ChapterPDF Available

Adjunct control and the poverty of the stimulus: Availability vs. evidence

Authors:

Abstract and Figures

Subject control in non-finite adjuncts is observed across languages (as in ‘John called Mary after drawing a picture’). Research on the acquisition of adjunct control has generally focused on the relevant grammatical components and when they are acquired. This paper considers these components in the context of the linguistic input to ask how control in adjuncts is acquired. Although adjunct control is available in the input, the instances themselves do not provide evidence for abstract syntactic relations. Implications are considered for linguistic dependencies and the evidence in the input.
Content may be subject to copyright.
Adjunct control and the poverty
of the stimulus
Availability vs. evidence
Juliana Gerard
Ulster University
Subject control in non-nite adjuncts is observed across languages (as in ‘John
called Mary aer drawing a picture’). Research on the acquisition of adjunct con-
trol has generally focused on the relevant grammatical components and when
they are acquired. is paper considers these components in the context of the
linguistic input to ask how control in adjuncts is acquired. Although adjunct
control is available in the input, the instances themselves do not provide evi-
dence for abstract syntactic relations. Implications are considered for linguistic
dependencies and the evidence in the input.
. Introduction
is paper focuses on obligatory control in non-nite adjuncts, as in (1):
(1) John1 called Mary2 aer PRO1/*2/*3 drawing a picture.
In particular, adjunct control is used as a case study for the role of the linguistic
input in acquiring dependencies: while some properties of adjunct control are ob-
served across languages, others are language-specic. Additionally, exceptions to
canonical control structures raise questions about the type of information needed
from the input.
In (1), the adjunct subject PRO is obligatorily controlled by the main clause
subject John. is pattern is observed across languages, and is captured by high at-
tachment of the adjunct clause and c-command by the controller (Chomsky 1981).1
erefore, evidence for these features must be available in the linguistic input or
they must be innate (Chomsky 1965).
. is paper is based on these components, but may also be considered in the context of other
frameworks; importantly, adjunct control involves a locality constraint which is structurally de-
ned. is constraint is the focus of this paper.
https://doi.org/./la..ger 🔒Available under the CC BY-NC . license
©  John Benjamins Publishing Company
 Juliana Gerard
e goal of this paper is to evaluate these features and their predictions for the
linguistic input, and the primary question is how the features of obligatory control
are acquired. As abstract features, they cannot be observed directly. erefore, if
evidence is available in the input then this evidence must be inferred from observ-
able features or patterns in the input. For example, this inference may be possible
based on the context or distribution of the surface features (Pullum & Scholz 2002;
Scholz & Pullum 2006; Ambridge etal. 2008; Ambridge 2019; Tomasello 2009;
Regier & Gahl 2004; Perfors, Tenenbaum & Regier 2011; Pullum 2020, i.a.), or on
frequencies of n-grams that make up a complex structure (Pearl & Sprouse 2013a;
Pearl & Sprouse 2013b; Pullum & Scholz 2002; Mintz, Newport & Bever 2002).
For adjunct control, this question depends on the availability of adjunct control
in the input, children’s perception of the input, and the relevant form of evidence
for the abstract features of control. If evidence is available for attachment height and
the c-command dependency, this evidence may be observed in sentences with ad-
junct control, specically; alternatively, the features may be generalized from other
structures. However, if evidence is not available in the input, then some aspects of
these features must be innate (i.e. specied in Universal Grammar, or UG), and
evidence is needed for language-specic aspects of the dependency.
ese factors are considered for adverbial adjuncts like (1) with obligatory con-
trol.2 e analysis is based on a critical review of predictions from previous studies,
with support from novel corpus data. Importantly, while the input does include
sentences with adjunct control, it does not provide evidence for the abstract com-
ponents of adjunct control, i.e. attachment height and a c-commanding controller.
is includes both direct evidence (from observing instances of adjunct control in
the input) and indirect evidence (by generalizing from similar structures).
If attachment height and c-command are innate, this makes further predictions
about the linguistic input. Finally, implications are considered for the acquisition of
non-obligatory control, linguistic dependencies in general, and the role of evidence
in the input.
. What is evidence?
e primary question of this paper is how control in adjuncts is acquired. e
following sections consider two preliminary issues: rst, evidence for adjunct con-
trol must be available in the input; and second, children must be receptive to this
evidence when they encounter it.
. ese are the most frequently used adjuncts in previous acquisition studies. Other adjuncts
with obligatory control are not discussed in this paper (e.g. rational clauses, purpose clauses, telic
clauses), although the paper’s conclusions have broader implications for control in general.
Adjunct control and the poverty of the stimulus 
is raises the question, what kind of input constitutes evidence for adjunct
control? Attachment height and c-command cannot be observed directly; therefore,
the availability of adjunct control in the input does not equate to evidence in the
input. Additionally, this question cannot be answered solely by observing children’s
behavior, either in naturalistic productions or in an experimental context: although
children’s behavior can be indicative of their linguistic knowledge, it does not reveal
how that knowledge is acquired. At the same time, children’s perception of the input
depends on their linguistic knowledge: for example, a child with a non-adult gram-
mar will access non-adultlike interpretations of the input; this has consequences
for the evidence that’s needed for the adult grammar.
e above issues therefore depend both on external factors– here, the syntactic
structures in the input– and internal factors– the grammatical competence needed
for interpreting the input. ese factors are discussed in the following sections.
. Considerations for the input
If evidence for adjunct control is available in the input, then the relevant input will
depend on several factors. First, the timeframe for the input is determined by the
ages when a child is receptive to the evidence. Next, the relevant input within this
timeframe depends on the source of the evidence. Also important is the signal to
noise ratio, with multiple sources of noise to consider.
.. e input: Timeframe
In previous studies, children have shown non-adultlike behavior for adjunct control
at age 4, but were generally adultlike by age 7 (Goodluck 1981; Hsu, Cairns & Fiengo
1985; Goodluck & Behne 1992; Janke & Bailey 2017; Janke & Perovic 2017; Janke
2018). erefore, evidence for the adult grammar must be available before this.
Meanwhile, a lower limit may be considered based on prerequisite knowledge
and parsing capacity (Sutton 2015). For example, evidence for attachment height
requires a distinction between arguments and adjuncts, while a c-commanding
controller assumes hierarchical structure and involves the deployment of binding
relations. Additionally, identifying control in non-nite (rather than nite) adjuncts
involves language-specic realization of tense.
Children are sensitive to argument structure by 24 months (Naigles 1990;
Gertner, Fisher & Eisengart 2006; Arunachalam etal. 2011; for a review see Fisher
etal. 2010); if this is indicative of a distinction between arguments and adjuncts,
then evidence may be available for some properties of adjunct control at this age.
Moreover, some binding relations may be computed by 30 months (Sutton, Fetters
& Lidz 2012; Lukyanenko, Conroy & Lidz 2014). However, evidence may also be
 Juliana Gerard
limited by children’s parsing capacity at a given age. For example, even when bind-
ing is available within clauses, cross-clausal binding relations may not yet be a
reliable source of evidence.
In general, if evidence is available for adjunct control in the input then it should
be available before age 7, but a lower limit will depend on the form of the evidence:
more salient, early-acquired forms like tense are likely to be available earlier than
more complex elements of control, like binding relations. As a tradeo, complex ele-
ments may provide more information about abstract features than the early acquired
forms. Either way, this evidence must be provided by a reliable source in the input.
.. e input: Sources of evidence
is paper is concerned primarily with evidence in the linguistic input. Importantly,
this is not the same type of evidence that is provided by an experiment for testing
children’s knowledge. is second type of evidence– experimental evidence– is
based on children’s behavior, and can be used by researchers to make inferences
about children’s grammatical knowledge at the time of testing.
Meanwhile, evidence in the input is used by children to acquire the adult gram-
mar. is evidence is therefore not based directly on children’s behavior, and does
not allow for direct inferences about children’s knowledge. However, since chil-
dren’s experience of the input depends on their grammatical knowledge, experi-
mental evidence can help to identify a potential mismatch between the input and
children’s perception of the input– i.e. the linguistic intake (Lidz & Gagliardi 2015;
Omaki & Lidz 2015); this mismatch can have implications for the evidence in the
input (discussed further below).
Another relevant contrast is between children’s own productions and the in-
put that they receive (from caretakers, sibling, etc.). Like experimental evidence,
children’s productions may be used to make inferences about their grammatical
knowledge; for example, if children produce only adultlike utterances at a given age,
this is likely evidence that children have acquired the adult grammar by that age.
In contrast, evidence in the input occurs in speech to children. erefore, for
a given child, the relevant evidence for adjunct control will not depend on their
own utterances.
.. e input: Signal to noise
Before moving on to internal factors, a nal external consideration is the noise in
the input from extragrammatical sources (Lidz & Gagliardi 2015; Omaki & Lidz
2015; Phillips 2013; for a review, see Pearl 2019). In addition to children’s grammat-
ical competence, important factors include speech errors in the input and parsing
errors in the intake, with implications for input frequency and the relative contri-
bution of a single instance.
Adjunct control and the poverty of the stimulus 
While non-adult interpretations are expected from a non-adult grammar, er-
rors may also be observed for adjunct control with the adult grammar, due to
extra-grammatical factors (Parker, Lago & Phillips 2015; Kwon & Sturt 2014; Kush
& Dillon 2020; Gerard etal. 2017). For example, speech errors like disuencies may
disrupt encoding of the input (Banbury etal. 2001), while a non-subject antecedent
of PRO will introduce noise for adjunct control, specically.
In addition, noise is likely to result from the deployment of immature parsing
mechanisms, independent of children’s grammatical knowledge. For sentences
with adjunct control, the antecedent of PRO must be retrieved from memory;
however, a similar referent in memory can interfere with the retrieval mechanism
(Gordon, Hendrick & Johnson 2001; Gordon, Hendrick & Johnson 2004; Warren
& Gibson 2002; Warren & Gibson 2005; Gordon etal. 2006; for a review, see
Gordon & Lowder 2012). is interference may occur for any grammar (adultlike
or non-adultlike), and the resulting interpretation may be consistent or inconsistent
with the child’sgrammar.
If an interpretation in the intake is inconsistent with the adult grammar, this
is a problem: such an interpretation should be taken as evidence against the adult
grammar (Belletti 2017; Pearl 2019). To avoid this conclusion, a learning strategy
is needed which can lter the input, depending on the likelihood of a parsing error
in the intake (Perkins, Feldman & Lidz 2017). For any single utterance in the input,
this likelihood is non-zero, with a higher likelihood of a parsing error for more
complex utterances (Boyle & Coltheart 1996). As a result, the relevant evidence
may also require multiple observations.
is strategy is important for adjunct control, since a single observation in the
input may be inconsistent with the adult grammar in the intake. Consequently, the
relative frequency of adultlike interpretations must be high enough to override the
non-adultlike ones, regardless of how they arise (non-adult grammar, speech error,
or parsing error). A further implication of this strategy is that a single observation
is not sucient for acquiring the adult grammar. is also avoids a transition to a
non-adult grammar for every non-adult observation in the intake.
is section has discussed several considerations for adjunct control in the
input. If evidence is available in the input, it is expected within a certain timeframe,
from an external source (rather than the child themself), and at a high enough
frequency to override expected noise in the input. ese factors are important for
determining the availability of evidence. In addition to availability, however, chil-
dren must also be receptive to this evidence to acquire the adult grammar.
