AgeofAcquisitionItsNeuralandComputationalMechanisms.pdf

Age of Acquisition: Its Neural and Computational Mechanisms

Arturo E. Hernandez
University of Houston

Ping Li
University of Richmond

The acquisition of new skills over a life span is a remarkable human ability. This ability, however, is
constrained by age of acquisition (AoA); that is, the age at which learning occurs significantly affects the
outcome. This is most clearly reflected in domains such as language, music, and athletics. This article
provides a perspective on the neural and computational mechanisms underlying AoA in language
acquisition. The authors show how AoA modulates both monolingual lexical processing and bilingual
language acquisition. They consider the conditions under which syntactic processing and semantic
processing may be differentially sensitive to AoA effects in second-language acquisition. The authors
conclude that AoA effects are pervasive and that the neural and computational mechanisms underlying
learning and sensorimotor integration provide a general account of these effects.

Keywords: bilingualism, second language acquisition, age of acquisition, sensorimotor learning,
computational modeling

Personal anecdotes and scientific evidence both confirm that it
is important to learn a second language (L2) early. In the 125th
anniversary Special Issue of Science (Kennedy & Norman, 2005),
age of acquisition (AoA) and critical periods were included in a list
of 100 important science questions to be addressed in the next few
decades. Linguists, psychologists, and cognitive scientists have
made significant progress in understanding AoA; however, impor-
tant questions remain unanswered: What neural substrates underlie
AoA, if any? Are AoA effects specific to L2 learning or are they
present in language in general? And, how is AoA reflected in both
linguistic and nonlinguistic domains? In this review, we approach
these questions broadly and consider general computational and
neural principles that may contribute to AoA effects in both
linguistic and nonlinguistic domains.

What Is Age of Acquisition?

AoA, in its broadest sense, refers to the age at which a concept
or skill is acquired. AoA effects have been addressed in at least
three distinct literatures: the age at which skills are acquired in
nonlinguistic domains, the age at which a lexical item is acquired
in monolingual learners, and the age at which L2 learning begins.
In the first and third literatures, researchers have attempted to

understand how early versus late learning affects successful ac-
quisition. This issue is often discussed in terms of a critical period,
or sensitive period, of learning. In the second literature, researchers
study the age at which lexical items are acquired in monolingual
learners and before AoA effects on the processing of these items.
Do these three types of AoA effects share a common mechanism?
If so, what might that mechanism be? Our review attempts to
provide an integrated answer to these questions.

Age of Acquisition in Nonlinguistic Domains

AoA effects have been found in many nonlinguistic domains.
For example, early deprivation or alteration of sensory input leads
to impaired sensory perception in many species. The most well-
known examples involve binocular deprivation during a critical
period leading to a reduction in stereoscopic depth perception
among cats, monkeys, rats, mice, ferrets, and humans (Banks,
Aslin, & Letson, 1975; Fagiolini, Pizzorusso, Berardi, Domenici,
& Maffei, 1994; Harwerth, Smith, Duncan, Crawford, & von
Noorden, 1986; Huang et al., 1999; Issa, Trachtenberg, Chapman,
Zahs, & Stryker, 1999; Olson & Freeman, 1980). Critical periods
are also found in the calibration of auditory maps by visual input
(Brainard & Knudsen, 1998). Moreover, sensory deprivation can
lead to problems in the motor system. For example, disruption of
binocular experience adversely affects smooth pursuit of moving
objects and diminishes stability of the eyes when viewing station-
ary targets (Norcia, 1996). Hence, problems in the sensory domain
lead to abnormalities of motor function.

AoA also affects song learning in birds. Learning generally
occurs in three phases: sensory, sensorimotor, and crystallized
(Brainard & Doupe, 2002). During the sensory period, a bird
listens to the song of a tutor and forms a template in memory. Lack
of exposure to adult song during this phase leads to irregular songs
that contain some species-specific characteristics. During the sen-
sorimotor phase, the bird learns to match the song to the template.
Songs are fine-tuned through practice; auditory feedback is crucial
during this time (note that sensory and sensorimotor phases may
overlap for some birds). In the final, crystallized phase, birds are

Arturo E. Hernandez, Department of Psychology, University of Hous-
ton; Ping Li, Department of Psychology, University of Richmond.

