Chapter 5

Wikis > Chapter 5

Age of Acquisition


Difficulties adults face when acquiring a second language

It is widely believed that the age at which we learn a second language affects our ability to acquire the language. However, recent studies have shown that the difficulties that adults face when learning a second language is due to the process of learning itself rather than age.

Age vs. Order

AoA effects refer to the shorter reaction time and higher accuracy for early learned words as compared to late learned words (Izura, Perez, Agallou, Wright, Marin, Stadthagen-Gonzalez & Ellis, 2010)

There are two different theories that account for AoA effects. One theory is that AoA effects would reflect the order in which language is acquired (early or late) while the other theory is that AoA effects would reflect whether a language is acquired before or after a certain age, or more specifically, a critical period (also known as the sensitive period).

According to Marinova-Todd, Marshall and Snow (as cited in Izura & Ellis, 2002), critical period refers to the “time when language can be acquired much more easily than when the critical period has passed.” However, there has been no consensus on the exact age range for the critical period as it has been defined differently by different researchers. According to Pinker (1994) (as cited in Seidenberg & Zevin, 2006), the critical period is up to 5 or 6 years old while according to Lenneberg (1967) (as cited in Seidenberg & Zevin, 2006), it is up to puberty. Based on this concept of critical period, the difficulty of learning a second language would then be attributed to the age at which the second language was learnt.

However, current views of AoA effects seem to be leaning towards the view that AoA effects reflect the order in which language is acquired especially since AoA effects have been seen even after the critical period has passed (Izura & Ellis, 2002). Research has shown that words learned early may result in faster recognition and production than words learned late (Hernandez & Li, 2007). This has been exhibited in various types of tasks, such as visual lexical decision and picture naming. A visual lexical decision task involves deciding whether a given word is an actual word or a non-word in the language, while a picture naming task refers to naming the object in the picture.

The factors and theories discussed below therefore assumes that AoA effects are seen due to the order of language acquisition rather than the age at which the language is acquired.

Go to top


Paradox of Success

This next section will introduce the theory that mainly explains how the AoA effects come about.

AoA effects can be analyzed from a computational or statistical perspective, usually achieved through the use of computer simulations. For example, connectionist models can account for the mechanism of AoA effects in this manner. Neural networks are computational models of mental processing of language. They are inspired by central nervous systems, the brain in particular, and are used to estimate functions of the brain that can depend on different inputs.

When learning a pattern for the first time, a neural network (ie. network pertaining to the nervous system) is highly sensitive to the new information and will be accustomed to that particular pattern (Seidenberg & Zevin, 2006). However, if afterwards another pattern is learnt, the network becomes less sensitive to the second pattern as it has already been accustomed to the first pattern. Hence, the network declines in plasticity and becomes “entrenched”. Entrenchment hence refers to the network progressively becoming less flexible after successfully learning a task (Seidenberg & Zevin, 2006).

This approach also suggests that learning a task successfully may make it more difficult to learn other new tasks afterwards. This process has been dubbed the ‘Paradox of Success’. Generalizations made from learning the first task makes it more difficult to assimilate the second task. Hence, this model helps to explain why early learning of L1 occurs at a rapid rate but learning of L2 is generally more difficult. Entrenchment of L1 therefore interferes with the learning of L2, especially if the two languages are very different.

An example of L1 entrenchment is provided by adult Japanese speakers who fail to learn to discriminate in English the phonemes /r/ and /l/. This discrimination is difficult for them as they do not have /r/-/l/ discrimination in Japanese. It can hence be seen that entrenchment of their Japanese interferes with their learning of the English phonemes.

Although it was previously mentioned that late learned words are at a disadvantage, this is not always the case. In fact, the learning of L2 may lead to the unlearning of L1. This is an extreme effect of entrenchment known as “catastrophic interference”. However, this only occurs if an individual is entirely removed from exposure to the first language since the onset of the learning of L2. Hence, even when exposure to L1 is minimal, unlearning of L1 would not occur. In real life, exposure to L1 is usually continued. Even so, studies have shown that catastrophic interference can occur. According to Pallier, Dehaene, Poline, LeBihan, Argenti, Dupoux and Mehler (2003) (as cited in Seidenberg & Zevin, 2006), a group of adults who were born in Korea but adopted very early on by French families when they were children experienced a form of catastrophic interference as their knowledge of Korean was lost afterwards. However, Hetherington and Seidenberg (1989) (as cited in Seidenberg & Zevin, 2006) hypothesized that even after catastrophic interference has occurred, individuals should be able relearn the L1 successfully with training as the L1 is most probably not completely replaced or erased.

Research has shown that to account for AoA effects using a connectionist network, the early and late learning sets must be introduced at different points and that the training for both sets must be cumulative and interleaved (where training for the first set still continued alongside the second set after the second set had been introduced, rather than focused (where the second set replaced the training of the first set) (Ellis & Lambon Ralph, 2000). Ellis and Lambon Ralph thus conducted a study using neural networks. There were two types of simulations; one simulation involved cumulative and interleaved training of the two sets, while the other involved focused training. When the training was interleaved, the early set performed better than the late set, displaying AoA effects. However, when the training was focused, knowledge of the early set was lost and only the late set was recognized. This hence shows the “catastrophic interference” effect discussed earlier.