 Juliana Gerard
. Considerations for grammatical competence
Previous studies on children’s acquisition of adjunct control have generally used
sentences with a structure like in (1), repeated below:
(1) John1 called Mary2 aer PRO1/*2/*3 drawing a picture.
Importantly, there are two animate antecedents in the main clause, both of which
are a semantically plausible antecedent for PRO (Goodluck 1981; Hsu, Cairns &
Fiengo 1985; McDaniel, Cairns & Hsu 1991; Goodluck & Behne 1992; Cairns etal.
1994; Broihier & Wexler 1995; Adler 2006; Janke & Bailey 2017; Janke & Perovic
2017; Gerard etal. 2017; Gerard etal. 2018; for a review see Janke 2018).
is isolates children’s syntactic knowledge as the source of their interpreta-
tion:3 in (1), the adult grammar identies the main clause subject as the anteced-
ent of PRO; however, for a non-adult grammar which does not rule out the main
clause object as an antecedent, (1) is ambiguous since there are multiple plausi-
ble antecedents. at is, a non-adult grammar of adjunct control can generate an
adultlike (subject control) interpretation of (1), or a non-adultlike (object control)
interpretation.
In previous studies on adjunct control, children have allowed both adultlike and
non-adultlike interpretations of (1). is is consistent with a non-adult grammar
which generates both interpretations. However, with a non-adult grammar, evi-
dence is required at some point for the adult grammar. Importantly, this evidence
must be available not only in the linguistic input, but also in the intake.4
. See work by Janke & Bailey (2017), Janke (2018) and Gerard etal. (2017, 2018) for pragmatic
and extragrammatical sources of children’s interpretations.
. One concern with sentences like (1) is that both plausible antecedents are sentence-internal,
making the sentence ambiguous for a non-adult grammar that allows object control. In contrast,
the following sentences have just one plausible sentence-internal antecedent:
(i) John1 called a taxi2 aer PRO1/*2/*3 drawing a picture.
(ii) John1 called aer PRO1/*2 drawing a picture.
ese sentences make contrasting predictions for dierent grammars: with a non-adult grammar
that allows any internal antecedent for PRO, but not an external antecedent, (i) and (ii) may
be disambiguated based on plausibility alone. However, a grammar which does allow external
antecedents, i.e. a free reference grammar, may still generate a non-adultlike interpretation for
(i) and (ii), if another referent is available in the discourse.
In previous studies, children who accepted a non-adultlike internal antecedent also tended
to accept an external antecedent for PRO, consistent with a free reference grammar (McDaniel,
Cairns & Hsu 1991; Cairns etal. 1994; Broihier & Wexler 1995; Adler 2006). erefore, if children
need evidence for the adult grammar of adjunct control, this evidence must be available even with
the interpretations allowed by a free reference grammar, i.e., with any referent in the discourse.
Adjunct control and the poverty of the stimulus 
e fact that a non-adult grammar generates non-adultlike interpretations
presents a puzzle: for adjunct control in the input, if children have a non-adult
grammar, then they will access both adultlike and non-adultlike interpretations,
as in previous experimental contexts (Wexler 1990).
Another consideration, however, is that the antecedent of PRO is a realization
of the abstract features of control, i.e. attachment height and c-command by the
controller. In previous studies, children’s interpretations were non-adultlike if they
identied a non-subject antecedent of PRO; accordingly, non-adult grammars have
been proposed with the incorrect attachment height (Goodluck 1981; Hsu, Cairns
& Fiengo 1985; McDaniel, Cairns & Hsu 1991; Cairns etal. 1994; Adler 2006) or
an immature representation of the control relation (Goodluck 2001; Goodluck &
Behne 1992; Broihier & Wexler 1995; Wexler 2019). Evidence for the adult gram-
mar would therefore relate to attachment height or the correct control relation,
respectively.
ese features cannot be observed directly, so this evidence must be available
indirectly, from observable features of the input. Additionally, the evidence must
be robust to children’s non-adultlike interpretations– that is, a non-subject ante-
cedent must not interfere with evidence for the adult grammar. Evidence for the
adult grammar must therefore involve other features of adjunct control, instead of
(or in addition to) the antecedent of PRO.
In this section, several issues have been considered for the linguistic input, as
well as children’s perception of the input. ese have implications in general for
the relevant input where evidence would be observed, and the form of evidence
for the adult grammar. e next sections consider these implications for adjunct
control, focusing rst on the availability of adjunct control in the input, followed
by evidence in the input.
. Availability
e linguistic input is represented here by transcripts of speech to children from
CHILDES (MacWhinney 2000). e analysis included all transcripts from the
North American English corpus,5 with the exception of transcripts from children
older than age 7 as discussed above, and transcripts with interviews between a
parent and interviewer with no child present.
. All transcripts are from the following corpora: Bates, Bernstein, Bloom, Braunwald, Brent,
Brown, Clark, Garvey, Gathercole, Gelman, Gleason, Hall, HSLLD, Kuczaj, MacWhinney, Moris-
set, Nelson, NewEngland, NewmanRatner, Peters, Post, Sachs, Snow, Soderstrom, Suppes, Tardif,
Valian, VanKleeck, and Weist.
 Juliana Gerard
Instances of adjunct control were identied by searching for each comple-
mentizer followed by the string “ing” (Broihier & Wexler 1995). Non-nite com-
plementizers included in the search were aer, before, while, when, without, and
instead of, which were then hand coded to exclude false positives (e.g. “what hap-
pens aer spring”). e results for each complementizer are presented in Table1,
which shows the number of utterances with adjunct control in the input (adult),
and the number produced by the target child.
Table1. Adjunct control in North American CHILDES, raw counts
Complementizer Adult Target child
aer 35 5
before 31 1
while 11 3
when 5 3
without 128 26
instead of 121 23
Tota l 331 61
Based on these counts, the following observations can be made for adjunct control
in this timeframe:
Adjunct control is available in the input before age 7.
Children produce adjunct control before age 7.
e instances with without and instead of are much more frequent than with
aer, before, while and when, for both children and adults; this contrast re-
ects the optionality for the lower frequency set of complementizers between a
non-nite or nite frame, compared to without and instead of, which can only
appear in a non-nite frame:
(2) a.
John called Mary
PRO drawing a pictu
re.
aer
before
while
when
b.
John called Mary he drew a picture.
aer
before
while
when
Adjunct control and the poverty of the stimulus 
(3) a.
John called Mary
PRO drawing a pictu
re.
without
instead of
b.
*John called Mary he drew a picture.
without
instead of
erefore, adjunct control is available in the input, and children are sensitive to
at least some aspects of the dependency, particularly the respective frequency by
complementizer.
Meanwhile, the counts in Table1 do not illustrate the frequency of the ut-
terances with adjunct control compared to other utterances in the input, or the
distribution of these counts over time (Gries 2008; Gries 2010; Wang & Trueswell
2019). is information is represented in Figure1, which plots children’s and adults’
utterances with adjunct control by two measures of development: children’s age in
years and children’s mean length of utterance (MLU). ese measures are corre-
lated, although in children’s own productions, adjunct control is better predicted
by MLU than by age. To illustrate the frequency of these utterances, Figure1 also
shows all transcripts in the corpus plotted by the age and MLU of the target child;
a mean of 341 utterances were produced in each transcript.
age of child (years)
no age
data
      
no child
utterances
(MLU = 0)
MLU of child
utterance source
adult (input)
target child
transcript
(mean of 341
utterances per
transcript)

Figure1. Instances of adjunct control in the input (produced by adults and siblings of the
target child), and instances produced by the target child, plotted by age and mean length
of utterance (MLU) of the target child
 Juliana Gerard
Importantly, adjunct control is available in the input at all ages, although at a rela-
tively low frequency throughout: from the ages of 2–5 years, children encounter one
utterance with adjunct control for every 2,000–3,000 utterances. For comparison
with other complex structures, this is less than 10% of the frequency of passive
constructions (Nguyen & Pearl 2018; Nguyen & Pearl 2019), which in turn are less
frequent than object relative clauses (Roland, Dick & Elman 2007).6 at is, adjunct
control does occur in the input, but at a lower frequency than other structures for
which non-adultlike behavior is reported at similar ages (for reviews, see Huang
etal. 2013; Adani, Stegenwallner-Schütz & Niesel 2017).
Next, sentences with adjunct control are generally produced by children with
an MLU of at least 4 (with the earliest productions between the ages of 3 and 4). is
shows that children produce the relevant non-nite contexts far younger than age
7; however, children’s non-adultlike behavior in previous studies was determined
based on interpretation (of the antecedent), rather than form (of the non-nite
adjunct). erefore, evidence in the input may also depend on the availability of
subject control, compared to other antecedents.
To assess this availability, the utterances from Table1 were hand coded for the
antecedent of PRO. In addition to subject control, possible antecedents included the
following categories (Wexler 1992; Goodluck 2001; Williams 1992; Landau 2015;
Landau 2017; Green 2018a):
a non-subject antecedent in an otherwise expected subject control context,
e.g. Mary in (1)
arbitrary PRO, as in (4)
logophoric PRO, as in (5)
an unclear antecedent– although this could be resolved in most cases by refer-
ring to previous discourse, this was not possible in a few cases when the utter-
ance wasn’t coherent, or when the speaker switched topics in the conversation
before completing the utterance
(4) It was good to call aer PRO drawing a picture.
(5) e ower wilted aer PRO drawing a picture of it.
Of the 392 utterances with adjunct control from Table1, nearly all had a subject
control interpretation. e instances which did not are presented in Table2 (input
utterances) and Table3 (target child’s utterances):
. In an analysis of the Brown and Valian corpora, Nguyen and Pearl (2018, 2019) reported 361
passive utterances in 113,024 total utterances, or 1 passive for every 313 utterances. Meanwhile,
Roland etal. (2007) reported even greater raw counts for object relative clauses in the Brown corpus
alone, with 608 object relatives, 1,460 reduced object relatives, and 658 object innitive relatives.
Adjunct control and the poverty of the stimulus 
Table2. Adjunct control with non-subject antecedent, input utterances
Child age
(years)
PRO
referent
Utterance
1arb at would be a good way to get <to things>. instead of reaching.
2arb I have a good rule that we have at school. to raise our hand instead
of yelling.
unclear Without nishing it.
unclear Nothing without spelling anything.
3non-subject I thought we could give her some tea before going to bed from this
pretty little tea pot.
(from discourse, PRO clearly refers to “her”)
logophoric So it won’t fall down without tying it to your chin.
unclear Aer (.) sliding though.
4 unclear &-um when you’re here alone with, when you, aer reading the four
seasons get him to just tell me for a few minutes about something
that you did and then we’ll do the same thing with Jake.
unclear Eleven o’clock at night aer sitting up in bed for two and three hours.
unclear Even aer being here all this time.
unclear Maybe aer (.) coming back [unintelligible].
5 logophoric An(d) I knew that if anyone would takes this home it would take up
too much room, so it would be easier to carry without dropping.
arb Going three days without making a juice circle really blew your mind.
arb Humming while eating noodles.
arb ere’s no breaking without breaking.
arb It helps to show that maybe these are muscles. without having to
draw all the in, all the muscles there.