The writing of this article has been made possible by National Institutes
of Health Grant 1 R03HD050313-01 to Arturo E. Hernandez and National
Science Foundation Grant BCS-0131829 to Ping Li.

We gratefully acknowledge our former mentor Elizabeth Bates, whose
many insights on language and cognition have shaped the ideas presented
here. We thank Michael Ullman, Leigh Leasure, and Brian MacWhinney
for helpful comments related to the theoretical issues discussed here. We
also thank Gedeon Deak and Catriona Morrison for comments on previous
versions of this article.

Correspondence concerning this article should be addressed to Arturo E.
Hernandez, Department of Psychology, 126 Heyne Building, University of
Houston, Houston, TX 77204-5022. E-mail: [email protected]

Psychological Bulletin Copyright 2007 by the American Psychological Association
2007, Vol. 133, No. 4, 638 – 650 0033-2909/07/$12.00 DOI: 10.1037/0033-2909.133.4.638

638

mature and can produce a species-specific song, but they often
cannot learn new songs. The fact that early acquisition of birdsong
is characterized by sensory and sensorimotor processing is of
particular importance in this review (see Doupe, Perkel, Reiner, &
Stern, 2005).

Finally, AoA effects have been observed in high-level nonlin-
guistic functions. AoA effects are found in musicians at both the
behavioral and the neural levels. Absolute pitch appears to be
learned by speakers of nontonal languages only before the age of
7 years (Deutsch, Henthorn, Marvin, & Xu, 2006; Trainor, 2005).
In addition, the ability to synchronize motor responses to a visually
presented flashing square differs significantly between groups of
professional musicians as a function of AoA, even when these
groups are matched for years of musical experience, years of
formal training, and hours of current practice (Watanabe, Savion-
Lemieux, & Penhune, 2007). At the neural level, early musical
training correlates with the size of digit representations in motor
regions of the cortex (Elbert, Pantev, Wienbruch, Rockstroh, &
Taub, 1995). Similarly, Schlaug, Jancke, Huang, Staiger, and
Steinmetz (1995) found that the anterior corpus callosum was
larger in musicians than nonmusicians and largest in those who
learned to play before the age of 7 years. Hence, AoA effects on
both behavior and neural representations in the music domain
appear to reflect sensorimotor processing.

AoA effects are generally considered evidence for critical peri-
ods, time windows within which learning outcomes are optimal
and after which the ability to learn drastically decreases. Critical
periods, however, may be only one instantiation of AoA effects. A
crucial aspect of these effects in nonlinguistic domains is that they
impact both sensory and motor systems (for further discussion of
critical and sensitive periods, see Knudsen, 2004).

Age of Acquisition in Monolingual Individuals

Researchers discovered over 30 years ago that early learned
words are processed differently than late learned words (Carroll &
White, 1973; Gilhooly & Watson, 1981); only recently, however,
has this difference attracted significant interests among psycholin-
guists as an AoA effect. Using a number of experimental para-
digms, researchers have shown that the age of word acquisition
significantly affects the speed and accuracy with which a word is
accessed and processed (Barry, Morrison, & Ellis, 1997; Cuetos,
Ellis, & Alvarez, 1999; Ellis & Morrison, 1998; Gerhand & Barry,
1998, 1999; Gilhooly & Gilhooly, 1979; Lewis, 1999; Meschyan
& Hernandez, 2002; Morrison, Chappell, & Ellis, 1997; Morrison
& Ellis, 1995, 2000). Early learned words typically elicit faster
response times than late learned words in word reading, auditory
and visual lexical decision, picture naming, and face recognition.
Researchers have not, however, agreed on the exact mechanisms
underlying this AoA effect. The controversy lies in the identifica-
tion of the locus of AoA effects, in particular, with regard to
whether AoA reflects endogenous properties of the lexicon or
exogenous properties of the learning process. We now turn to the
various theoretical accounts.