Using connectionist models to account for AoA effects shows that the process of learning itself, rather than age, can in principle make later learning more difficult due to a process of entrenchment. Under this account, AoA effects are hence reliant on the order of learning a particular task. Connectionist models are general accounts and hence can account for AoA effects which occur phonologically, semantically etc.

Go to top


The next section will look at factors that may affect the extent of AoA effects.

Frequency trajectory

Another factor studied was the frequency trajectories of the words that were being taught to participants. Izura et al. (2010) set up experiments to see if this difference affected the effects of AoA. In teaching English monolinguals Spanish words, they exposed the participants to early words for two consecutive sessions, with each word being trained six times each session. The ‘late’ set of words were introduced in the third session, with each word being trained 12 times each for the first two sessions, while being interleaved with the early learned words being continually trained six times each. Thus, while the words from both the early and late sets were trained an equal number of times, the early set had a “flat medium trajectory” while the late words had “high to medium trajectories”.

Fig. 1 Early items with flat frequency trajectory and late items with high to medium trajectories

Fig. 1 Early items with flat frequency trajectory and late items with high to medium trajectories

In a second experiment, early words were presented in the first two sessions, while the late words were presented in both the third and fifth session. Early words were interleaved with late words in session 4. This way, both the early and late words had equal training as there were two early-only and late-only trainings each. However, the results of these two experiments show that the difference in frequency trajectories did not contribute to the AoA effect. The second experiment served to strengthen the argument that early learned words had the advantage over late learned words.

Fig. 2 Interleaved training with two early-only sessions and two late-only sessions

Fig. 2 Interleaved training with two early-only sessions and two late-only sessions

Similarity between L1 and L2

When L1 and L2 are largely different, AoA effects would be larger. Based on the “Paradox of Success” introduced earlier, this is because generalizations would have been made from the early learned words (L1). The networks in our mind would then attempt to apply these generalizations onto the late learned words (L2). As such, if L1 and L2 differ greatly, the generalizations would be inapplicable and it would hence be more difficult to acquire the L2. For instance, according to a study done by Chen, Shu, Liu, Zhao and Li (as cited in Hernandez & Li, 2007), it was difficult for L1 speakers of Chinese to acquire the subject-verb agreement when learning English as an L2. This was attributed to the absence of subject-verb agreement in their L1 (Chinese).

However, differences between the two languages may also be advantageous. For instance, Seidenberg & Zevin (2006) suggest that the absence of click sounds in English may make it easy for an English speaker to learn Zulu clicks as these sounds are distinctively different.

AoA effects do not necessarily have a negative impact on L2. When there is a high similarity between L1 and L2, learning of L2 could be enhanced. Similarities across the two languages include phonological, morphological, syntactical and lexical similarities.

Go to top

Basis of representation

AoA effects have been said to be based on phonological representation and semantic representation.

Semantic representation

It has been proposed that AoA effects are based on semantic representations. Semantic representations refer to the way meaning is represented in the mind. Since most research has shown evidence supporting a single semantic store, which is where meaning is encoded in the mind (Hernandez & Li, 2007). This means that even though meaning can be expressed in two separate languages, there is only a single shared concept. Thus, based on the semantic representation, since the concept is first encountered when learning L1, the AoA effects of L2 learned would reflect the AoA effects of L1. For instance, if the word “fairy” is learned early in L1, the word “fairy” in L2 should reflect the AoA effect of a word learned early even though it may have been learned late in L2. However, it has been shown that this is not true. The AoA effects of L2 does not reflect the order of which words were learned early or late in L1 but instead reflects the order of words learnt in L2 itself (Izura et al., 2010).

This hence shows that although there is still a possibility that there is only a single semantic store, it can be seen that AoA effects are not based on how words are semantically represented. Instead this further reiterates that AoA effects are based on the order of learning, or more specifically whether the learning concerned is the ‘early’ learning or the ‘late’ learning.

Phonological representation

Phonological representation refers to the whole phonological shape of the word as represented in the mind. Words learned early are represented holistically in the mind and in contrast, words learned late are represented in a fragmented way and would therefore need reconstruction for an output of the whole phonological shape of the word. As a result, words learned early are pronounced faster than words learned late.

However, this manner of representing words phonologically has been contradicted. Contrary to what this representation predicts, when participants were asked to perform a segmentation task, their reaction times were quicker for early learned words as compared to late learned words (Monaghan & Ellis, 2002a). If words learned late are in fact learned in a fragmented way, they should be easier and therefore faster to segment than early learned words.

Go to top


Studies which reflect real world situations

Many of the tests and experiments conducted in the studies described here involved sessions that were concentrated over a span of a few weeks, and words trained were tightly controlled. This is rather unlikely in the real world as words are taught over a longer period of time. Thus, it would be beneficial if tests were more similar to the real world situation to more effectively study AoA effects.


In conclusion, although previous studies showed that biological age was the main reason behind the difficulties which adults face when learning a second language, more modern research has shown that this is not true. In fact, these difficulties can be attributed to the order of learning. However, the extent of AoA effects depends on various factors.

First created by Christina Amrita Arul, Nur Ayesha Bte Solehan, Nur Atiqah Binte Othman, AY2014/15 Semester 1


Sub Wikis