Table3. Adjunct control with non-subject antecedent, target child’s utterances
Child age
(years)
PRO
referent
Utterance
2 unclear And aer playing +… with with all my
3non-subject
Yeah but when trying to catch daddy (.) daddy put me under the water.
arb Instead of eating a lot (.) that would be good.
5 logophoric (In)stead of walking, car is better going to school.
arb And [/] and that was the most important [: important] [* d] job
instead of doing the prayer.
arb ere’s no making without breaking.
unclear Without catching.
unclear Maybe aer (.) coming back [unintelligible].
6arb at’s what’s fun about [unintelligible] looking out the window
without having to be driving.
 Juliana Gerard
e utterances in Table2 demonstrate that non-subject antecedents occur in the
input, both due to speech errors, and also in non-obligatory control constructions.
In children’s own productions in Table3, the counts of these categories occur in
similar proportions. Further conclusions from Tables2 and 3 are limited, however,
before considering the evidence that would be available from observations with
obligatory control, or other forms of evidence in the input. is evidence is the
focus of the following sections, which consider the following hypotheses:
a. evidence for attachment height and c-command is available in the input, either
i. by observing instances of adjunct control directly or
ii. by generalizing the relevant features from similar structures.
b. evidence for these features is not in the input, and the features are specied in UG.
. Evidence
If either attachment height or c-command by the controller are acquired from the
linguistic input, then explicit predictions are made about the evidence in the input.
Two types of evidence will be considered here: rst, the conditions are spelled out
for inferring the correct attachment height or c-command by observing instances of
adjunct control directly. Next, these features may be generalized to adjunct control
from similar structures, which may be more frequent or salient in the input.
. Direct observation
For attachment height or c-command to be inferred by observing instances of
adjunct control, there must be instances of adjunct control available in the in-
put. Based on the CHILDES data in Section3 above, this requirement is satised.
However, while adjunct control is necessary, it is not sucient; other factors to
consider include the prerequisite linguistic knowledge and children’s perception
of the input. ese factors are discussed in the following sections.
.. Attachment height
If children need evidence for adjunct attachment height, then incorrect attach-
ment is predicted before the relevant evidence is encountered in the input. During
this stage of incorrect attachment, non-adultlike interpretations are predicted
for adjunct PRO.7 Indeed, children in previous studies have accepted a range of
. Crucially, this is not reversible: if children have non-adultlike attachment, then non-adultlike
interpretations of PRO are expected. However, if children have non-adultlike interpretations of
PRO, this does not entail that they have attached the adjunct incorrectly – this is one possibility,
among others.
Adjunct control and the poverty of the stimulus 
interpretations, and one prominent account is misattachment of the adjunct to
the main clause (Goodluck 1981; Hsu, Cairns & Fiengo 1985; McDaniel, Cairns
& Hsu 1991; Cairns etal. 1994; Adler 2006). Two primary forms of evidence have
been considered for attachment height in previous studies, which make dierent
assumptions about children’s pre-existing knowledge.
... Lexical learning (Cairns etal. 1994)
To account for children’s behavior, Cairns etal. (1994) propose dierent non-adult
grammar types, which predict non-adultlike interpretations before children acquire
the adult grammar. ese grammar types involve high attachment of the adjunct to
the main clause (coordination) or low attachment (with c-command by the main
clause object). Here, an important distinction is made between types of accounts:
these non-adult grammar types can explain children’s behavior in the study; how-
ever, the grammar types alone do not provide an account of acquisition– i.e. how
a learner can transition from a non-adult grammar to the adult grammar.
To account for children’s acquisition, Cairns etal. (1994) cite the Lexical
Learning Hypothesis (Wexler & Chien 1985), noting that children must link each
complementizer form with its selectional properties. ey suggest that incorrect
attachment results from mapping a complementizer form rst to a coordinating
structure, before acquiring the correct mapping for a non-nite adjunct. Evidence
for the correct attachment would therefore be available with any instance of a given
complementizer (not just as a non-nite adjunct), with the transition to the adult
grammar resulting from “accretion of lexical and semantic knowledge” for each
complementizer (Cairns etal. 1994: 264).
is description accounts for the transition to the adult grammar; however,
it does not involve the acquisition of syntactic structure. It assumes instead that
children already have the relevant abstract knowledge of coordination and subor-
dination, with incorrect form-structure mappings. If adjunct attachment height is
assumed as preexisting knowledge, then another source of evidence is needed for
attachment height, or it is innate.
... Adjunct misanalysis (Adler 2006)
In a dierent misattachment account, Adler (2006) suggests that the syntactic con-
trasts between non-nite adjuncts and coordinated clauses may be used as cues to
attachment height. For example, the verb form in non-nite adjuncts contrasts with
the nite form in coordinated clauses:
(6) a.
John eats cake before presents.
opening
*opens
b.
adapted from Adler (2006)
 Juliana Gerard
Other contrasts involve transformations; for example, cle structures are possible
with adjuncts but not coordinate clauses:
(7) a. It was before opening presents that Mary cut the cake.
b. *It was and John opened presents that Mary cut the cake.
Similarly, dierent proles are observed for extraction:
(8) a. Whati did you eat ti before John opened presents?
b. *Whati did you eat ti and (then) John open presents?
Importantly, these examples involve positive evidence (Berwick 1985): in (6) the
contrast in verb form (or niteness) is a cue to the contrast in clause type, while in
(7) and (8), the transformation itself is a cue– since the sentences are not possible
with a coordinated clause, any instances in the input would need to be represented
with an adjunct clause (Adler 2006).
However, the above evidence is still problematic for learning attachment height.
In (6), the contrast in verb form aligns with the contrast in attachment height:
that is, coordinated clauses and non-nite clauses have dierent verb forms and
dierent attachment heights. is strategy makes the wrong predictions for nite
adjuncts, though, which also have a nite verb form (grouping nite adjuncts with
coordinated clauses):
(9)
John eats cake before he presents.
*opening
opens
is miscategorization may be avoided if the contrast in (6) is applied to a subset
of the input data. However, this would involve domain-specic knowledge about
which data to use for learning, merely shiing the learning problem rather than
addressing it.
Meanwhile, the sentences in (7) and (8) must be represented accurately in order
to be used as evidence for the correct attachment height. However, the inuence
of an immature parser, along with high sentence processing costs may aect the
reliability of this evidence.
More broadly, both types of evidence discussed by Adler (2006) rely on prior
knowledge of a contrast in attachment height between adjuncts and coordinated
structures. Moreover, similar to the approach by Cairns etal. (1994), the relevant
learning strategies involve mapping a lexical item (complementizer) to abstract
structure (adjunct clause), by abandoning an initial incorrect mapping (coordinated
clause). ese mappings are important, but they require the attachment height
for adjuncts to have already been acquired. Again, attachment height must either
be innate here, or acquired using another form of evidence. A nal possibility for
attachment height is discussed in the following section.
Adjunct control and the poverty of the stimulus 
... Binding across clauses
e next type of evidence to consider for attachment height involves binding rela-
tions across clauses, as in (10) and (11):
(10) He1 called Mary before John*1/2 le for the store.
(11) John called her1 before PRO meeting Mary1 at the store.
In (10), the pronoun he c-commands John, and co-reference is ruled out by Princi-
pleC (Chomsky 1981). However, co-reference is possible if the adjunct is attached
high. us, if children have a grammar with high attachment, negative evidence is
needed against co-reference in sentences like (10), which may then be used to infer
the correct (lower) attachment height.8
Meanwhile, syntactic evidence against a low attaching adjunct is seen in sen-
tences like (11), with co-reference between her and Mary. If children have a gram-
mar with low attachment, then co-reference in the input with this conguration
would provide positive evidence for the correct (higher) attachment height.
For both (10) and (11), the relevant evidence involves several assumptions
which are problematic for acquisition. First, evidence against the co-reference in
(10) might be available in the form of indirect negative evidence (Xu & Tenenbaum
2007); however, previous research on children’s acquisition of Principle C nds
that children already reject co-reference in this conguration from as young as 3
years of age (Crain & McKee 1985; Crain & ornton 1998; for reviews, see Lust,
Eisele & Mazuka 1992; Guasti 2017). is timeline is inconsistent with studies on
adjunct control, where children’s non-adultlike interpretations were observed until
5–6 years of age.
Alternatively, children might acquire a high attachment grammar initially but
get evidence for the adult grammar before age 3. However, if the relevant evidence
involves referential dependencies across multiple clauses, the timeframe is further
limited by children’s parsing abilities at this age.
More importantly, using binding across clauses as evidence for attachment
height involves the crucial assumption that the relevant congurations will be avail-
able in the linguistic input. However, for both (10) and (11), the critical anaphoric
relations are highly infrequent, especially if the relevant timeframe is limited by
other factors like the developing parser (Sutton 2015; Gerard 2016). Furthermore,
this type of evidence depends on the co-reference interpretation, which children
may not always access: if a dierent referent is assigned the intake than from the
input, then this will provide evidence for the incorrect attachment height (Lidz &
. As (10) is nite, this strategy involves an additional generalization from nite to non-nite
adjuncts (discussed further below).
 Juliana Gerard
Gagliardi 2015; Omaki & Lidz 2015). us, it is unlikely that binding relations alone
are used as evidence for attachment height for non-nite adjuncts.
Attachment height will be addressed again in the section on generalization; the
following section considers the evidence for inferring a c-commanding controller.
.. C-command by the controller
Inferring the c-command relation between the main clause subject and adjunct
PRO is a two-step process:
1. Identify the set of possible antecedents for adjunct PRO (i.e. the main clause
subject).
2. Determine that the dependency is due to c-command, as opposed to e.g. a dis-
course or agent preference or based on a property like animacy, which are also
likely to involve the main clause subject.
It is assumed that before reaching step 2, a learner has already acquired the correct
attachment height, either from other evidence in the input, or attachment is spec-
ied in UG (Goodluck & Behne 1992). Otherwise, the inference in step 2 cannot
be made based on a hierarchical relation.
Meanwhile, these steps must be indirect on some level: with just a single in-
stance of adjunct control in the input, the interpretation of PRO is consistent with
multiple grammars. For example, in addition to a strict subject (adult) grammar,
the co-reference in (1), repeated below as (12), is also consistent with an agent
grammar, a sentence-internal grammar, a free reference grammar, and others.
(12) John1 called Mary2 aer PRO1/*2/*3 drawing a picture.
All things equal, inferring that the antecedent of PRO is the main clause subject
therefore requires multiple instances of adjunct control. However, children’s inter-
pretations in previous studies suggest that this inference will be problematic, for
any type of learning mechanism (domain-specic or domain-general).
Traditionally, children with a non-adult grammar will encounter some form in
the input which is consistent with the adult grammar but not with the non-adult
grammar, and this form will be evidence for the adult grammar (Gold 1967;
Pinker 1979; Grimshaw & Pinker 1989; Pinker 2009). is logic is discussed in
Section4.1.1.2 above for encountering syntactic evidence against a coordination
grammar. However, as discussed in Section2.2 above, children with a non-adult
grammar of adjunct control will access adultlike interpretations and non-adult in-
terpretations of the linguistic intake. As a result, the set of interpretations generated
by the non-adult grammar is a superset of the interpretations generated by the adult
grammar. ese relations are illustrated in Figure2.