Theoretical accounts of age of acquisition. Brown and Watson
(1987) proposed a phonological completeness hypothesis to ac-
count for AoA effects in word learning. In this view, early learned
words are stored and represented holistically, whereas late learned
words are represented in a fragmented fashion and require recon-
struction or reassembly before the whole phonological shape can

be produced. Thus, early learned words are pronounced more
quickly than late learned words. This hypothesis, however, has not
been supported in a number of studies in several domains. First,
the phonological completeness hypothesis has difficulty account-
ing for AoA effects in tasks that do not involve overt naming, such
as face recognition (Moore & Valentine, 1998, 1999) and object
processing (Moore, Smith-Spark, & Valentine, 2004). Second,
reaction times are faster to early than to late learned words when
participants are asked to perform a segmentation task (Monaghan
& Ellis, 2002a). This pattern is contrary to what the hypothesis
predicts. If late learned words are acquired in a fragmented form,
they should be easier to segment than early learned words. These
findings have led researchers to consider alternative hypotheses, in
particular, hypotheses about whether lexical AoA effects are due to
a more general mechanism.

Several general accounts of AoA effects have been proposed
(for a recent review of the literature see Juhasz, 2005). The
cumulative frequency hypothesis maintains that word frequency
consists of additive effects across the lifetime of a word. Hence,
early learned words will be encountered more times across many
years of use than late learned words, even if they are low in
frequency (Lewis, Gerhand, & Ellis, 2001). Lewis et al. have
provided evidence for this hypothesis using mathematical model-
ing: However, research with old adults has not supported it. AoA
effects should decrease as the language user becomes older. The
difference, for example, between words learned at age 3 years
versus words learned at age 8 years should be large when the
learner reaches age 14 years (these words have been encountered
for 11 and 6 years, respectively), but the difference should be
smaller when the learner reaches age 60 years (the same words
have been encountered for 57 and 52 years, respectively). Morri-
son et al. (Morrison, Hirsh, Chappell, & Ellis, 2002) found the
standard AoA effect but also found that it did not increase with
age. Such findings provide compelling evidence against the cumu-
lative frequency hypothesis.1

The semantic locus hypothesis claims that early learned words
have a semantic advantage over late learned words because they
enter the representational network first and affect the semantic
representations of later learned words (Brysbaert, Van Wijnen-
daele, & De Deyne, 2000; Steyvers & Tenenbaum, 2005). Brys-
baert et al. (2000) found that participants generated associates
faster to early learned words than to late learned words. Similarly,
Morrison and Gibbons (2006) found AoA effects in a “living”
versus “nonliving” semantic categorization task but only for the
“living” items. Research with neural networks has found that early
learned words have more semantic connections to other words than
do late learned words (Steyvers & Tenenbaum, 2005) and that
early learned words establish a basic semantic structure that allows
later word learning to accelerate (for a discussion of the “vocab-
ulary spurt” in lexical acquisition, see Li, Zhao, & MacWhinney,

1 Recently, Zevin and Seidenberg (2004) have suggested a variant of the
cumulative frequency hypothesis in which both cumulative frequency and
frequency trajectory play an important role. Frequency trajectory, unlike
cumulative frequency, refers to whether a word is encountered more
frequently in childhood than in adulthood (e.g., potty, stroller) or vice versa
(fax, merlot). In their view, AoA is difficult to quantify because it corre-
lates highly with other types of information; hence, it may be impossible to
isolate. According to Zevin and Seidenberg, frequency trajectory may be a
more accurate measure of true AoA.

639AGE OF ACQUISITION MECHANISMS

in press). Hence, AoA effects may be due, at least in part, to
differences in semantic processing.

According to the semantic locus hypothesis, early learned words
are conceptually more enriched than late learned words (e.g., have
more semantic connections to other words) and these representa-
tions affect later learning.2 In monolingual individuals, a linguistic
form maps in a consistent and relatively straightforward manner to
its corresponding conceptual representation. In bilingual indvidu-
als, however, each concept maps to two forms, one for each
language. Thus, the semantic locus hypothesis suggests that AoA
effects should transfer to a second language.

Bilingual researchers have long argued for a unitary semantic
store with separate lexical form representations for each language
(Altarriba, 1992; Kroll & de Groot, 1997, 2005; Kroll & Tokow-
icz, 2005; Kroll, Tokowicz, & Nicol, 2001; Potter, So, von Eck-
ardt, & Feldman, 1984; Schreuder & Weltens, 1993; Sholl, San-
karanarayanan, & Kroll, 1995). Furthermore, they have argued that
connections between concepts and L2 lexical items are mediated
initially through the learned first language. As proficiency (i.e.,
language ability) improves, connections between L2 and the con-
ceptual store are strengthened. The conceptual/semantic process-
ing of L2 items should reflect the overall organization of the
conceptual system because semantic processing occurs at the con-
ceptual level.3 If AoA effects are purely a product of early items
having more semantic connections to other items than do late
items, then an L2 lexical item should inherit the L1’s AoA

