1
The Savings Paradigm Applied on Language Loss in Monolingual Elderly
Does Residual Lexical Knowledge Exist?Charlotte Osté*
Abstract
This article reports on a study that applied the savings paradigm to vocabulary loss in monolingual older adults who did not have any diagnosed brain damage. The idea behind this savings paradigm is that language, once learned, is never completely forgotten but residues of knowledge remain. This study seeks to find out whether words that are no longer used are truly forgotten, or whether residues of this knowledge still exist. Therefore, an experiment is conducted in which both older and younger adults learned 30 words of two different conditions. First, they learned Dutch words that existed in the past and likely known at some point in time by the older adults only, since these words do not occur in everyday language anymore. Second, they acquired pseudo words which are previously unknown to all participants. The included items were pretested before learning to make sure the existing words were not known. Two posttests were conducted after the learning session: the first posttest took place immediately after the learning task, the second posttest was conducted approximately 20 minutes after learning. Based on the results, it can be concluded that older adults experience a relearning advantage for words that may be previously known over previously unknown words. The results did not show a significant interaction between age group and condition at the first posttest. However, the interaction between age group and condition was significant at the second posttest, i.e. the difference between the older and younger group was larger for pseudo words than for existing words. Moreover, the outcomes showed that existing words were more likely to be retained by the older age group at the second posttest than pseudo words. It is concluded that this study provides evidence that the savings paradigm is also applicable to language loss in monolingual and healthy elderly. This study also supports the idea that lexical knowledge of words is never completely forgotten. In other words, residues of this knowledge remain.
Keywords: Language loss, attrition, monolingual, L1, aging, elderly, savings paradigm, lexical knowledge, vocabulary
* Charlotte Osté (10361936), Research Master Linguistics, University of Amsterdam. This study is supervised by dr.
S.J. Andringa (UvA) and prof. dr. F. Kuiken (UvA). Prof. dr. J.H. Hulstijn (UvA) additionally supervised this master thesis.
2
1 Introduction
Many language users have experienced language loss. Language loss may occur due to various reasons, which can be generally considered as pathological or non-pathological, and so it occurs in different contexts. Pathological loss is caused by, for example, dementia that damages certain brain areas that are somehow related to language (Martin & Fedio 1983; Mesulam 2001; Mesulam 2003; Scherder 2001; Sloot & Jonkers 2011). De Bot & Weltens (1985) discussed different contexts in which language loss can appear. Depending on the language that is lost (first or second language) and the language environment, the authors distinguished four types of loss. Table 1 gives an overview of the different contexts of language loss.
Table 1. Different contexts of language loss depending on the language environment
Language loss L an gu age en vironme n t L1 L2
L1 e.g. language death or language loss caused by normal aging
L2 attrition
L2 L1 attrition e.g. language reversion
Table 1 shows that there are different contexts in which loss may occur. An example of the loss of a first language (L1) in a L1 environment is language death (De Bot & Weltens 1985). Language death is the result of the extinction of a language (Sasse 1992). A more individual process of L1 loss in a L1 environment is language loss caused by normal aging. This latter example will be discussed in detail later, because this specific type of loss is the main focus of this study. A L1 may also be lost in a second language (L2) environment, i.e. L1 attrition. In short, L1 attrition is the loss of some linguistic features of the L1 due to decreasing language use (De Bot 2004; Schmid 2011; Schmid 2013). It is generally linked to the acquisition of a L2 since it most commonly occurs in situations of migration. The declining use of someone’s L1 often co-occurs with the acquisition of the L2 (De Bot 2004; Schmid 2011; Schmid 2013). The general idea of L1 attrition is that as the L2 is used increasingly and becomes more dominant in the speaker’s mind, the L1 attrites (Schmid & Köpcke 2007). The L2 can be lost as well, both in a L1 and L2 environment. If someone
3 loses the L2 in a L1 environment, this is called ‘L2 attrition’. L2 attrition may occur as a result of insufficient use of the L2 (De Bot & Weltens 1991). It is possible to lose a L2, for example, when a speaker remigrates after having lived in a country where the L2 was dominant (De Bot & Weltens 1985), or when a L2 is not used sufficiently after it is learned in a formal classroom setting (Gardner et al. 1985). Finally, a L2 can be lost in a L2 environment. De Bot & Weltens (1985) discussed a well-known example about elderly who migrated years ago to the L2 environment and lost their L2 when they became demented. The elderly fell back on their L1, even while they are still in a place where their L2 dominates. In this type of language loss, L1 reversion goes along with L2 attrition.
Because of the many types of language loss and the contexts in which they occur, there are different ideas about what language loss exactly is. A number of studies suggested that a language can be completely forgotten, for example when a child is adopted abroad by a family that speaks another language (Isurin 2000; Pallier et al. 2003; Ventureyra et al. 2004). The early-acquired language can be forgotten if the speakers are not exposed to the L1 for a prolonged period of time. However, other studies on adopted children and the loss of their early-learned language showed contradictory results. For example, Bowers et al. (2009) reported leftover traces of exposure to the L1 in adults who were adopted in their early childhood. Even when the adopted speakers could not speak or recognize words of their early-acquired language, leftover traces were observed as an improved ability to relearn the phonology of the first language in adulthood.