Adjunct control and the poverty of the stimulus 
adult
grammar
non-adult
grammar
data2
data1
Figure2. Subset-superset relation between the adult grammar and non-adult grammars
for adjunct PRO (sentence-internal and free reference). While the adult grammar only
includes data1, the non-adult grammar includes both data1 and data2
is is inconsistent with the Subset Principle, which posits that children will select
the subset language over the superset language (Berwick 1985; Manzini & Wexler
1987; Wexler & Manzini 1987; Wexler 1990). Additionally, transitioning to the
adult grammar requires negative evidence (Berwick 1985; Gold 1967; Baker 1979;
Manzini & Wexler 1987; Pinker 2013; Heinz & Riggle 2011).
One potential option for this involves the size principle, where smaller hypoth-
eses are considered to be more likely than larger hypotheses (which generate a su-
perset of the data generated by a smaller hypothesis), and exponentially more likely
as more data that is observed that is compatible with both hypotheses (Tenenbaum
1999; Tenenbaum & Griths 2001; Xu & Tenenbaum 2007). However, this logic
does not work with evidence for the subject as the antecedent of PRO, and high-
lights a more general problem with acquiring syntactic constraints on anaphora.
A non-adult grammar which allows a superset of the interpretations in the
adult grammar is represented in Figure2– for example, a free reference grammar.
e subset grammar is the strict subject (adult) grammar, which allows only a
subject control interpretation. Under the size principle, children should transi-
tion from the superset grammar to the subset grammar by observing instances of
adjunct control in the input with a subject control interpretation, represented by
data1 in Figure2. e subset grammar should be considered to be more likely if
data like data1 occur in the input. Other than the few instances of speech errors and
non-obligatory control in Table2, data1 (subject control) was indeed the only type
of data in the input. However, this overlooks the additional noise introduced in the
intake from extragrammatical factors, and the nding from previous studies that
children allowed non-adultlike interpretations of adjunct PRO. If these children’s
grammars were not adultlike, then they would also allow non-adultlike interpre-
tations of the input, represented by data2 in Figure1. Crucially, data2 will provide
evidence against the adult grammar and for the non-adult grammar (Fodor 1989;
Fodor 1994; Grodzinsky 1989).
 Juliana Gerard
As a result, children’s interpretations of adjunct PRO are not a reliable cue for
inferring the c-command relation. Moreover, other syntactic dependencies face
a similar dilemma: if children accept a wider range of interpretations in an ex-
perimental context, then the same interpretations are likely to be available in the
linguistic input. Further implications are discussed in the nal sections.
If the grammatical components of adjunct control are not inferred directly–
from instances of adjunct control in the input– then evidence may instead be
available from other structures, which may be generalized to structures with ad-
junct control.
. Generalization from similar structures
e following sections will consider the possibility of generalizing attachment
height and c-command to sentences with adjunct control from two similar struc-
tures: complement control, where the dependent element has the same form; and
nite adjuncts, with a similar syntactic context.
.. Complement control
In sentences with complement control (as in (13), below), the same c-command
relation is generally observed for the controller– that is, the closest c-commanding
NP– with the same (null) form of PRO:
(13) a. John1 wanted PRO1 to run to the store.
b. John1 told Mary2 PRO*1/2 to run to the store.
In previous studies, children have exhibited adultlike behavior for complement
control before adjunct control (Hsu, Cairns & Fiengo 1985; McDaniel, Cairns &
Hsu 1991; Cairns etal. 1994); however, children still accepted a wider range of in-
terpretations initially, albeit at a younger age than for adjunct control. is suggests
that children do not infer the antecedent of PRO from sentences with complement
control, since the non-adultlike interpretations would provide incorrect evidence
in the input in the same way as discussed above for adjunct control.
A generalization strategy also makes several assumptions: rst, if children did
infer the antecedent for complement control, then the same inference must not also
be made for adjunct control. Next, if children generalize from complement control
to adjunct control, this assumes that the relevant generalization is not made in the
reverse direction, from adjunct control to complement control. Finally, adjunct
control and complement control share various features; if children do generalize the
correct features, then they must avoid generalizing others (e.g. attachment height
or verb form).
Adjunct control and the poverty of the stimulus 
ese arbitrary assumptions about what is generalized suggest that children do
not generalize from complement control to adjunct control, at least for a property
like the antecedent of PRO.
.. Finite adjuncts
For the purposes of identifying the controller, nite adjuncts have the same attach-
ment height as non-nite adjuncts, as demonstrated by the co-reference in (14)
between her and Mary:
(14) John1 called her2 before he1 met Mary2 at the store.
erefore, if children could acquire the attachment height for nite adjuncts from
the linguistic input, then this might then be generalized to non-nite adjuncts.
However, the evidence needed for attachment height with nite adjuncts has
the same problems discussed above for non-nite adjuncts– for example, evidence
in the form of binding relations across clauses is unlikely to occur in the input,
falling short of explaining how attachment height is acquired in general.
Additionally, the same assumptions are made for nite adjuncts as the ones
outlined above for complement control: if children did infer attachment height
for nite adjuncts, then the same inference must not also be made for non-nite
adjuncts. Next, if children did generalize from nite adjuncts to non-nite adjuncts,
this assumes that the relevant generalization is not made in the reverse direction,
from non-nite adjuncts to nite adjuncts. Finally, nite adjuncts and non-nite
adjuncts share various features; if children do generalize the correct features, then
they must avoid generalizing other ones (e.g. the antecedent of the adjunct subject,
or the verb form).
For example, the subject in nite adjuncts can grammatically corefer with any
sentence-internal NP (barring contexts that would result in a Principle C violation,
as in (10)), or sentence-external NP. Based on the input distribution in CHILDES
(MacWhinney 2000), these interpretations are realized in the linguistic input
(Table4), with relatively matched frequencies for internal and external anteced-
ents.9 erefore, generalization from the antecedent of subjects in nite adjuncts
would result in the wrong conclusion about adjunct PRO.
ese concerns suggest that children do not generalize a feature like attachment
height from nite adjuncts to non-nite adjuncts. Furthermore, the sources of evi-
dence considered above are not evidence for the abstract features of control (lexical
learning and adjunct reanalysis), or they are not reliable (binding across clauses
. Finite adjuncts were identied by searching for each complementizer followed by a pronoun,
a bare noun, or determiner, and coded by hand for the antecedent.
 Juliana Gerard
and negative evidence from the size principle). Nevertheless, all children acquire
a grammar with the correct attachment height and c-command by the controller.
ese abstract features must then be innate, i.e. part of Universal Grammar.
. Universal grammar
Even though adjunct control itself is available in the input, evidence is not availa-
ble for the main syntactic components of adjunct control, attachment height and
c-command by the controller. is suggests that these properties are part of UG,
which has implications for the hypothesis space of possible grammars considered
by a learner. In particular, a learner will only consider the grammars where these
properties are adultlike.10
If evidence for attachment height and c-command is not in the input, this
raises the question of what is in the input. What features of adjunct control must
be acquired? Predictions are also made for children’s acquisition which may be
tested empirically.
. Role of the input
If the properties of adjunct control are abstract universals, then the input is needed
for any variation. For example, niteness distinguishes non-nite adjuncts from
nite adjuncts and conjoined clauses. If tense can be used as a cue for the type
of dependency, then it may be one of the features to acquire from the input for
adjunct control.
. c8-fn9A reviewer notes that these two properties alone may not be sucient for obligatory control, as
a learner must also recognize that control occurs in non-nite clauses. However, acquisition from
the perspective of the learner does not distinguish between adjunct control contexts and non-nite
adjuncts: in the input, a learner will perceive a non-nite adjunct with an empty subject, prompting
a search for an antecedent to the subject. e task for the learner is to recognize the non-nite
context, while UG identies the antecedent in this context as the closest c-commanding NP.
Table4. Frequencies of nite adjunct subjects in the input, by complementizer and
subject antecedent. Counts are from the CHILDES transcripts discussed in Section3
Tota l Co-reference with
Main clause subject Other internal referent External referent
aer 346 193 25 128
before 717 383 83 251
while 307 92 30 185
Adjunct control and the poverty of the stimulus 
.. Finiteness
Compared to the abstract syntactic properties, morphological tense is more ac-
cessible in the input: the contrast between nite and non-nite verbs is generally
realized overtly, and is not limited to adjunct control. For example, the contrast
between nite and non-nite clauses is also relevant for complement control, as
well as syntactic bootstrapping for verb learning (Harrigan, Hacquard & Lidz 2019).
An additional cue to adjunct control is the form of the subject– while -
nite adjuncts generally have an overt subject, in non-nite adjuncts the subject is
not pronounced (from the point of view of the learner). erefore, a learner may
look for an empty subject or non-nite morphology to identify an adjunct con-
trol dependency. Of course, this raises an additional question: would these cues
be weighted dierently in a language depending on their availability or reliability
(Kempe & MacWhinney 1999)? For example, for languages which allow the sub-
ject to be dropped (e.g. pro drop, topic drop), the empty subject would not be as
helpful for identifying an adjunct control dependency, since nite verbs may also
appear without a subject (Haegeman 2000; Holmberg, Nayudu & Sheehan 2009;
Huang 1984; Sundaresan 2014; Nunes 2014; Wu 1992). However, the probability of
an empty subject is much higher in a non-nite clause than in a nite clause, even
for languages which allow subject drop (since the probability of an overt subject
in a non-nite clause is essentially zero). Children are sensitive to these contrasts
in probability (for a review see Newport 2016). erefore, if children use tense or
subject form as a cue for adjunct control, then cross-linguistic predictions may be
made for acquisition based on (a) the availability of tense (for languages which
express tense overtly vs covertly), and (b) the reliability for predicting an empty
subject in non-nite vs nite verbs.
For example, the cue to retrieve an antecedent is the missing subject, but if a
missing subject may occur in a nite or non-nite clause (as in languages which
allow the subject to be dropped), then tense information is also needed to identify
the grammatical antecedent. Meanwhile, in languages which do not allow subject
drop, if empty subjects are associated with non-nite clauses then an antecedent
may be identied without tense information. If the retrieval mechanism is deployed
as soon as possible, then children’s parsing strategies may vary depending on these
cues (to be tested in future research).
.. Complementizers
Another feature of adjunct control which varies cross-linguistically is the specic
complementizers and the clauses that they select. For example, without may appear
in a nite frame in both German and Dutch, but not in English:
 Juliana Gerard
Non-nite
(15) a. John cooks without PRO sleeping
b. Fritz kocht ohne PRO zu schlafen
Fritz
cooks
without PRO to sleep
“Fritz cooks without sleeping” adapted from Ller (1995)
c. Hij gaf, zonder PRO het te
weten,
het juiste
antwoord
He
gave,
without PRO it to know the right answer
“He gave, without knowing it, the right answer.
adapted from dutchgrammar.com
Finite
(16) a. *John cooks without that he sleeps
b. Fritz kocht ohne dass er
schlä
Fritz
cooks
without that he sleeps
“Fritz cooks without ‘that he sleeps’”
c. Hij gaf, zonder dat hij het wist, het juiste
antwoord
He
gave,
without that he it knew the right answer
“He gave, without ‘that he knew it,’ the right answer.
erefore, children must learn the form for each complementizer, and whether it
selects a nite clause, non-nite, or both. Alternatively, some complementizers may
be categorized based on a particular feature to be learned in groups, although that
would introduce the additional question of how this feature is acquired.
e issue of adjunct complementizers is relevant for any acquisition account
of adjunct control: complementizers must be distinguished from conjoined clauses
and complement clauses. If attachment height is an expected (innate) contrast, then
the learning problem will involve identifying the complementizer forms and their
selected clauses, and other lexical and semantic properties as discussed by Cairns
etal. (1994). is has implications, then, for children’s competence and the expected
developmental trajectory. ese are discussed further in the following sections.