Empirical studies with L2 speakers, however, have not sup-
ported this prediction. Researchers have found that the speed of L2
lexical access is determined by the age at which words are ac-
quired in the second language (L2 AoA) and not the age at which
the corresponding words are learned in the native language (L1
AoA; Hirsh, Morrison, Gaset, & Carnicer, 2003; Izura & Ellis,
2004). Thus, these effects appear to be due to the in which
words enter a particular language, irrespective of when the lan-
guage was learned (Hirsh et al., 2003). In to account for this
finding, the semantic locus hypothesis would have to assume
separate semantic stores for each language. Researchers have not
yet determined the exact mode of bilingual lexical representation
(for a review, see French & Jacquet, 2004; Kroll & Tokowicz,
2005); however, most of the evidence favors a single semantic
store. Thus, it seems reasonable to assume that AoA exerts its
effects at the lexical level rather than at the semantic level (for
further discussion along these lines with monolingual individuals,
see Belke, Brysbaert, Meyer, & Ghyselinck, 2005).

Computational accounts of age of acquisition. Some connec-
tionist models have been designed to explicitly capture mechanisms
of AoA (Ellis & Lambon Ralph, 2000; Li, Farkas, & MacWhinney,
2004; M. A. Smith, Cottrell, & Anderson, 2001). Ellis and Lambon
Ralph trained an auto-associative network on sets of words that were
introduced at different times. They showed that the network displayed
strong AoA effects, as indicated by lower recognition errors for early
than for late learned words when the words were presented to the
network in stages, that is, trained on one set of words before a second
set was introduced. Using the same model without staged learning,
M. A. Smith et al. (2001) showed that recognition errors decreased as
a function of learning ; early learned words had lower final
recognition errors than did late learned words. Li, Farkas, and Mac-
Whinney (2004) further explored AoA effects using a self-organizing
neural network relying on Hebbian learning. AoA effects appeared
such that early and late acquired words showed structural differences

in organization as a natural outcome of learning . More recently,
Lambon Ralph and Ehsan (2006) showed that their connectionist
network could capture AoA effects as a function of the consistency or
predictability in the input-to-output mapping relations: arbitrary map-
pings elicited larger AoA effects compared with less arbitrary map-
pings. In each case, AoA effects appeared to reflect increased rigidity
(reduced plasticity) of the network as a result of the learning process.
Early learned words influenced the structural organization of the
distributed mental lexicon more than late learned words, and had
better optimized representations (e.g., as captured by word density
measures in Li, Farkas, & MacWhinney, 2004).4

These connectionist models provide a general account of AoA that
is not specific to any particular domain (i.e., phonology, semantics,
etc.); as such, the account is compatible with aspects of several
hypotheses. For example, the semantic locus hypothesis also posits
that early learned words help shape the (semantic) network. Similarly,
the phonological completeness hypothesis conceptualizes early
learned words as more complete than late learned words and posits
that these words form a foundation for the less complete words
acquired later. Hence, loss of plasticity may be a property of learning
that is reflected in a number of domains.

Neuroimaging studies of age of acquisition. Relatively few
studies have investigated the neural basis of AoA effects. Fiebach,
Friederici, Müller, von Cramon, and Hernandez (2003) examined
AoA with functional magnetic resonance imaging (fMRI), a tech-
nique that allows researchers to measure the oxygenation level of
blood and thereby determine which neural areas are activated during
a task. Participants were asked to make visual and auditory lexical
decisions to words and pronounceable pseudowords during fMRI
scanning. Results in both the visual and auditory modalities revealed
increased activity for late relative to early learned words in the left
inferior prefrontal cortex (IPFC; Brodmann’s Area [BA] 45) extend-
ing to the lateral orbitofrontal cortex (BA 47/12). The precuneus was
more activated for early learned relative to late learned words (see
Figure 1). In addition, increased activity in the region of the left

2 This is not true for some proposals based on statistical learning or
neural networks. For example, Li et al. (in press) argued that semantic
representations become enriched over time as a function of learning, much
like filling holes in Swiss chess; initially, there may be more holes than
cheese (shallow representations), but the holes fill quickly as the word
context accumulates during learning (rich representations). This perspec-
tive, however, does not contradict the idea that early learned words estab-
lish the basic lexical–semantic structure.