The idea that language cannot be completely forgotten although it seems to be lost was also supported by studies on other contexts of language loss, such as L2 attrition as a result of cessation of using the L2 (De Bot et al. 2004; De Bot & Stoessel 2000; Hansen 2001, Hansen et al. 2002; Van der Hoeven & De Bot 2012). For example, De Bot et al. (2004) and Van der Hoeven (2012) focused on the question to what extent knowledge of lexical items that seemed to be lost was still available after a certain amount of time. The authors found evidence for the existence of residual lexical knowledge of words that seemed to be lost and, therefore, these studies supported the idea that language is never completely forgotten.
While language loss is typically associated with cessation of using a language or brain damage, it also occurs in a monolingual and non-pathological context, i.e. language loss in monolingual elderly who are mentally healthy. Many older adults without any brain damage claim to forget, for example, words. In general, elderly may have word finding problems and experience
4 more occurrences of the tip of the tongue (TOT) phenomenon than younger adults (Brown 1991; Burke 2002; Scherder 2001). According to Burke (2002), elderly indicate both problems in word finding and TOTs as their most serious cognitive problems and link these language problems to aging since they did not experience these problems in similar amounts when they were younger. There are two reasons for the fact that older adults forget more words than younger speakers. First, aging weakens certain connections that should be made for producing words and, therefore, elderly procedure more TOTs than younger people (Burke 2002). Second, words are more likely to be lost when they occur very little in everyday language. In other words, frequently and recently used words are more accessible for a speaker than infrequently used items (Ecke 2004). Since the vocabulary of a language changes over time, some words are not used in everyday language anymore. It is more likely that words that were known and used at some point in time by the older adults are lost more often because of their infrequent use.
This article reports on a study on language loss in older monolingual adults who are mentally healthy. Therefore, this study focused on loss of the L1 in a L1 environment (see Table 1). It seeks to find out whether words that are no longer used are truly forgotten, or whether residues of this knowledge still exist. Previous studies on the existence of residual lexical knowledge drew attention to a bilingual framework. Those studies investigated the existence of residual knowledge for second languages by applying the savings paradigm (see De Bot et al. 2004; De Bot & Stoessel 2000; Hansen 2001, Hansen et al. 2002; Van der Hoeven & De Bot 2012). In short, the savings paradigm corresponds to the idea that once words are learned, they cannot be completely forgotten because there will always be residues of lexical knowledge (De Bot et al. 2004; Schmid & Mehotcheva 2012). To our knowledge, the savings paradigm has not been studied in the context of a monolingual and non-pathological language use until now. By doing so, this study provides a new perspective on language loss and the savings paradigm. Therefore, this study complements existing research on language loss and it provides insight in how language is stored. Moreover, this article helps to define language loss more precisely. As discussed before, it is still unclear what language loss exactly is. By investigating the possible existence of residual lexical knowledge, this study may provide more clarity about the loss of words.
This article is organized as follows: section 2 provides the theoretical background which discusses two opposing interpretations of language loss. Subsequently, the savings paradigm and different studies that applied this paradigm to language loss will be discussed. Section 3 discusses
5 the aim of the current study and presents the hypotheses. The method of the study is described in section 4. Results will be presented in section 5. Then, section 6 discusses the main findings of this study and section 7 concludes this article.
2 Background
This section provides the theoretical background of this study. Subsection 2.1 discusses the definition of language loss and describes two opposing interpretations of language loss and to what extent language can be completely forgotten. Subsection 2.2 explains the savings paradigm and discusses previous studies that applied this paradigm to language loss.
2.1 Interpretations of language loss
Language loss refers to a process in which individual speakers lose different linguistic features or items to various degrees. Language loss can relate to both the production and comprehension of language and includes different linguistics levels, such as vocabulary, phonology and syntax (see i.a. De Bot et al. 2004; Burke 2002; El Aissati 2011; Ellis et al. 1983; Van der Hoeven & De Bot 2012; Mesulam 2001; Mesulam 2003; Parr et al. 1997; Prins & Bastiaanse 1999; Scherder 2001). As described in the introduction, language loss may occur in various contexts. Furthermore, the course of the loss can be very different. Some types of language loss, such as L1 attrition, are gradual processes in which a language is lost step-by-step (De Bot 2004; Schmid 2011; Schmid 2013; Schmid & Köpcke 2007), while other types cause abrupt language loss, for example as result of a stroke (Prins & Bastiaanse 1999). Because of all these variations in language loss, the phenomenon of loss could be interpreted in different ways. This subsection discusses two opposing views on language loss.