. Competence and acquisition
In previous studies on the acquisition of adjunct control, children’s behavior has
generally been attributed to a non-adultlike grammar. However, if both attachment
height and c-command by the controller are already part of UG, then these proper-
ties of adjunct control would not need to be acquired from the input. Instead, the
input would be used for mapping overt forms (like tense and complementizers) to
the abstract structure in UG. is predicts that children might sometimes make the
wrong mappings, but no stage should be observed with non-adultlike attachment
height or a non-adultlike controller.
Adjunct control and the poverty of the stimulus 
is prediction presents a puzzle for explaining children’s non-adultlike be-
havior in previous studies. If children’s competence was adultlike, why would they
access non-adultlike interpretations?
One option is that children’s non-adultlike interpretations were indeed due
to a non-adult grammar, and the adult grammar is acquired independent of the
linguistic input, via language-specic maturation (Manzini & Wexler 1987; Wexler
& Manzini 1987; Wexler 1990; Wexler 1992; Wexler 2019). is is consistent with
children’s behavior, as well as the lack of evidence in the input.
Another consideration is that children’s interpretations reect their linguis-
tic competence ltered through an immature parser. at is, children may have
acquired the adult grammar, but processing limitations may interfere with the
deployment of this grammatical knowledge in an experimental setting. ese
processing limitations may involve parsing mechanisms for antecedent retrieval
(Gerard etal. 2017), as well as the complexity of the task itself (Gerard etal. 2018).
For children to access adultlike interpretations consistently, development will then
involve domain-general memory mechanisms, which can interface with language
and with other specic domains (Nairne 1988; Nairne 1990). is development is
likely to aect children’s interpretations (for reviews, see Feigenson 2007; Cowan
2001; Courage & Cowan 2008).11
Finally, other processes may be more sensitive to specic input frequencies,
as discussed above for potential cues for adjunct control in the input (for further
discussion, see Van Dyke & Johns 2012; Omaki & Lidz 2015; Gerard 2016). For
example, children may not have a strong enough link between the overt forms of
tense or complementizers and the corresponding structures.12 is explanation
may also be given along with an account of limited processing resources: in both
cases, non-adultlike interpretations are due to problems with deploying adultlike
syntactic knowledge. Also, their predictions can be tested in an experimental con-
text (discussed further below).
Importantly, the source of children’s non-adultlike interpretations does not
aect the arguments above about the lack of evidence in the input for attachment
height or a c-commanding controller; for example, children are still likely to ac-
cess non-adultlike interpretations of adjunct control in the input, regardless of the
source of these non-adultlike interpretations.
. See also Frank (1998) on non-adultlike behavior due to processing limitations with lan-
guage-specic development.
. is second option is similar to the account proposed by Cairns etal. (1994) in that adultlike
behavior is achieved by forming adultlike mappings between lexical forms and abstract structure.
 Juliana Gerard
. Predictions for the input
Although most types of adjunct control exhibit subject control, exceptions exist
depending on various aspects of the dependency. To account for this variation,
evidence must be available in the input in some form. For example, in (17), the
controller is the main clause patient, rather than the subject:
(17) a. John1 thanked Mary2 for PRO*1/2 running to the store.
b. John1 was thanked by Mary2 for PRO1/*2 running to the store.
is exception with the complementizer for is observed across languages with the
corresponding complementizer. is means that some aspect of the meaning of for
is associated with control by the patient, or that evidence in the input is available
for this exception.
To test this prediction, an additional corpus search was conducted for non-nite
adjuncts with the complementizer for, using the same methods as described above.
e raw counts are presented in Table5.
Table5. Frequency of antecedents in non-nite for by adults (speech to children) and
children (speech by children) in CHILDES
Tota l Co-reference with
Main clause subject Other internal referent External referent
adult (input) 326 42 281 3
target child 36 8 28 0
e data here raise two main points. First, compared to the other non-nite comple-
mentizers, the adjuncts with for occur at a high frequency (comparable to without
and instead of), and should therefore be more salient than the lower frequency
adjuncts.
Next, unlike the other non-nite complementizers, which occurred in the input
with only subject control interpretations, an overwhelming majority of adjuncts
with for have an object or other internal NP as the controller, as in the following
examples:13
. e search of for followed by the string “ing” also returned utterances such as the following:
(iii) ey’re not for eating.
(iv) Where’re the songs for dancing?
(v) is one’s for something else.
(vi) Mommies are not for hitting.
ese instances were not included in the analysis.
Adjunct control and the poverty of the stimulus 
(18) a. Can you1 scold Jennifer2 for PRO*1/2 stepping on the truck?
b. What did Aunt Carey1 buy you2 at the store for PRO*1/2 being a good sharer?
c. You1 yelled at him2 today for PRO*1/2 chewing your slippers.
d. I1 have a little present for you2 for PRO*1/2 coming today.
If children are sensitive to dierent distributions of antecedents, this is the kind of
striking contrast that might be relevant for acquisition. is would be in compar-
ison to a contrast between strict subject control and e.g. a discourse bias for the
subject interpretation, which would only be detectable in a minority of instances.
However, while some variety is observed within the instances of for adjuncts,
70% of the instances occurred in the frame ‘thank you for ___ing,’ as in:
(19) a. ank you for helping me.
b. ank you for letting Mommy nish her breakfast.
c. ank you for carrying socks.
is frequent frame may start out as a larger chunk, to be linked later to the for
non-nite frame. Meanwhile, the discourse contexts for the utterances in (18)
strongly support a patient interpretation for the adjunct subject. ese utterances,
along with the instances with the patient as the subject, may provide the relevant
evidence against strict subject control for for adjuncts.
is predicts, however, that similar evidence will be available in the input for
other languages. It also predicts that children would treat for adjuncts like the
other non-nite adjuncts until the relevant evidence is available. Alternatively, the
meaning of for as a complementizer may be associated already with the patient
antecedent, so that identifying the complementizer form-meaning mapping would
be sucient for acquisition; this would involve additional language-specic infor-
mation to be specied in UG.
. Discussion
is paper has considered the options for acquiring adjunct control. Although
adjunct control is available in the input, this is not sucient for acquiring the main
syntactic properties of adjunct control. Observing instances of adjunct control di-
rectly may provide information about overt features in the dependency, but not
abstract features like the correct attachment height of the adjunct or the controller
as the closest c-commanding NP. Similar issues arise when considering the possi-
bility of generalizing from other structures, which involve arbitrary assumptions
about generalization.
Without evidence in the input for these key components of adjunct control,
they must be innate– considered here as principles in UG. is argument from the
 Juliana Gerard
poverty of the stimulus instead involves a dierent type of evidence in the input
for acquiring adjunct control, and makes further predictions about the input. e
following sections consider the implications of this account– for control, for other
dependencies, and for acquisition.
. Other types of control
Accounting for the adjunct control as a dependency requires a syntactically de-
ned locality constraint. is is supported by crosslinguistic judgments, as well as
in experiments which control for the discourse context (Parker, Lago & Phillips
2015; Kwon & Sturt 2014; Kush & Dillon 2020; Broihier & Wexler 1995; Adler 2006;
Gerard etal. 2018; but see Green 2018b). ese judgments are also represented in
the linguistic input, which consists nearly exclusively of subject control.
ese instances of adjunct control are generally considered to be obligatory
control in that they require a local antecedent. Meanwhile, non-obligatory control
is also observed in temporal adjuncts (Williams 1992; Landau 2015; Landau 2017;
Green 2018a) as in (4) and (5), repeated below as (20) and (21):
(20) It was good to call aer PRO drawing a picture.
(21) e ower wilted aer PRO drawing a picture of it.
As observed in Section3, both of these occur in the input, and are produced by
children. However, there are several reasons not to consider these occurrences as
evidence in the input for non-obligatory control.
In previous studies, children have accepted an external antecedent for sentences
with obligatory control (McDaniel, Cairns & Hsu 1991; Cairns etal. 1994; Broihier
& Wexler 1995; Adler 2006). erefore, development must involve a change to
strict subject interpretations for obligatory control, while still allowing external
interpretations for non-obligatory control as in (20) and (21). If children’s external
interpretations are due to a non-adult grammar, then these interpretations in the in-
put are of type data2 in Figure2. With a free reference grammar, sentences like (20)
and (21) may also be parsed as data2; that is, these sentences would be consistent
with the non-adult grammar and would not provide evidence for non-obligatory
control until aer the adult grammar is acquired.
Meanwhile, regardless of the source of children’s non-adultlike external in-
terpretations, they are likely to occur at comparable frequencies to the counts in
Table2. erefore, if a learner uses instances like those in Table2 as evidence
for non-obligatory control, then non-adultlike external interpretations are just as
likely to provide incorrect evidence against obligatory control. Future research will
Adjunct control and the poverty of the stimulus 
further examine these implications for acquiring obligatory and non-obligatory
control (Landau 2021).
. Other dependencies
is paper discusses the acquisition of adjunct control based on a hierarchical rela-
tion (c-command by the controller) and attachment height. In addition to adjunct
control, other dependencies are also dened in terms of hierarchical relations, so
much of the logic discussed here may be applied more generally.
For example, for any referential dependency, an antecedent must be identi-
ed to resolve the dependency. Consider a syntactic dependency between X and
Y, where the grammatical antecedent may be identied by some constraint (e.g.
c-command and/or locality):
(22)
X [ Y ]
c-command
If the relevant constraint has not yet been acquired, then an alternative strategy is
needed to resolve the dependency; for example, by retrieving an antecedent from
the discourse:
(23)
V
W X
[ Y ]
discourse
Additionally, there must be evidence available in the input to (eventually) acquire
the relevant syntactic constraint. Otherwise, without this evidence, some aspect of
the dependency must be available in UG; this will make further predictions similar
to adjunct control about factors like exceptions, experimental contexts, etc.
Languages vary in their inventories of syntactic dependencies, with some de-
pendencies observed more universally than others. Positing a domain-specic fea-
ture in UG may account for more widely observed dependencies, while evidence
is needed in the input in other cases. Arguments identifying which features are in
UG oen (reasonably) appeal to this universality, or lack thereof; this paper is con-
cerned also with the transparency of a given feature in the input: for abstract prop-
erties which are not directly observable from the linear input, evidence for these
properties may be more elusive, even when the relevant structures are available in
the input. Attachment height and c-command are examples of such properties (with
the same logic for locality in other frameworks).
 Juliana Gerard
. Role of the argument of the poverty of the stimulus
is paper presents an argument from the poverty of the stimulus that the abstract
components of adjunct control are innate. Evidence for these components does not
occur in the input, so they must be available from another source. If attachment
height and the controller are part of UG, then common features of control across
languages may be explained without requiring redundancy in the input.
