NP and the PP, but they varied in the length of these
phrases—in one condition (the 5-word, long NP condition), the absolute lengths of the NP
and PP were 10 words and 5 words respectively, whereas in the 5-word, short NP condition,
the NP was 7 words long and the PP was 2 words long. These conditions were compared to
a third condition in which the PP and NP were each two words long (so that there was no
length difference between the two phrases). If relative length is a driving force of heavy-NP
shift, then the percentage of shifted orders produced in the two 5-word difference conditions
Stallings and MacDonald Page 5
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
should be larger than in the 0-word difference condition. If object NP length also affects the
production of heavy-NP shift structures above and beyond the effects of relative length, then
the 5-word difference condition with the longer (10-word) NPs should yield more shifting
than the 5-word difference condition with shorter (7-word) NPs.
Participants—Thirty-six undergraduates were either paid or received extra credit in
psychology courses for their participation. All were native speakers of English.
Materials and Procedure—A sample of the materials is shown in Table 2. The stimuli
were similar to those from Experiment 1. Specifically, they included subject-verb phrases,
NPs and PPs, with length modifications to form three conditions. The 0-word difference
condition contained an NP and a PP of two words each. The 5-word difference/Short NP
condition contained a 7-word NP and a 2-word PP, and the 5-word difference/Long NP
contained a 10-word NP and 5-word PP.
All 10 practice items and 37 of the filler items were retained from Experiment 1. The shorter
filler items from Experiment 1 were removed because the experimental conditions in this
experiment already included short items.
The procedure was identical to that in Experiment 1. Screen position was again
counterbalanced for the experimental items. For the 37 fillers, all of which were
grammatical with only one ordering of the two phrases at the top and bottom of the screen,
19 had a required phrase ordering of middle-top-bottom, and the other 18 required a middle-
bottom-top ordering.
Results and Discussion
Before analyses, trials were excluded when participants omitted or changed the structure of
one of the sentence constituents, had ready-to-speak response times of less than 500 ms, or
when the audio recording of the utterance was inaudible. A total of 4.91% of the utterances
were excluded. There were no reliable effects of screen position and no interaction of screen
position with the relative length factor, Fs < 1, so this factor will not be discussed further.
We first analyzed the number of words speakers produced in the three conditions. The 0-
word difference condition, with two-word NPs and PPs, was produced at essentially 100%
accuracy. In the 5-word/Short-NP condition, participants produced an average of 6.55 words
(SD = 0.79) for the 7-word NP and 2.06 words (SD = 0.21) for the 2-word PP, for a mean
relative length difference of 4.50 words. In the 5-word/Long-NP condition, speakers
produced an average of 8.44 words (SD = 1.32) for the 10-word NP and 4.34 words (SD =
0.83) of the 5-word PP, for a mean relative weight difference of 4.10 words.
Phrase Order Choice—The rates of heavy-NP shift in the three conditions are shown in
Fig. 2. Analyses showed that there was a robust effect of length F 1(2, 70) = 50.339, p < .
001; F 2(2, 58) = 51.360, p < .001, such that the two 5-word difference conditions yielded
greater rates of shifted utterances compared to the 0-word difference condition (all ps < .
001), but they did not differ from one another, Fs < 1.
These results confirm the effects of relative length on shifting that were found in Experiment
1. The experiment did not find any evidence for NP length having an effect on shifting
beyond the relative length effects. While caution should be used when considering a null
result, and other combinations of length and relative length manipulations might further
Stallings and MacDonald Page 6
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
inform this issue, at this point there is no reason to assume that any length effects beyond the
relative length of constituents affect shifting.
General Discussion
Two experiments showed that the choice of the heavy-NP shift versus the more typical Verb
NP PP structure during speech production is constrained by the relative length of the
constituents in the verb phrase. In both Experiment 1 and Experiment 2, as the difference
between the length of the NP and PP increased, shifting rates increased. There was no
evidence that length of the NP alone contributed to shifting beyond the relative length
effects. These results therefore form an important companion to corpus analyses (Hawkins
1994; Wasow 1997), in which more spontaneous utterances and writing offer naturalistic
examples but less control of factors thought to be driving the production mechanisms during
utterance planning.