3 One could argue, however, that semantic structure is not equivalent to
conceptual structure, with the former tied to specific properties of a given
language and the latter more language independent (for further discussion
see Lyons, 1977). Most bilingual lexical memory research does not make
this fine-grained distinction, and considers semantic and conceptual struc-
ture at the same level.

4 Zevin and Seidenberg (2002) have argued that AoA effects may be
restricted to tasks in which early learned information does not aid in
acquiring items introduced later. In the simulations discussed above, the
networks must “memorize” each pattern. However, Zevin and Seidenberg
have simulated reading acquisition and found that the practice effects can
diminish AoA effects. Hence, AoA effects may be robust for tasks such as
object naming and face recognition but may be small for skilled tasks such
as reading (for additional evidence in favor of this view, see Lambon Ralph
& Ehsan, 2006; Monaghan & Ellis, 2002b). Studies of AoA effects in
transparent orthographies, however, call into question this “mapping”
hypothesis (Raman, 2006).

640 HERNANDEZ AND LI

temporal operculum near Heschl’s gyrus was observed for early
relative to late learned words in the visual modality. Because auditory
association cortices were activated, Fiebach et al. concluded that
participants automatically coactivated auditory representations when
making lexical decisions to early learned words that were visually
presented, possibly to facilitate word recognition. The increase in
inferior frontal activity during processing of late learned words is
compatible with findings regarding the role of the left IPFC in
semantic processing. Left IPFC appears to be critical in the effortful
or strategic activation of information from the semantic knowledge
system (Fiez, 1997; Thompson-Schill, D’Esposito, Aguirre, & Farah,
1997). Hence, processing of late learned words, at least when making
lexical decisions, is likely to involve complex semantic retrieval or
selection processes instantiated by inferior frontal brain areas. An
interesting implication of this result is that semantics may play a
strong role in learning words late in life, whereas auditory processing
may play a strong role in learning words early in life. This makes
sense especially in light of our hypothesis regarding early sensorimo-
tor integration in L2 acquisition (see discussions presented later in the
section, Integration of Age of Acquisition Effects Across Domains).

Recent studies have extended Fiebach et al.’s (2003) research using
word reading (Hernandez & Fiebach, 2006) and picture-naming tasks
(Ellis, Burani, Izura, Bromiley, & Venneri, 2006). Ellis et al. (2006)
found increased activity in separate portions of the middle occipital
gyrus for early compared to late learned words, suggesting that both
sets engage visual processing to a certain extent. Of particular interest,
was the fact that late learned words elicited activity in the fusiform
gyrus and early learned words elicited activity in the most inferior
portions of the temporal lobe. Ellis et al. (2006) interpreted activity in
the temporal pole for early learned items as reflecting the highly
interconnected nature of these items. This inference is based on
evidence that damage to the temporal poles leads to semantic demen-
tia (Rogers, Lambon Ralph, Hodges, & Patterson, 2004; Thompson,
Patterson, & Hodges, 2003). The increase in activity for late learned
compared with early learned items in the fusiform gyrus reflects an

increased need for visual form processing (Devlin, Jamison, Gonner-
man, & Matthews, 2006; Price & Devlin, 2003). These results seem
consistent with the view that early learned items have more semantic
interconnections than do late learned items, whereas late learned items
require more visual form processing than do early learned items
during picture naming.

Hernandez and Fiebach (2006) asked participants to read single
words during fMRI scanning. Increased activity to late as com-
pared with early items was found in the left planum temporale
(posterior superior temporal gyrus) and in the right globus pallidus,
putamen, middle frontal gyrus (BA 9) and inferior frontal gyrus
(BA 44). The authors suggested that late learned words engage
brain areas in the left hemisphere that are involved in mapping
phonological word representations and areas in the right hemi-
sphere that aid articulatory and motor processing.