More broadly, based on the type of evidence that is not available and because
these features of control are not learned, the conclusions about evidence in the input
are applicable to linguistic dependencies more generally: if the actual elements of a
dependency are not reliable for inferring the properties of the dependency, then a
dierent form of evidence is needed for these properties. is was the case for ad-
junct control, as children’s non-adultlike interpretations of adjunct PRO were likely
to provide incorrect evidence about the adult grammar. Similarly, non-adultlike in-
terpretations have also been observed for other types of anaphora (Chien & Wexler
1990; McKee 1992; for a review see Conroy etal. 2009), as well as A movement
(Manzini & Wexler 1987; Ortelli 2012; Mateu 2016, i.a.) and A-bar movement
(Tavakolian 1981; Friedmann, Belletti & Rizzi 2009; Adani etal. 2010, inter alia;
but see Hamburger & Crain 1982; Gagliardi, Mease & Lidz 2016).
For many of these general phenomena, innate components have been proposed,
based on the poverty of the stimulus. Meanwhile, children’s non-adultlike behav-
ior is oen accounted for by a non-adult grammar. ese accounts may achieve
descriptive adequacy for children’s non-adultlike behavior; however, if evidence is
not available in the input for the non-adult grammar and for the transition to the
adult grammar, then this casts doubt on the explanatory adequacy of the grammar.
If both forms of evidence are not available, then either a dierent non-adult gram-
mar or extragrammatical sources are needed to account for children’s behavior.
. Conclusion
is paper considered how adjunct control is acquired and compared dierent
sources of evidence in the linguistic input. ese options did not provide evidence
for the key grammatical components of adjunct control, suggesting that these com-
ponents are innate, with other more overt forms of evidence in the input. Future
research will further investigate the predictions of this evidence, as well as the more
general implications for the content of UG.
Adjunct control and the poverty of the stimulus 
References
Adania, Flavia & Stegenwaller-Schütz, Maja & Niesel, Talea. 2017. e peaceful co-existence
of input frequency and structural intervention eects on the comprehension of complex
sentences in German-speaking children. Frontiers in Psychology 8.
https://doi.org/10.3389/fpsyg.2017.01590
Adani, Flavia, van der Lely, Heather K., Forgiarini, Matteo & Guasti, Maria Teresa. 2010. Gram-
matical feature dissimilarities make relative clauses easier: A comprehension study with
Italian children. Lingua 120(9): 2148–2166. https://doi.org/10.1016/j.lingua.2010.03.018
Adler, Allison Nicole. 2006. Syntax and Discourse in the Acquisition of Adjunct Control. PhD
dissertation, MIT.
Ambridge, Ben. 2019. Against stored abstractions: A radical exemplar model of language acqui-
sition. First Language 40(5–6): 509–559. https://doi.org/10.1177/0142723719869731
Ambridge, Ben, Pine, Julian Mark, Rowland, Caroline F. & Young, Chris R. 2008. e eect of
verb semantic class and verb frequency (entrenchment) on children’s and adults’ graded
judgements of argument-structure overgeneralization errors. Cognition 106(1): 87–129.
https://doi.org/10.1016/j.cognition.2006.12.015
Arunachalam, Sudha, Escovar, Emily, Hansen, Melissa & Waxman, Sandra R. 2011. Verb learn-
ing from syntax alone at 21 months. In Proceedings of the 35th Annual Boston University
Conference on Language Development, Nick Danis, Kate Mesh & Hyunsuk Sung (eds), 21–
24. Somerville MA: Cascadilla Press.
Baker, Carol L. 1979. Syntactic theory and the projection problem. Linguistic Inquiry 10(4):
533–581.
Banbury, Simon P., Macken, William J., Tremblay, Sébastien & Jones, Dylan M. 2001. Auditory
distraction and short-term memory: Phenomena and practical implications. Human Factors
43(1): 12–29. https://doi.org/10.1518/001872001775992462
Belletti, Adriana. 2017. Internal grammar and children’s grammatical creativity against poor
inputs. Frontiers in Psychology 8. https://doi.org/10.3389/fpsyg.2017.02074
Berwick, Robert C. 1985. e Acquisition of Syntactic Knowledge. Cambridge MA: e MIT Press.
https://doi.org/10.7551/mitpress/1074.001.0001
Boyle, Robyn & Coltheart, Veronika. 1996. Eects of irrelevant sounds on phonological coding
in reading comprehension and short-term memory. e Quarterly Journal of Experimental
Psychology. A, Human Experimental Psychology 49(2): 398–416.
https://doi.org/10.1080/713755630
Broihier, Kevin & Wexler, Kenneth. 1995. Children’s acquisition of control in temporal adjuncts.
In Papers on Language Processing and Acquisition [MIT Working Papers in Linguistics 26],
Carson Schütze, Jennifer Ganger & Kevin Broihier (eds), 193–220. Cambridge MA: MIT.
Cairns, Helen S., McDaniel, Dana, Hsu, Jennifer R. & Rapp, Michele. 1994. A longitudinal study
of principles of control and pronominal reference in child English. Language 70(2): 260–288.
https://doi.org/10.2307/415829
Chien, Yu-Chin & Wexler, Kenneth. 1990. Children’s knowledge of locality conditions in bind-
ing as evidence for the modularity of syntax and pragmatics. Language Acquisition 1(3):
225–295. https://doi.org/10.1207/s15327817la0103_2
Chomsky, Noam. 1965. Aspects of the eory of Syntax. Cambridge MA: e MIT Press.
Chomsky, Noam. 1981. Lectures on Government and Binding: e Pisa Lectures [Studies in Gen-
erative Grammar 9]. Berlin: De Gruyter.
 Juliana Gerard
Conroy, Anastasia, Takahashi, Eri, Lidz, Jerey & Phillips, Colin. 2009. Equal treatment for
all antecedents: How children succeed with Principle B. Linguistic Inquiry 40(3): 446–486.
https://doi.org/10.1162/ling.2009.40.3.446
Courage, Mary & Cowan, Nelson. 2008. e Development of Memory in Infancy and Childhood.
Hove: Psychology Press. https://doi.org/10.4324/9780203934654
Cowan, Nelson. 2001. e magical number 4 in short-term memory: A reconsideration of men-
tal storage capacity. Behavioral and Brain Sciences 24(1): 87–114.
https://doi.org/10.1017/S0140525X01003922
Crain, Stephen & McKee, Cecile. 1985. e acquisition of structural restrictions on anaphora.
In Proceedings of NELS 15, Stephen Berman, Jae-Woong Choe & Joyce McDonough (eds),
94–110. Amherst MA: GLSA.
Crain, Stephen & ornton, Rosalind. 1998. Investigations in Universal Grammar: A Guide to
Experiments on the Acquisition of Syntax and Semantics. Cambridge MA: e MIT Press.
Feigenson, Lisa. 2007. Continuity of format and computation in short-term memory develop-
ment. In Short-and Long-Term Memory in Infancy: Taking the First Steps towards Memory,
Lisa M. Oakes & Patricia J. Bauer (eds), 51–74. Oxford: OUP.
Fisher, Cynthia, Gertner, Yael, Scott, Rose M. & Yuan, Sylvia. 2010. Syntactic bootstrapping. Wiley
Interdisciplinary Reviews: Cognitive Science 1(2): 143–149. https://doi.org/10.1002/wcs.17
Fodor, Janet D. 1989. Learning the periphery. In Learnability and Linguistic eory, Robert J.
Matthews & William Demopoulos (eds), 129–154. Dordrecht: Springer.
https://doi.org/10.1007/978-94-009-0955-7_7
Fodor, Janet D. 1994. How to obey the Subset Principle: Binding and locality. In Syntactic e-
ory and First Language Acquisition, Vol. 2: Binding, Dependencies and Learnability, Barbara
Lust, Gabriella Hermon & Jaklin Kornlt (eds), 429–451. Hillsdale NJ: Lawrence Erlbaum
Associates.
Frank, Robert. 1998. Structural complexity and the time course of grammatical development.
Cognition 66(3): 249–301. https://doi.org/10.1016/S0010-0277(98)00024-9
Friedmann, Naama, Belletti, Adriana, & Rizzi, Luigi. 2009. Relativized relatives: Types of inter-
vention in the acquisition of A-bar dependencies. Lingua 119(1): 67–88.
https://doi.org/10.1016/j.lingua.2008.09.002
Gagliardi, Annie, Mease, Tara M. & Lidz, Jerey. 2016. Discontinuous development in the ac-
quisition of ller-gap dependencies: Evidence from 15- and 20-month-olds. Language Ac-
quisition 23(3): 234–260. https://doi.org/10.1080/10489223.2015.1115048
Gerard, Juliana. 2016. e Acquisition of Adjunct Control: Grammar and Processing. PhD dis-
sertation, University of Maryland. https://doi.org/10.13016/M2621K
Gerard, Juliana, Lidz, Jerey, Zuckerman, Shalom & Pinto, Manuela. 2017. Similarity-based in-
terference and the acquisition of adjunct control. Frontiers in Psychology 8.
https://doi.org/10.3389/fpsyg.2017.01822
Gerard, Juliana, Lidz, Jerey, Zuckerman, Shalom & Pinto, Manuela. 2018. e acquisition of
adjunct control is colored by the task. Glossa: A Journal of General Linguistics 3(1): 75.
https://doi.org/10.5334/gjgl.547
Gertner, Yael, Fisher, Cynthia & Eisengart, Julie. 2006. Learning words and rules: Abstract knowl-
edge of word order in early sentence comprehension. Psychological Science 17(8): 684–691.
https://doi.org/10.1111/j.1467-9280.2006.01767.x
Gold, E. Mark. 1967. Language identication in the limit. Information and Control 10(5): 447–474.
https://doi.org/10.1016/S0019-9958(67)91165-5
Adjunct control and the poverty of the stimulus 
Goodluck, Helen. 1981. Children’s grammar of complement-subject interpretation. In Language
Acquisition and Linguistic eory, Susan Tavakolian (ed.), 139–166. Cambridge MA: e
MIT Press.
Goodluck, Helen. 2001. e nominal analysis of children’s interpretations of adjunct pro clauses.
Language 77(3): 494–509. https://doi.org/10.1353/lan.2001.0159
Goodluck, Helen & Behne, Dawn. 1992. Development in control and extraction. In eoretical
Issues in Language Acquisition, Jürgen Weissenborn, Helen Goodluck & omas Roeper
(eds), 151–171. Hillsdale NJ: Lawrence Erlbaum Associates.
Gordon, Peter C., Hendrick, Randall & Johnson, Marcus. 2001. Memory interference during
language processing. Journal of Experimental Psychology: Learning, Memory, and Cognition
27(6): 1411–1423.
Gordon, Peter C., Hendrick, Randall, & Johnson, Marcus. 2004. Eects of noun phrase type on
sentence complexity. Journal of Memory and Language 51(1): 97–114.
https://doi.org/10.1016/j.jml.2004.02.003
Gordon, Peter C., Hendrick, Randall, Johnson, Marcus, & Lee, Yoonhyoung. 2006. Similari-
ty-based interference during language comprehension: Evidence from eye tracking dur-
ing reading. Journal of Experimental Psychology: Learning, Memory, and Cognition 32(6):
1304–1321. https://doi.org/10.1037/0278-7393.32.6.1304
Gordon, Peter C., & Lowder, Matthew W. 2012. Complex sentence processing: A review of e-
oretical Perspectives on the Comprehension of Relative Clauses: Review of Complex Sen-
tence Processing. Language and Linguistics Compass 6(7): 403–415.
https://doi.org/10.1002/lnc3.347
Green, Jerey. 2018a. Adjunct Control: Syntax and Processing. PhD dissertation, University of
Maryland.