These results contribute to prior work suggesting that properties of the heavy-NP alone
cannot entirely explain shifting behavior (Hawkins 1994; Stallings et al. 1998; Wasow
1997); indeed we found that NP length alone had no effect when relative length is
considered in Experiment 2. Interestingly, the effect of relative length does not appear to be
linear—the shifting rates in the short-, mid-, and long-PP conditions of Experiment 1
(38.7%, 37.0%. 29.7% respectively) did not vary in direct proportion to the actual numbers
of words uttered in the short-, mid-, and long-PP conditions (6.10, 3.85, and 3.04 word
differences between NP and PP length uttered in the short-, mid-, and long-PP conditions
respectively). This pattern, along with prior work showing that properties of lexical items
can affect shifting rates (Stallings et al. 1998; Wasow 1997) suggest that other factors are
also contributing to shifting rates. We consider the potential role of other factors below, but
first we frame the relative length effects, and shifting more generally, within language
production research.
Relative Length in Incremental Production
Within current theories of language production, constituents are ordered in an utterance plan
as a function of their accessibility—essentially the ease with which they can be planned and
articulated (Bock and Levelt 1994). On this view, “shifting” is not literally any shifting of
sentence constituents in a mental representation but rather a tendency for the production
planning processes to place relatively short elements earlier in the utterance plan compared
to longer elements, because, all else being equal, shorter phrases are easier to plan than
longer ones (De Smedt 1994). It is also not an obligatory grammatical rule—indeed, our
shifting rates never exceeded 40% of utterances, even in the most conducive experimental
conditions. Instead, shifted utterances tend to become more frequent as the difference in
length between the phrases grows. The role of accessibility on production planning provides
a useful perspective for viewing shifting behavior in several respects. First, whereas at least
some linguistic accounts of shifting offer no basis for expecting that absolute versus relative
weight would determine shifting behavior, relative weight has a natural explanation in a
production system in which the relative accessibility of phrases is determining their order in
the utterance plan. Second, because phrase length is only one factor that appears to modulate
accessibility in other constructions (e.g. McDonald et al. 1993), incremental production
naturally accommodates suggestions that non-length effects modulate shifting. For example,
a number of researchers have argued that syntactic complexity, rather than phrase length,
affects shifting rates (e.g., Ross 1967; Quirk et al. 1972; Kimball 1973; Emonds 1976) or
that both length and complexity affect shifting. Wasow and Arnold (2003) collected
acceptability judgments for shifted and unshifted sentences in which they controlled both
length and syntactic complexity and found higher preferences for shifting with complex
phrases than syntactically simpler ones matched for length. Many other factors may also
Stallings and MacDonald Page 7
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
affect shifting, including lexical and conceptual factors that have been shown to affect
structure choice in other domains, including noun animacy, the coherence of a concept
expressed by a phrase, or the frequency of words in the phrase (Bock and Warren 1985;
McDonald et al. 1993). For example, NPs containing inanimate, low frequency nouns might
be more likely to be produced at the end of a clause compared to nouns of the same length
and syntactic complexity containing animate high-frequency nouns. Bresnan (2007), using a
large corpus of the two dative constructions (prepositional dative, e.g. gave a present to
Mary, versus double object dative, e.g., gave Mary a present), found that speakers’ choices
of the alternative forms were predicted by a combination of fourteen length-based, semantic,
and discourse factors. It seems plausible that structure choice in other constructions,
including heavy-NP shift, may be similarly subject to many influences.