These results implicate neuroanatomical substrates that may be
associated with plasticity. In all of the studies reviewed above,
processing of late learned items involved brain areas thought to be
involved in effortful retrieval, including effortful semantic re-
trieval in lexical decision, articulatory and motor processing during
reading, and visual form processing in picture naming. By contrast,
early learned words appeared to be strongly connected to seman-
tics in picture naming and to auditory word representations in
lexical decision. Together, these results are consistent with the
notion that the neural substrate of early learned words is at a basic
level, albeit semantic or auditory, depending on the task. Late
learned words build on these representations and require additional
processing during lexical tasks.

Age of Acquisition in Second-Language Learning

The term AoA has also been used by scholars of L2 acquisition.
The meaning of the term, however, is different when researchers
use it to describe L2 learning than when they use it to describe L1
processing. In L1 processing, AoA refers to a stimulus property of
linguistic items (early vs. late learned words), whereas in L2
learning AoA usually refers to learner characteristics (early vs. late
starting age for acquiring L2). In the L2 literature, AoA5 is often
examined along with other learner characteristics, such as level of
L2 proficiency.6

Behavioral studies have long documented differences between
early and late learners of a second language. They have consis-

5 Some authors (e.g., Johnson & Newport, 1989) have used “age of
arrival” rather than AoA to indicate the age at which L2 acquisition begins.
The former term is conceptually relevant to immigrant learners whose L2
learning coincides with their arrival in the target language country, whereas
the latter is a more general term. Here we use “L2 AoA” for consistency.

6 Language proficiency can be defined as the degree of control one has
over a language. Proficiency can be defined in four domains: listening,
speaking, reading, and writing. These skills, although interrelated, are
independent in that one skill may develop separately from the others.
Cummins (1983) has argued that language proficiency has two levels: basic
interpersonal communicative skills (BICS) and cognitive and academic
language proficiency (CALP). BICS involves personal, face-to-face,
“context-embedded” communication and typically requires 2 years to
acquire, whereas CALP involves skills in understanding and using lan-
guage in academic settings (context-reduced) and requires 5 to 7 years to
acquire. Studies in the psycholinguistic and neuroimaging literature gen-
erally use some standardized test to assess proficiency; hence, proficiency
involves CALP in Cummins’s terminology.

Figure 1. Neural activity associated with early and late learned words.
Increased activity is evident for early and late learned words in monolingual
German speakers. The blue-to-green scale represents areas of increased activ-
ity for early learned words. The red-to-yellow scale represents areas of in-
creased activity for late learned words. BA � Brodmann’s area; IFG �
inferior frontal gyrus; ant. � anterior; lat. � lateral. From “Distinct brain
representations for early and late learned words,” by C. J. Fiebach, A. D.
Friederici, K. Müller, D. Y. von Cramon, & A. E. Hernandez, 2003, Neuro-
image, 42, p. 1631. Copyright, 2003 by Elsevier. Adapted with permission.

641AGE OF ACQUISITION MECHANISMS

tently found an AoA on the ultimate attainment of L2 (Flege,
Munro, & MacKay, 1995; Flege, Yeni-Komshian, & Liu, 1999;
Mackay & Flege, 2004; Munro, Flege, & MacKay, 1996).

Although critical period effects in L2 learning are still being
debated (Hakuta, Bialystok, & Wiley, 2003; Harley & Wang,
1997; Johnson & Newport, 1989; Liu, Bates, & Li, 1992; Snow &
Hoefnagel-Höhle, 1978), researchers generally agree that late
compared with early learning of L2 is associated with lower
ultimate proficiency, even though some individuals may achieve
native-like proficiency (Birdsong, 1992). Moreover, behavioral
work by Hernandez and colleagues suggests that proficiency, and
not AoA, determines naming latencies in lexical tasks when L2
acquisition occurs early in life (Hernandez, Bates, & Avila, 1996;
Hernandez & Kohnert, 1999; Hernandez & Reyes, 2002; Kohnert,
Hernandez, & Bates, 1998). This is consistent with the view that
L2 AoA affects the processing of syntax, morphology, and pho-
nology more than it affects lexical and semantic processing (John-
son & Newport, 1989; Weber-Fox & Neville, 1996).

Evidence supporting the role of AoA in behavioral studies has
been overwhelming; however, findings regarding the neural bases
of L2 AoA effects have been mixed. First, language recovery in
those with bilingual aphasia is not driven exclusively by L2 AoA
(see Fabbro, 1999, for a review). Second, recent fMRI studies have
yielded conflicting results, with some finding that AoA …