Though the immediate effectiveness of the keyword method is supported by many studies, Wang et al. (1992) claimed that the performance on the delayed test may be improved by the immediate test. That is, Wang et al. (1992) suggested that the
- 14 -
immediate test may serve as a potential learning trial and thereby boost learners’
performance on the delayed test. They speculated that the keyword learners’
superiority may not last over time if they were not tested immediately.
To explore whether an immediate test boosts learners’ later performance, Wang et al. (1992) conducted an experiment using a between-subjects design. The purpose of the research design was to prevent any effects of the immediate test by having half of the participants take the immediate test and the other half the delayed test. The participants were adult beginners who learned the vocabulary through either rote repetition or the keyword method. The researchers then compared the scores on the immediate and delayed tests. The results revealed that the keyword learners forgot newly learned words faster than the rote learners did. The keyword learners
significantly outperformed the rote learners on the immediate test, but among the participants who only took the delayed test, no significant difference was found. They concluded that the keyword learners forgot things faster than the rote rehearsal
learners.
Later, Wang and Thomas (1995) conducted another study to test if the faster forgetting rate also occurred when comparing the keyword method to context learning.
The participants learned vocabulary either with the keyword method or in context.
The keyword learners were given keywords and then asked to form mental images,
- 15 -
while context learners were presented flashcards with a two-sentence context and the L1 translation of the target word. Participants were tested either immediately or after a two-day delay. The keyword learners in the study outperformed the context learners on the immediate test; however, on the delayed test the keyword learners performed worse than the context learners. The results thus revealed that the effectiveness of the keyword method vanished within two days. The reason that the keyword method could not maintain its advantage over time was not explored in the study.
The faster forgetting rate of the keyword method was confirmed by Carney and Levin (1998). The participants in their study were undergraduates. A faster forgetting rate for the keyword learners over rote learners was also observed in their study. They separated the target words into two categories, some of which were tested on the immediate test and the others not. The results indicated that the tested words were better recalled during the delayed test, and it was found that the advantage of the keyword method was not seen on the delayed test when the target words were not on the immediate test list. This echoes Baddeley’s (2012) assertion—phonological coding is easy to build and the effect is rapid, but phonological coding is also readily forgotten [in the short-term memory]. That is, the immediate advantage of the keyword method was built by the phonological encoding connected to the keyword but rapidly disappeared over time.
- 16 -
Researchers and practitioners have sought to maintain the immediate advantage of the keyword method. Several ways have been used to reinforce the connection between the target word and its semantic link based on the keyword.
These techniques include, but are not limited to: encouraging learners to generate their own keywords (Shapiro & Waters, 2005) or form vivid and interactive images (Campos, González, & Amor, 2003, Experiments 1 and 2; Wang & Thomas, 1995, Experiment 1), repetitively providing stimuli of the target words (Karpicke & Smith, 2012; Wang & Thomas, 1995, Experiment 3), and attaching a relevant sound to the keyword (e.g., lamb → “baa, baaing”) in a picture related to the meaning of the target word (Carney & Levin, 1998, Experiment 3). Though many techniques have been examined, none of them led to a significant advantage in later retrieval when learners were not tested immediately. The retention issue remains unresolved.
While some researchers have used different techniques to reinforce the link between a target word and its semantic link, Wang et al. (1992) speculated that the problem may stem from forgetting the keywords provided by the researchers. They conjectured that when the keywords were provided by the researchers instead of generated by the learners, they would be forgotten easily. If learners forgot the keyword, it would be unlikely for them to rebuild the mental image or link the image to the meaning of the to-be-learned target word. However, the results revealed that the
- 17 -
participants could still recall most of the keywords during later recall (Wang et al., 1992; Wang & Thomas, 1995). Though the participants could recall the keywords linked to the target words, they failed to reconnect the keyword to the meaning of the target words.
Wang et al. (1992) thus claimed that the keywords remembered by the participants may become a hindrance during later recall. They speculated that when learners recalled the keywords to draw on their acoustic similarity to the target words,
irrelevant associations may occur. For instance, in their study, their participants were asked to connect the keyword “egg” to the mental image of a church by encoding the
keyword as a white, oval hen’s egg. The French word église (church in English) is acoustically similar to the English word egg. Hence, “egg” worked as a keyword and its white figure was connected to the church. However, participants might recall different kinds of eggs (e.g., scrambled eggs) one week later. Though Wang et al.
(1992) speculated that the keyword may become hindrance, they did not further investigate the reason causing the hindrance.
Wang et al. (1992) believed the acoustically similar keywords may introduce confusion over time and thus impair the participants’ performance. The hindrance of the keywords may be caused by the phonological (acoustic) similarity effect. Lian, Karlsen, and Winsvold (2001) discovered that when words were connected to
- 18 -
irrelevant semantic information, the phonological similarity effect was observed. In their study, they divided words into those with high and low associative values. To define the associative value, they presented a list of words one by one to the
participants and asked them to link a word to a semantic associate as fast as possible and then calculated the time to make the connection. If the reaction time of a word was shorter than the mean reaction time, the word was defined as having a high associative value. On the other hand, if the reaction time of a word was longer than the mean reaction time, the word had a low associative value.