The existence of multiple factors may explain one surprising effect of length on shifting,
which is that in Japanese, length-based shifting appears to go in the opposite direction, with
longer phrases appearing before shorter ones (Hawkins 1994; Dryer 1980; Yamashita and
Chang 2001). If factors in addition to relative length shape accessibility, then this result may
make more sense. Chang (2009) presented a computational model of several length-based
structure choices, including heavy-NP shift. He argued that lexical-semantic properties such
as animacy and learning about language statistics from other structures in the language could
contribute to accessibility and therefore word ordering. The model also addressed the
opposite effects of length on shifting in English and Japanese. Chang suggested that other
properties of English and Japanese conspired to promote the different length-based ordering
preferences in the two languages. For example, the verb precedes the rest of the verb phrase
in English, so that during the sentence planning process, it will be activated and can
influence ordering of subsequent constituents (Stallings et al. 1998; Wasow 1997). By
contrast, in Japanese, the verb is at the end of the verb phrase, and is not likely to be
activated during the planning of the verb phrase. As a result, the verb has little influence on
word order of other elements in the verb phrase. Chang suggested that this and other
differences between the two languages affect the extent to which phrase length modulates
accessibility—in English, length has a substantial effect, with long phrases much less
accessible than short ones, leading to the standard heavy-NP shift effect. In Japanese,
however, length has a much weaker effect, and other elements such as lexical-semantics and
discourse status have a greater role in modulating accessibility of phrases. This view is
consistent with the one advocated here, that the relative accessibility of sentence elements
influences word order during sentence planning and that at least in English, relative length is
a strong contributor to the relative accessibility differences that drive heavy-NP shift. The
investigation of relative length in other languages is likely to increase our understanding of
utterance planning during the process of grammatical encoding in language production.
This research was supported by NSF Grant DBS-9120415 and NICDH R01HD047425. We thank Jack Hawkins for
helpful discussions and Jerry Cortrite, Jared Layport, Miranda Lim, Summer Montague, and Daniel Stallings for
assistance in data collection and transcription
Arnold JE, Wasow T, Losongco A, Ginstrom R. Heaviness versus newness: The effects of structural
complexity and discourse status on constituent ordering. Language. 2000; 76:28–55.
Bock, JK.; Levelt, W. Language production: Grammatical encoding. In: Gernsbacher, MA., editor.
Handbook of psycholinguistics. San Diego, CA: Academic Press; 1994. p. 945-984.
Bock JK, Warren RK. Conceptual accessibility and syntactic structure in sentence formulation.
Cognition. 1985; 21:47–67. [PubMed: 4075761]
Stallings and MacDonald Page 8
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
Bresnan, J. Is syntactic knowledge probabilistic? Experiments with the English dative alternation. In:
Featherston, S.; Sternefeld, W., editors. Roots: Linguistics in search of its evidential base. Berlin:
Mouton de Gruyter; 2007. p. 77-96.
Chang F. Learning to order words: A connectionist model of heavy NP shift and accessibility effects in
Japanese and English. Journal of Memory and Language. 2009; 61:374–397.
De Smedt, K. Parallelism in incremental sentence generation. In: Adriaens, G.; Hahn, U., editors.
Parallel natural language processing. Norwood, NJ: Ablex; 1994. p. 421-447.
Dryer M. The positional tendencies of sentential noun phrases in Universal Grammar. The Canadian
Journal of Linguistics. 1980; 25:123–195.
Emonds, J. A transformational approach to English syntax. New York: Academic Press; 1976.
Givón, T. The pragmatics of word order: Predictability, importance, and attention. In: Hammond, M.;
Moravcsik, EA.; Wirth, J., editors. Studies in syntactic typology. Amsterdam: John Benjamins;
1988. p. 243-284.
Hawkins, JA. A performance theory of order and constituency. Cambridge, England: Cambridge
University Press; 1994.
Kimball J. Seven principles of surface structure parsing in natural language. Cognition. 1973; 2:15–47.
Lombardi L, Potter MC. Syntactic priming in immediate recall of sentences. Journal of Memory and
Language. 1998; 38:265–282.
McDonald J, Bock K, Kelly MH. Word and world order: Semantic, phonological, and metrical
determinants of serial position. Cognitive Psychology. 1993; 25:188–230. [PubMed: 8482072]
Quirk, R.; Greenbaum, S.; Leech, G.; Svartvik, J. A grammar of contemporary English. London:
Longman; 1972.
Race, DS.; MacDonald, MC. The use of “that” in the production and comprehension of object relative
clauses. Proceedings of the 25th annual meeting of the cognitive science society; 2003.
Ross, JR. Unpublished PhD Dissertation. MIT; 1967. Constraints on variables in syntax.