Green, Jerey. 2018b. Non-subject control of temporal adjuncts. In Proceedings of the 53rd An-
nual Meeting of the Chicago Linguistic Society 54. Chicago IL: CLS.
Gries, Stefan T. 2008. Dispersions and adjusted frequencies in corpora. International Journal of
Corpus Linguistics 13(4): 403–437. https://doi.org/10.1075/ijcl.13.4.02gri
Gries, Stefan T. 2010. Dispersions and adjusted frequencies in corpora: Further explorations. In
Corpus-linguistic Applications, Stefan T. Gries, Stefanie Wul & Mark Davies (eds), 197–212.
Leiden: Brill Rodopi. https://doi.org/10.1163/9789042028012
Grimshaw, Jane & Pinker, Steven. 1989. Positive and negative evidence in language acquistion.
Behavioral and Brain Sciences 12(2): 341–342. https://doi.org/10.1017/S0140525X00048950
Grodzinsky, Yosef. 1989. e language learner: A trigger-happy kid? Behavioral and Brain
Sciences 12(2): 342–343. https://doi.org/10.1017/S0140525X00048962
Guasti, Maria Teresa. 2017. Language Acquisition: e Growth of Grammar. Cambridge MA:
e MIT Press.
Haegeman, Liliane. 2000. Adult null subjects in non pro-drop languages. In e Acquisition of
Syntax: Studies in Comparative Developmental Linguistics, Marc-Ariel Friedemann & Luigi
Rizzi (eds), 129–169. New York NY: Pearson Education.
CIT0429Hamburger, Henry & Crain, Stephen. 1982. Relative acquisition. Language Development 1: 245–274.
Harrigan, Kaitlyn, Hacquard, Valentine & Lidz, Jerey. 2019. Hope for syntactic bootstrapping.
Language 95(4): 642–682. https://doi.org/10.1353/lan.2019.0069
Heinz, Jerey & Riggle, Jason. 2011. Learnability. In e Blackwell Companion to Phonology, 5
Volume Set, Vol. 1, Chapter 2, Marc van Oostendorp, Colin J. Ewen, Elizabeth V. Hume &
Keren Rice (eds), 54–78. Malden MA: Wiley & Sons.
https://doi.org/10.1002/9781444335262.wbctp0003
 Juliana Gerard
Holmberg, Anders, Nayudu, Aarti & Sheehan, Michelle. 2009. ree partial null-subject lan-
guages: A comparison of Brazilian Portuguese, Finnish and Marathi*. Studia Linguistica
63(1): 59–97. https://doi.org/10.1111/j.1467-9582.2008.01154.x
Hsu, Jennifer R., Cairns, Helen S. & Fiengo, Robert. W. 1985. e development of grammars
underlying children’s interpretation of complex sentences. Cognition 20(1): 25–48.
https://doi.org/10.1016/0010-0277(85)90003-4
Huang, C. James. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry
15(4): 531–574.
Huang, Yi T., Zheng, Xiaobei, Meng, Xiangzhi & Snedeker, Jesse. 2013. Children’s assignment
of grammatical roles in the online processing of Mandarin passive sentences. Journal of
Memory and Language 69(4): 589–606. https://doi.org/10.1016/j.jml.2013.08.002
Janke, Vikki. 2018. Discourse eects on older children’s interpretations of complement control
and temporal adjunct control. Language Acquisition 25(4): 366–391.
https://doi.org/10.1080/10489223.2017.1359271
Janke, Vikki & Bailey, Laura R. 2017. Eects of discourse on control. Journal of Linguistics 53(3):
533–565. https://doi.org/10.1017/S0022226716000281
Janke, Vikki & Perovic, Alexandra. 2017. Contrasting complement control, temporal adjunct
control and controlled verbal gerund subjects in ASD: e role of contextual cues in refer-
ence assignment. Frontiers in Psychology 8. https://doi.org/10.3389/fpsyg.2017.00448
Kempe, Vera & MacWhinney, Brian. 1999. Processing of morphological and semantic cues in
Russian and German. Language and Cognitive Processes 14(2): 129–171.
https://doi.org/10.1080/016909699386329
Kush, Dave & Dillon, Brian. 2020. Principle B Constrains the Processing of Cataphora: Evidence
for Syntactic and Discourse Predictions Open Science Framework.
https://doi.org/10.31219/osf.io/mxtc6
Kwon, Nayoung & Sturt, Patrick. 2014. e use of control information in dependency formation:
An eye-tracking study. Journal of Memory and Language 73: 59–80.
https://doi.org/10.1016/j.jml.2014.02.005
Landau, Idan. Forthcoming. A Selectional eory of Adjunct Control.
Landau, Idan. 2015. A Two-tiered eory of Control. Cambridge MA: e MIT Press.
https://doi.org/10.7551/mitpress/9780262028851.001.0001
Landau, Idan. 2017. Adjunct control depends on Voice. In A Pesky Set: Papers for David Pesetsky,
Claire Halpert, Hadas Kotek & Coppe van Urk (eds), 93–102. Cambridge MA: e MIT Press.
Lidz, Jerey & Gagliardi, Annie. 2015. How nature meets nurture: Universal grammar and sta-
tistical learning. Annu. Rev. Linguist 1(1): 333–353.
https://doi.org/10.1146/annurev-linguist-030514-125236
Lukyanenko, Cynthia, Conroy, Aanastasia, & Lidz, Jerey. 2014. Is she patting Katie? Constraints
on pronominal reference in 30-month-olds. Language Learning and Development 10(4):
328–344. https://doi.org/10.1080/15475441.2013.853529
Lust, Barbara, Eisele, Julie, & Mazuka, Reiko. 1992. e Binding eory module: Evidence from
rst language acquisition for Principle C. Language 68(2): 333–358.
https://doi.org/10.2307/416944
MacWhinney, Brian. 2000. e CHILDES Project: e Database, Vol. 2. New York NY: Psychol-
ogy Press.
Manzini, M. Rita & Wexler, Kenneth. 1987. Parameters, Binding eory, and learnability. Lin-
guistic Inquiry 18(3): 413–444.
Adjunct control and the poverty of the stimulus 
Mateu, Victoria E. 2016. Intervention Eects in the Acquisition of Raising and Control: Evidence
from English and Spanish. PhD dissertation, UCLA.
McDaniel, Dana, Cairns, Helen S. & Hsu, Jennifer R. 1991. Control principles in the grammars
of young children. Language Acquisition 1(4): 297–335.
https://doi.org/10.1207/s15327817la0104_1
McKee, Cecile. 1992. A comparison of pronouns and anaphors in Italian and English acquisition.
Language Acquisition 2(1): 21–54. https://doi.org/10.1207/s15327817la0201_2
Mintz, Toben H., Newport, Elissa L. & Bever, omas G. 2002. e distributional structure of
grammatical categories in speech to young children. Cognitive Science 26(4): 393–424.
https://doi.org/10.1207/s15516709cog2604_1
Naigles, Letitia. 1990. Children use syntax to learn verb meanings. Journal of Child Language
17(2): 357–374. https://doi.org/10.1017/S0305000900013817
Nairne, James S. 1988. A framework for interpreting recency eects in immediate serial recall.
Memory & Cognition 16(4): 343–352. https://doi.org/10.3758/BF03197045
CIT0457Nairne, James S. 1990. A feature model of immediate memory. Memory & Cognition 18(3): 251–269.
https://doi.org/10.3758/BF03213879
Newport, Elissa L. 2016. Statistical language learning: Computational, maturational, and linguistic
constraints. Language and Cognition 8(3): 447–461. https://doi.org/10.1017/langcog.2016.20
Nguyen, Emma & Pearl, Lisa. 2018. Do you really mean it? Linking lexical semantic proles and
the age of acquisition for the english passive. In e Proceedings of the 35th West Coast Con-
ference on Formal Linguistics, Wm. G. Bennett, Lindsay Hracs & Dennis Ryan Storoshenko
(eds), 288–295. Somerville MA: Cascadilla Press.
Nguyen, Emma & Pearl, Lisa. 2019. Using developmental modeling to specify learning and
representation of the passive in English children. In Proceedings of the Boston University
Conference on Language Development 43, Megan M. Brown & Brady Dailey (eds), 469–482.
Somerville MA: Cascadilla Press.
Nunes, Jairo. 2014. Adjunct Control and edge features. In Minimalism and Beyond: Radicalizing
the Interfaces [Language Faculty and Beyond 11], Peter Kosta, Steven L. Franks, Teodora
Radeva-Bork & Lilia Schürcks (eds), 79–108. Amsterdam: John Benjamins.
https://doi.org/10.1075/lfab.11
Omaki, Akira & Lidz, Jerey. 2015. Linking parser development to acquisition of syntactic knowl-
edge. Language Acquisition 22(2): 158–192. https://doi.org/10.1080/10489223.2014.943903
Ortelli, Robyn M. 2012. Argument Intervention in the Acquisition of A-movement. PhD dis-
sertation, UCLA.
Parker, Daniel, Lago, Sol & Phillips, Colin. 2015. Interference in the processing of adjunct con-
trol. Frontiers in Psychology 6. https://doi.org/10.3389/fpsyg.2015.01346
Pearl, Lisa. 2019. Poverty of the stimulus without tears. University of California, Irvine. (Preprint).
Pearl, Lisa & Sprouse, Jon. 2013a. Computational models of acquisition for islands. In Experimen-
tal Syntax and Island Eects, Jon Sprouse & Norbert Hornstein (eds), 109–131. Cambridge:
CUP. https://doi.org/10.1017/CBO9781139035309.006
Pearl, Lisa & Sprouse, Jon. 2013b. Syntactic islands and learning biases: Combining experimental
syntax and computational modeling to investigate the language acquisition problem. Lan-
guage Acquisition 20(1): 23–68. https://doi.org/10.1080/10489223.2012.738742
Perfors, Amy, Tenenbaum, Joshua B. & Regier, Terry. 2011. e learnability of abstract syntactic
principles. Cognition 118(3): 306–338. https://doi.org/10.1016/j.cognition.2010.11.001
 Juliana Gerard
Perkins, Laurel, Feldman, Naomi & Lidz, Jerey. 2017. Learning an input lter for argument
structure acquisition. In Proceedings of the 7th Workshop on Cognitive Modeling and Com-
putational Linguistics (CMCL 2017), Ted Gibson, Tal Linzen, Asad Sayeed, Martin van
Schijndel & William Schuler (eds), 11–19. Valencia: Association for Computational Lin-
guistics. https://doi.org/10.18653/v1/W17-07
Phillips, Colin. 2013. On the nature of island constraints, II: Language learning and innateness.
In Experimental Syntax and Island Eects, Jon Sprouse & Norbert Hornstein (eds), 132–158.
Cambridge: CUP. https://doi.org/10.1017/CBO9781139035309.007
Pinker, Susan. 1979. Formal models of language learning. Cognition 7(3): 217–283.
https://doi.org/10.1016/0010-0277(79)90001-5
Pinker, Susan. 2009. Language Learnability and Language Development, with new commentary
by the author, Vol. 7. Cambridge MA: Harvard University Press.
Pinker, Susan. 2013. Learnability and Cognition: e Acquisition of Argument Structure. Cambridge
MA: e MIT Press. https://doi.org/10.7551/mitpress/9700.001.0001
Pullum, Georey K. 2020. Waiting for Universal Grammar. In Current Controversies in Philos-
ophy of Cognitive Science, Adam J. Lerner, Simon Cullen & Sarah-Jane Leslie (eds), 29–43.