The results revealed that when the words had a high associative value, the phonological similarity effect was observed. In other words, if a word could be linked easily to irrelevant semantic information, the phonological similarity effect would occur. This echoes Wang et al.’s (1992) speculation. They speculated that the keywords became linked to irrelevant semantic information over time (e.g., eggs linked to scrambled eggs instead of the shape of a mosque) and hindered the retrieval of the meaning of the target words. Nevertheless, Wang et al. (1992) did not explore whether the hindrance was caused by the negative phonological similarity effect.
The phonological similarity effect was first observed by Conrad and Hull (1964). They discovered that when learners are provided with a list of acoustically similar words, it is very likely that the words may cause interference and confusion
- 19 -
during subsequent recall. Later, Nimmo and Roodenrys (2005) conducted an experiment to examine if the phonological similarity effect occurred when the students were provided with a list of acoustically similar words. The result revealed that acoustically similar words could not be recalled as well as phonologically dissimilar words. In other words, greater impairment occurred as the similarity in sound increases (Nimmo & Roodenrys, 2005).
The phonological similarity effect may explain the failure of the immediate effectiveness of the keyword method to be retained. According to Baddeley (2012), there are two storage systems involved in memory: phonological and semantic. The phonological system mainly involves short-term memory (working memory) and the semantic system involves long-term memory. When the target words are connected to
acoustically similar keywords, phonological encoding is involved and the information is stored temporarily and immediately “using a process that is rapid, and requires minimal attention” (Baddeley, 2012, p. 12). This explains the immediate effectiveness
of the keyword method.
However, when learners need to recall the learned information over time, their performance might be impaired by the phonological similarity effect caused by the keywords. According to Lian et al. (2001), the phonological similarity effect might occur when “the irrelevant lexical clusters and their neighbors” in long-term memory
- 20 -
were activated during recall (Lian et al., 2001, p. 297). In other words, if the target words made the participants recall the keywords over time, this might trigger the phonological similarity effect and thus cause confusion. The confusion caused by the phonological similarity effect impaired the participants’ recall of the semantic
meanings of the words. This explains why the keyword learners failed to recall the definitions of the target words over time, even though they could still recall almost all of the target words (Carney & Levin, 1998; Wang & Thomas, 1995; Wang et al., 1992). That is to say, the keyword learners successfully retrieved the phonologically similar keywords but failed to retrieve the definitions due to the negative phonological similarity effect.
Some researchers have suggested that by encouraging semantic encoding prior to phonological encoding, it may be possible to make the learned information more
robust and hence facilitate long-term retention (Campoy & Baddeley, 2008; Savill, Ellis, & Jefferies, 2016). In Campoy and Baddeley’s (2008) study, they discovered
that the phonological similarity effect may be alleviated when the phonologically similar words were encoded semantically. When the participants connected the
phonologically similar words with word meanings, the phonological similarity effect was inhibited and the participants’ performance was improved. They stated that
“semantic encoding appeared to protect participants from the effects of phonological
- 21 -
similarity” and thus made the learned information more robust (Campoy & Baddeley,
2008, p. 334).
In other words, the phonological similarity effect may be alleviated by changing the implementation procedure of the keyword method. Encouraging the semantic encoding prior to the phonological encoding might inhibit the phonological similarity effect caused by the keywords, hence facilitating later retrieval. Moreover, Savill et al.
(2016) discovered that newly acquired words were more robust when semantic encoding was built prior to phonological encoding. This implies that if semantic encoding was first encouraged during the learning session, learners’ later performance might be improved. They concluded the semantic information stored in long-term memory may promote the reintegration of missing phonological elements (kept in short-term memory). Hence, encouraging semantic encoding prior to phonological encoding might facilitate the retention of the newly acquired words.
In the aforementioned studies, which aimed to retain the immediate effective of the keyword method (Carney & Levin, 1998; Wang & Thomas, 1995; Wang et al., 1992), the participants were given word lists that show target words and keywords simultaneously and were then instructed to directly link the target word to the keyword. The form or semantic meaning of the target word was provided next to allow learners to build mental images linking the keyword to the meaning of the target
- 22 -
word. Throughout the above procedure, a negative phonological similarity effect may occur before semantic encoding is built. When this phonological similarity effect occurs, semantic encoding may be harder because the phonological similarity effect would lead to inaccurate memory traces, which may lead in turn to later recall problems (Baddeley, 2012).
Hence, to avoid the negative phonological similarity effect, semantic encoding should be encouraged prior to phonological encoding. In other words, the L1
translation or the meaning-related image should be provided before the keyword association. When students are familiar with the form or semantic meaning, the keyword may cause less interference because the semantic encoding was encouraged prior to the provision of the keyword. In this way, the target word can be recalled with the dual memory traces without the hindrance of the phonological similarity effect.
Though previous studies verified that encouraging the semantic encoding prior to phonological encoding led to better performance (Campoy & Baddeley, 2008;
Savill et al., 2016), none of the studies connected the issue of the encoding sequence to the keyword method. This study aims to examine if the failure of later retrieval for keyword learners was caused by the phonological similarity effect. If so, preventing such an effect might be a possible solution.
- 23 -