Stallings L, MacDonald MC, O’Seaghdha PG. Phrasal ordering constraints in sentence production:
Phrase length and verb disposition in heavy-NP shift. Journal of Memory and Language. 1998;
Staub A, Clifton C, Frazier L. Heavy NP shift is the parser’s last resort: Evidence from eye
movements. Journal of Memory and Language. 2006; 54:389–406. [PubMed: 17047731]
Wasow T. End-weight from the speaker’s perspective. Journal of Psycholinguistic Research. 1997;
Wasow, T.; Arnold, J. Post-verbal constituent ordering in English. In: Rohdenburg, G.; Mondorf, B.,
editors. Determinants of grammatical variation in English. The Hague: Mouton; 2003. p. 119-154.
Yamashita H, Chang F. “Long before short” preference in the production of a head-final language.
Cognition. 2001; 81:B45–B55. [PubMed: 11376643]
Zec, D.; Inkelas, S. Prosodically constrained syntax. In: Inkelas, S.; Zec, D., editors. The phonology-
syntax connection. Chicago: CSLI and the University of Chicago Press; 1990. p. 365-378.
Stallings and MacDonald Page 9
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
Fig. 1.
Production of heavy-NP shifted structures as a function of prepositional phrase (PP) length
Stallings and MacDonald Page 10
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
Fig. 2.
Production of heavy-NP shifted structures as a function of relative phrase length and NP
Stallings and MacDonald Page 11
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
Stallings and MacDonald Page 12
Table 1
Sample materials for Experiment 1
Subject-verb phrase The radio listeners accepted
10-word object NP The whole story about the defects in the new Mazda
Manipulated PPs
Short PP (2 words long) Without doubt
Mid PP (5 words long) Without any doubt or concern
Long PP (7 words long) Without doubt or any bit of concern
NP noun phrase, PP prepositional phrase
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
NIH-PA Author Manuscript NIH-PA Author Manuscript NIH-PA Author Manuscript
Stallings and MacDonald Page 13
Table 2
Sample materials for Experiment 2
Subject verb phrase: The radio listeners accepted
5-word/Long-NP condition
10-word NP The whole story about the defects in the new Mazda
5-Word PP Without any doubt or concern
5-word/Short-NP condition
7-word NP The whole story on the recent defects
2-Word PP Without doubt
0-word condition
2-word NP The story
2-Word PP Without doubt
J Psycholinguist Res. Author manuscript; available in PMC 2012 June 1.
... While several factors may contribute to HNPS structures, there is evidence that the propensity for HNPS in the production of English utterances increases with the length of the NP in words. For our purposes, the key features of HNPS are first that it affords two viable response options (phrase orders), and that the tendency to produce a PP-first form increases with the difference in word length between the PP and the NP (Hawkins, 1994;Stallings & MacDonald, 2011). The phenomenon of HNPS thus creates an opportunity to design a language task in which behaviour (use of the NP-first or PP-first form) may be expected to vary on a continuum. ...
... Consequently, production choices at the extremes of the continuum were expected to be more variable. For example, previous studies of HNPS suggest that participants produce 40%-50% PP-first phrasing on trials with long NPs (Hawkins, 1994;Stallings et al., 1998;Stallings & MacDonald, 2011;Wasow & Arnold, 2003). ...
Full-text available
Hysteresis in motor planning and syntactic priming in language planning refer to the influence of prior production history on current production behaviour. Computational efficiency accounts of action hysteresis and theoretical accounts of syntactic priming both argue that reusing an existing plan is less costly than generating a novel plan. Despite these similarities across motor and language frameworks, research on planning in these domains has largely been conducted independently. The current study adapted an existing language paradigm to mirror the incremental nature of a manual motor task in order to investigate the presence of parallel hysteresis effects across domains. We observed asymmetries in production choice for both the motor and language tasks that resulted from the influence of prior history. Further, these hysteresis effects were more exaggerated for subordinate production forms implicating an inverse preference effect that spanned domain. Consistent with computational efficiency accounts, across both tasks participants exhibited reaction time savings on trials in which they reused a recent production choice. Together, these findings lend support to the broader notion that there are common production biases that span both motor and language domains.