New York NY: Routledge. https://doi.org/10.4324/9781003026273-4
Pullum, Georey K. & Scholz, Barbara C. 2002. Empirical assessment of stimulus poverty argu-
ments. e Linguistic Review 18(1–2): 9–50. https://doi.org/10.1515/tlir.19.1-2.9
Regier, Terry & Gahl, Susanne. 2004. Learning the unlearnable: e role of missing evidence.
Cognition 93(2): 147–155. https://doi.org/10.1016/j.cognition.2003.12.003
Roland, Douglas, Dick, Frederic & Elman, Jerey L. 2007. Frequency of basic English grammat-
ical structures: A corpus analysis. Journal of Memory and Language 57(3): 348–379.
https://doi.org/10.1016/j.jml.2007.03.002
Scholz, Barbara C. & Pullum, Georey K. 2006. Irrational nativist exuberance. In Contemporary
Debates in Cognitive Science, Robert Stainton (ed), 59–80. New York NY: John Wiley & Sons.
Sundaresan, Sandhya. 2014. Making sense of silence: Finiteness and the (OC) pro vs. pro distinc-
tion: Commentary on Kissock (2013). Natural Language & Linguistic eory 32(1): 59–85.
https://doi.org/10.1007/s11049-013-9216-6
Sutton, Megan. 2015. Competence and Performance in the Development of Principle C. PhD
dissertation, University of Maryland. https://doi.org/10.13016/M23C9S
Sutton, Megan, Fetters, M., & Lidz, Jerey. 2012. Parsing for Principle C at 30 months. In Proceed-
ings of the 36th Annual Boston University Conference on Language Development 13, Alia K.
Biller, Esther Y. Chung & Amelia E. Kimball (eds), 581–593. Somerville MA: Cascadilla Press.
Tavakolian, Susan. 1981. e conjoined-clause analysis of relative clauses. Language Acquisition
and Linguistic eory, Susan Tavakolian (ed), 167–187. Cambridge MA: e MIT Press.
Tenenbaum, Joshua B. 1999. Bayesian modeling of human concept learning. In Advances in
Neural Information Processing Systems 11, Michael J. Kearns, Sara A. Solla, & David A. Cohn
(eds), 59–68. Cambridge MA: e MIT Press.
Tenenbaum, Joshua B., & Griths, omas L. 2001. Generalization, similarity, and Bayesian
inference. Behavioral and Brain Sciences, 24(4): 629–640.
https://doi.org/10.1017/S0140525X01000061
Tomasello, Michael. 2009. Constructing a Language. Cambridge MA: Harvard University Press.
Van Dyke, Julie A., & Johns, Clinton L. 2012. Memory interference as a determinant of language
comprehension: Interference in comprehension. Language and Linguistics Compass 6(4):
193–211. https://doi.org/10.1002/lnc3.330
Adjunct control and the poverty of the stimulus 
Wang, Felix H. & Trueswell, John C. 2019. Spotting Dalmatians: Children’s ability to discover
subordinate-level word meanings cross-situationally. Cognitive Psychology 114(101226).
https://doi.org/10.1016/j.cogpsych.2019.101226
Warren, Teresa & Gibson, Edward. 2002. e inuence of referential processing on sentence
complexity. Cognition 85 (1): 79–112. https://doi.org/10.1016/S0010-0277(02)00087-2
Warren, Teresa & Gibson, Edward. 2005. Eects of NP type in reading cle sentences in English.
Language and Cognitive Processes 20(6): 751–767.
https://doi.org/10.1080/01690960500051055
Wexler, Kenneth. 1990. On unparsable input in language acquisition. In Language Processing and
Language Acquisition, Lyn Frazier & Jill De Villiers (eds), 105–117. Dordrecht: Springer.
https://doi.org/10.1007/978-94-011-3808-6_4
Wexler, Kenneth. 1992. Some issues in the growth of control. In Control and grammar, Richard
Larson, Sabine Iatridou, Uptal Lahiri, & James Higginbotham (eds), 253–295. New York
NY: Springer. https://doi.org/10.1007/978-94-015-7959-9_8
Wexler, Kenneth. 2019. Arbitrary control instead of obligatory control in temporal adjuncts in
Child Grammar: An ATTRACT analysis. In Linguistic Variation: Structure and Interpreta-
tion, Ludovico Franco & Paolo Lorusso (eds), 691–716. Berlin: De Gruyter.
https://doi.org/10.1515/9781501505201-033
Wexler, Kenneth & Chien, Yu-Chin. 1985. e development of lexical anaphors and pronouns.
Papers and Reports on Child Language Development 24: 138–149.
Wexler, Kenneth & Manzini, M. Rita. 1987. Parameters and learnability in Binding eory. In
Parameter Setting, omas Roeper & Edwin Williams (eds), 41–76. Dordrecht: Springer.
https://doi.org/10.1007/978-94-009-3727-7_3
Williams, Edwin. 1992. Adjunct Control. In Control and Grammar, Richard Larson, Sabine Iatri-
dou, Uptal Lahiri, & James Higginbotham (eds) 297–322. Dordrecht: Springer.
https://doi.org/10.1007/978-94-015-7959-9_9
Wu, Guobin. 1992. Discourse anaphora in Chinese: A rhetorical predicate account. In Yor k Pa -
pers in Linguistics 16, Steve J. Harlow & Anthony R. Warner (eds), 185–202.
Xu, Fei & Tenenbaum, Joshua B. 2007. Word learning as Bayesian inference. Psychological Review
114(2): 245–272. https://doi.org/10.1037/0033-295X.114.2.245
... This point brings up the question of overall frequency. Although structures with adjunct control are not frequent in the input, they are not absent (Gerard 2021). However, if children do not use these instances to learn the adjunct control dependency, then the frequency does not contribute to the acquisition problem. ...
Article
Full-text available
Previous research on 4-6-year-olds' interpretations of adjunct control has observed non-adult-like behavior for sentences like John called Mary before running to the store. Several studies have aimed to identify a grammatical source of children's errors. This study tests the predictions of grammatical and extragrammatical accounts by comparing children's behavior on two truth value judgment tasks: a high-demand task, with a true/false judgment based on event ordering, and a low-demand task, with a true/false judgment based on the color of an item. Children's behavior is more adultlike on the low-demand task, suggesting that children's interpretations may be influenced by extragrammatical factors. Implications are discussed for children's behavior in previous studies and for the role of the linguistic input. ARTICLE HISTORY
Preprint
Full-text available
Previous research on 4-6-year olds' interpretations of adjunct control has observed non-adultlike behavior for sentences like 'John called Mary before running to the store.' Several studies have aimed to identify a grammatical source of children's errors. This study tests the predictions of grammatical and extragrammatical accounts by comparing children's behavior on two truth value judgment tasks: a high demand task, with a true/false judgment based on event ordering; and a low demand task, with a true/false judgment based on the color of an item. Children's behavior is more adultlike on the low demand task, suggesting that children's interpretations may be influenced by extragrammatical factors. Implications are discussed for children's behavior in previous studies and for the role of the linguistic input.
Article
Full-text available
Poverty of the stimulus has been at the heart of ferocious and tear-filled debates at the nexus of psychology, linguistics, and philosophy for decades. This review is intended as a guide for readers without a formal linguistics or philosophy background, focusing on what poverty of the stimulus is and how it’s been interpreted, which is traditionally where the tears have come in. I discuss poverty of the stimulus from the perspective of language development, highlighting how poverty of the stimulus relates to expectations about learning and the data available to learn from. I describe common interpretations of what poverty of the stimulus means when it occurs, and approaches for determining when poverty of the stimulus is in fact occurring. I close with illustrative examples of poverty of the stimulus in the domains of syntax, lexical semantics, and phonology, and discuss the value of identifying instances of poverty of the stimulus when it comes to understanding language development.
Article
Full-text available
The goal of this article is to make the case for a radical exemplar account of child language acquisition, under which unwitnessed forms are produced and comprehended by on-the-fly analogy across multiple stored exemplars, weighted by their degree of similarity to the target with regard to the task at hand. Across the domains of (1) word meanings, (2) morphologically inflected words, (3) n-grams, (4) sentence-level constructions and (5) phonetics and phonology, accounts based on independently-represented abstractions (whether formal rules or prototype categories) fail for two reasons. First, it is not possible to posit abstractions that delineate possible and impossible form; e.g. that (1) rule in pool tables and data tables, but rule out chairs, (2) rule in the past-tense forms netted and bet but rule out * setted and * jet, (3) rule in the bigram f+t but rule out (probabilistically) v+t, (4) rule in both John feared Bill and John frightened Bill but rule out * John laughed Bill, (5) rule in Speaker A but rule out Speaker B as the person who produced a particular word (e.g. Sa’urday). Second, for each domain, empirical data provide evidence of exemplar storage that cannot be captured by putative abstractions: e.g. speakers prefer and/or show an advantage for (1) exemplar variation even within word-meaning ‘category boundaries’, (2) novel inflected forms that are similar to existing exemplars, (3) n-grams that have occurred frequently in the input, (4) SVO sentences with he as SUBJECT and it as OBJECT and (5) repeated productions of ‘the same’ word that are phonologically similar or, better still, identical. An exemplar account avoids an intractable lumping-or-splitting dilemma facing abstraction-based accounts and provides a unitary explanation of language acquisition across all domains; one that is consistent with models and empirical findings from the computational modelling and neuroimaging literature.
Preprint
We tested whether comprehenders can use Binding Principle B (Chomsky, 1981) to guide antecedent search during the processing of cataphoric pronouns. We implemented two self-paced reading experiments using the gender mismatch paradigm (van Gompel & Liversedge, 2003) as an index of active prediction of coreference between a cataphor and a main clause subject. In both experiments, we find gender mismatch effects at the main clause subject when coreference with the cataphor is grammatically acceptable. We do not find comparable gender mismatch effects in conditions where coreference is ruled out by Principle B. Our results are broadly consistent with models in which grammatical constraints guide active referential processing in comprehension. We illustrate how such the parser integrates syntactic and discourse-level information to achieve grammatical sensitivity during incremental referential processing.
Article
Even when children encounter a novel word in the situation of a clear and unique referent, they are nevertheless faced with the problem of semantic uncertainty: when "puziv" refers to a co-present spotted dog, does the word mean Fido, Dalmatian, dog, animal, or entity? Here we explored the extent to which children (3-5 years of age) can reason about a novel word's meaning from information they have gathered cross-situationally, from a series of simple ostensive labeling events ("I see a puziv!"). Of particular interest were the conditions under which children arrive at a subordinate level meaning (e.g., Dalmatian) rather than a basic level meaning (e.g., dog). Experiment 1 showed that children (N = 32) were capable of using lexical contrast and/or mutual exclusivity cross-situationally, such that they arrived at subordinate level meanings only when the words being learned contrasted at the subordinate level, otherwise they strongly preferred basic level meanings (e.g., dog) even when the word had previously referred to subordinate level exemplars (always Dalmatians). Experiment 2 showed that some children in this same age range (N = 20) can also arrive at subordinate level meanings cross-situationally when offered relatively minimal linguistic support ("It's a kind of dog."). The findings are interpreted with respect to current theories of cross-situational word learning, and suggest that word meanings rather than sets of referential exemplars are tracked and used for cross-situational comparison.