... Our protocol follows a wellknown task used in various experimental studies on word order preferences (e.g. Stallings & MacDonald 2011). Participants were asked to make a sentence with given constituents that appear on a computer screen and to produce it from memory with the help of a cue from the sentence, after a short lapse of time during which, for distraction, they were presented with a basic arithmetic operation. ...
Full-text available
This study addresses the issue of unmarked word order in Modern Eastern Armenian (MEA), typologically considered an (S)OV flexible language due to its being strongly left-branching as well as due to the syntactic properties of its VP (focus, bare objects, low adverbs). However, Armenian grammars generally consider (S)VO to be the canonical order. To tackle this controversy, we have conducted a corpus study and two sentence production experiments. These studies show that the placement of DOs is mainly triggered by definiteness. While definite DOs are overwhelmingly postverbal, indefinite DOs display a strong preference for preverbal placement. This implies a 'typological discrepancy': although MEA is a strongly left-branching language, the unmarked placement of definite DOs is postverbal. We account for this 'discrepancy' based on areal, historical and cognitive factors. Contact with OV languages has resulted in a consistent shift from right to left-branching in Armenian, whereas word order at the clausal level has resisted the shift, because MEA makes an optimal use of each order with respect to their cognitive advantages. The evolution of word order in MEA is an illustration of the universal cross-linguistic bias toward SVO.
... That is, the length of the direct object NP correlates with particle placement: shorter NPs prefer the discontinuous order (4.4.1) and longer NPs prefer the continuous order (4.4.2). Length effect, along with pronominal anaphoric status of the object NP and placement of phonological stress, is a typical information structuring strategy (Kuno and Takami 1993;Stallings and MacDonald 2011) constrained by attention allocation and processing costs as proposed by Gries (2002Gries ( , 2003. As such it operates independent of construal. ...
... Information about the date of writing is of course useful for studying and plotting the diachronic development. The weight of the object, measured in words, is interesting because heavy objects are more likely to undergo rightward extraposition, an independent type of a syntactic transformation which affects our VO/OV diagnostics (Wasow 1997;Stallings, MacDonald, and O'Seaghdha 1998;Thráinsson 2007;Stallings and MacDonald 2011). ...
This article describes PaCQL (Parsed Corpus Query Language), a novel query language for carrying out research on parsed historical corpora, an important task for the digital humanities. PaCQL implements and enhances many of the most important features of earlier software that is designed for computational research in historical syntax and combines such functionality with a search engine which employs a fast in-memory index that cuts down waiting time in many realistic research scenarios. A web interface is provided with an automatically created summary of the main quantitative findings. The primary goal of this project is to contribute to the development of software tools which are designed from the ground up specifically with the needs of the digital humanities in mind.
This collection presents novel insights into the micro- and macro-variation of causal clauses from a cross-linguistic perspective. It contains a general introduction to the topic setting the scene and nine chapters based on data from Dutch, German, English, Icelandic, Chinese, and Japanese. Topics discussed in the individual chapters involve, inter alia, external, internal and linear syntax of adverbial clauses expressing a causal relation, their semantic interpretation and information-structural properties, verb position, volitionality, and the development of particular causal conjunctions. The findings gained here are of synchronic and diachronic nature and offer new theoretical perspectives on how causal dependency relationships are expressed by inherent causal morpho-syntactic patterns. They also provide a deeper comprehension of how sentential modifiers work, emerge, and develop in general. This volume is an asset to grammarians, syntacticians, theoretical, and historical linguists.
All living beings try to save effort, and humans are no exception. This groundbreaking book shows how we save time and energy during communication by unconsciously making efficient choices in grammar, lexicon and phonology. It presents a new theory of 'communicative efficiency', the idea that language is designed to be as efficient as possible, as a system of communication. The new framework accounts for the diverse manifestations of communicative efficiency across a typologically broad range of languages, using various corpus-based and statistical approaches to explain speakers' bias towards efficiency. The author's unique interdisciplinary expertise allows her to provide rich evidence from a broad range of language sciences. She integrates diverse insights from over a hundred years of research into this comprehensible new theory, which she presents step-by-step in clear and accessible language. It is essential reading for language scientists, cognitive scientists and anyone interested in language use and communication.
Bringing together cutting-edge research, this Handbook is the first comprehensive text to examine the pivotal role of working memory in first and second language acquisition, processing, impairments, and training. Authored by a stellar cast of distinguished scholars from around the world, the Handbook provides authoritative insights on work from diverse, multi-disciplinary perspectives, and introduces key models of working memory in relation to language. Following an introductory chapter by working memory pioneer Alan Baddeley, the collection is organized into thematic sections that discuss working memory in relation to: Theoretical models and measures; Linguistic theories and frameworks; First language processing; Bilingual acquisition and processing; and Language disorders, interventions, and instruction. The Handbook is sure to interest and benefit researchers, clinicians, speech therapists, and advanced undergraduate and postgraduate students in linguistics, psychology, education, speech therapy, cognitive science, and neuroscience, or anyone seeking to learn more about language, cognition and the human mind.
This chapter examines variation patterns across the world's grammars in relation to working memory (WM) models in psycholinguistics. It distinguishes: (1) constrained capacity proposals in which certain limits in WM are used to explain why some grammatical phenomena are (or are supposed to be) nonoccurring; (2) more versus less WM proposals where greater or lesser demands on WM are reflected in dispreferred versus preferred structures and in declining typological distributions across languages; and (3) integrated WM proposals where WM considerations interact with other factors that facilitate processing. These include prediction, and communicative efficiency, with the result that structures that add to WM load can sometimes be preferred across languages. It is proposed that grammars have conventionalized the preferences of language performance and so have been shaped by the same processing considerations, including WM load, that determine usage data within individual languages and that lead to the preferences in performance. As a result grammars and cross-linguistic comparison can provide relevant evidence for certain issues in language processing that are being debated currently, such as the precise nature and role of WM and its interaction with other processing considerations. These grammatical data come from a vast range of languages that psycholinguists have yet to consider
Acceptability rating questionnaires are a highly accessible tool for the experimental study of different syntactic phenomena, however, for the study of word order preferences, they are not as efficient as sentence production experiments. In this paper, we present a constrained sentence production task implemented via web-based self-administrated questionnaires. In order to highlight the advantages of such a paradigm, we compare the results of two experiments, an acceptability rating and a sentence completion task, using the same experimental material in order to study constituent ordering preferences in French and Persian. Our results show that acceptability rating data reflect the variation observed in production data only in a reduced manner and, consequently, make it difficult to study factors with small effect size or to identify a canonical word order among different grammatically possible alternatives.
Full-text available
In construction grammar, the term multiple inheritance has been used to talk about constructions that inherit features that can be traced back to more than one construction. The constructions involved are organized hierarchically, in that the more specific construction inherits features from multiple more general and abstract constructions (Hudson 2007; Trousdale 2013). In recent years, increased attention has been drawn to lateral or constructional relations, which connect constructions at the same level of abstraction (Cappelle 2006; Van de Velde 2014; Traugott 2018; Diessel 2019). Adopting a nested-network approach (Diessel 2019), this dissertation shows that constructions can be motivated by multiple constructions at the same level of abstraction, i.e., via lateral relations. This is explored by means of two case studies. The first study investigates the change from object-verb to verb-object order in Old to Middle English subject relative clauses. It is argued that the principle of end-weight motivated the existence of a postverbal slot, which could expand under the influence of declarative main clauses. It is shown that Old English had a group of non-prototypical subject relative clauses that bore formal and functional similarity to declarative main clauses. This group proves essential in the analysis of the analogical transfer of verb-object order from main clauses to subject relative clauses. The second study investigates Norwegian definite noun phrases of the type han mannen (lit. he man-the) in relation to three other constructions: den mannen (lit. that/the man-the), mannen (lit. man-the), and (ha)n Per (lit. he Per). The relation of han mannen to its neighbors is considered in terms of structural similarity and contrast, which is statistically evaluated with the method of partial dependence plots based on random forests. It is shown that han mannen shares relations with all three constructions individually and is partially motivated by all three. Both studies support the idea that multiple source constructions are able to motivate changes and variation of a target construction to which they are related by lateral relations, i.e., they exist at the same level of abstraction.
Full-text available
Variations in postverbal constituent ordering have been attributed to both grammatical complexity (heaviness) and discourse status (newness), although few studies compare the two factors explicitly. Through corpus analysis and experimentation, we demonstrate that both factors simultaneously and independently influence word order in two English constructions. While past investigations of these factors have focused on their effects in language comprehension, we argue that postponing heavy and new constituents facilitates processes of planning and production.
A number of different factors influence the ordering of constituents after the verb in English. These include the syntactic complexity of the phrases, the discourse status of the information expressed, how tightly the meaning of each constituent is connected to that of the verb, and idiosyncratic lexical biases. Perhaps surprisingly, English speakers do not appear to use ordering as a way of avoiding potential attachment ambiguities. Some of the factors that do influence order seem to be stronger than others, and none of them is categorical. For example, syntactic complexity is a gradient notion, and its influence on order depends on the relative complexity of postverbal constituents. Moreover, the different factors interact. Understanding what determines ordering choices requires studying usage, both through corpus studies and through laboratory experiments. These observations support a model of language processing based on interacting defeasible constraints, rather than one based on categorical and independent modules.
The pragmatics of word-order involves two general cognitive principles: (a) Anaphoric coherence: The predictability/accessiblity of the referent. (b) Cataphoric coherence: The importance/topicality of the referent. Other considerations also weight in, such the structure of clause-chains and contrast/emphasis. Pragmatically-controlled ("flexible") word order is a matter of degree. But even the most rigid-ordered languages allow some flexibility (pre-posing or post-posing) vis-a-vis the statistically-dominant word order. Lastly, the pragmatics of word-order is a major factor in word-order change.
Theoretical linguistics traditionally relies on linguistic intuitions such as grammaticality judgments for data. But the massive growth of language technologies has made the spontaneous use of language in natural settings a rich and easily accessible alternative source of data. Moreover, studies of usage as well as intuitive judgments have shown that linguistic intuitions of grammaticality are deeply flawed, because (1) they seriously underestimate the space of grammatical possibility by ignoring the effects of multiple conflicting formal, semantic, and contextual constraints, and (2) they may reflect probability instead of grammaticality. Both of these points are richly exemplified by studies of the English dative alternation (Green 1971; Gries 2003, 2005; Fellbaum 2005; Bresnan & Nikitina 2003; Bresnan, Cueni, Ni- kitina & Baayen in press; Lapata 1999; Bresnan & Hay 2006; Hay & Bresnan 2006), which is the linguistic domain of the present study. The present study discusses two experiments following up Bresnan et al. (in press). The first indicates that the "soft" generalizations found in corpus studies of the dative alternation reappear in subjects' intuitions of gram- maticality in context, and that language users have substantial knowledge on the basis of these generalizations of what others are going to say (mean- ing here the choice of syntactic structure to convey the message). The sec- ond experiment shows that rare constructions that have been considered ungrammatical by many linguistic theorists are judged natural by speakers when the appropriate soft conditions are met. Intuitive contrasts in gram- maticality that many linguists have reported seem to reflect probabilities rather than categorical constraints.
In two previous papers (Lombardi & Potter, 1992; Potter & Lombardi, 1990) we reported evidence that immediate recall of a sentence requires regeneration from the message level, rather than from a verbatim representation. However, participants tended to reproduce the surface syntax even when there were two meaning-equivalent surface structures available (e.g., for dative verbs, “gave the letter to her mother,” “gave her mother the letter”). In three experiments we tested the hypothesis that this verbatim bias is the result of syntactic priming (Bock, 1986). In Experiment 1 single sentences were recalled; the prime sentence preceded the target dative sentence. In Experiments 2 and 3 two-clause sentences were recalled; the second clause served as a prime that had been perceived but not yet recalled when the first clause was produced, or vice versa. When the prime sentence or clause was a dative that mismatched the surface structure of the target there was an increase in changes to the alternate (primed) structure in recall of the target, compared with control primes. These results support the hypothesis that simply perceiving a sentence is enough to prime its surface syntactic structure, contributing to verbatim recall.