In the first interpretation, it is assumed that a language can be completely forgotten if contact with that language is severed. This interpretation was suggested in different studies on adults who were adopted during early childhood by a family that speaks another language (Isurin 2000; Pallier et al. 2003; Ventureyra et al. 2004). For example, Pallier et al. (2003) included eight adults who were born in Korea and who were adopted by French families when they were between 3 to 8 years old. The participants performed behavioural tests on language identification and word
6 recognition. In the language identification task, 60 sentences from different languages (Korean, French, Polish, Swedish and Wolof) were presented to the participants and the participants were asked to indicate to what extent they thought presented sentences were Korean or not. During the word recognition test, two auditory Korean words were presented to the participants, who were asked to correct translation of a presented French word. Furthermore, Pallier et al. (2003) performed brain imaging using event-related fMRI while the participants listened to 128 sentences in four different languages: French, Korean, Japanese and Polish, of which the latter two were unknown for all participants. The adoptees did not distinguish the Korean sentences from sentences coming from other unknown languages. Compared to a control group of eight native monolingual French speaker, the adopted adults could not identity Korean words in the word recognition task better. Moreover, the fMRI data revealed no differences in brain activation when the adopted participants listened to Korean relate to an unknown language. The brain areas that responded more to the known language, French, than the other languages were similar in the adopted participants and in the control group. Ventureyra et al. (2004) also reported a study on native Koreans who were adopted by French-speaking families and stopped using their L1 for many years. The results of this study showed that the adopted participants did not perceive the differences between Korean phonemes better than the control group of native speakers of French who were never exposed to Korean. The findings suggested that a language can be completely forgotten.
The second interpretation of language loss assumes that a language is never completely forgotten. Bowers et al. (2009) conducted a study on adopted children and the loss of their early-learned language as well, and they showed contrasting results than the studies that were previously discussed. Bowers et al. (2009) reported leftover traces of exposure to the first language in adults who were adopted in their early childhood. Their study included four native speakers of English who acquired Hindi in their childhood and three native speakers of English who learned Zulu in their childhood. None of these speakers had any contact to Hindi or Zulu during adulthood until the moment of testing. Moreover, they did not have remaining knowledge of the language at the time of testing. This study included a control group of monolingual native English speakers who were never exposed to Hindi or Zulu. The participants performed a vocabulary test in which they listened to 10 everyday words from their background language. These 10 words were characterized by their specific phoneme contrasts that are difficult to distinguish for monolingual native speakers
7 of English. After the participants listened to the stimuli, for each stimulus 10 possible English translations were presented. The participants were asked to point to the translation that matched the stimulus. Bowers et al. (2009) found that the test groups were still sensitive to the phoneme contrast, although they showed no preserved knowledge of their childhood language. Their results provided evidence of preserved implicit knowledge of a forgotten childhood language. Also other studies, among which Hyltenstam et al. (2009) and Oh et al. (2010), showed that the L1 was not totally forgotten. Hyltenstam et al. (2009) and Oh et al. (2010) found that it is easier to learn specific sounds of a language for adults who were adopted during childhood from the region where this language is spoken than for adults who are not adopted from this region. It seemed that adopted participants still had some knowledge about the phonetic structure of their L1, although they were not aware of this. This knowledge made it easier for them to (re)learn the sounds. Apart from phonetics, the adopted participants did not learn other linguistic aspects, such as vocabulary or grammar, easier than the non-adopted ones.
The contrasting interpretations were not only discussed in studies on the forgetting of early-acquired languages of adoptees, but also in other studies on language loss, such as studies on L2 attrition as a result of cessation of using the L2 (De Bot et al. 2004; De Bot & Stoessel 2000; Hansen et al. 2002; Van der Hoeven & De Bot 2012). All of the mentioned studies supported the idea that language cannot be completely forgotten. The researchers suggested that residues of knowledge always remain. This interpretation corresponds to the so-called savings paradigm (Nelson 1978), that will be discussed in the next section.
2.2 Savings paradigm
The savings paradigm is a theory based on the idea that information, once acquired, is never completely forgotten (De Bot et al. 2004; Schmid & Mehotcheva 2012). In other words, there will always be some knowledge left according to the savings paradigm. Although the savings paradigm is applied to language loss of natural languages nowadays, this was not always the case. Before the 21st century, the savings paradigm was often applied to human memory in general. Since current studies on the savings paradigm applied to language loss usually rely on two important experiments that were conducted during the previous centuries, this section discusses these experiments before the previous studies on the savings paradigm with respect to language loss.
8 The first study that will be discussed is the famous study of Ebbinghaus from 1885 in which Ebbinghaus found a relearning effect, i.e. the advantage of relearning previously known information over learning previously unknown information. Ebbinghaus was a psychologist who generally studied human memory. In Über das Gedächtnis (1885), the researcher conducted an experiment on learning and forgetting. It was Ebbinghaus himself who participated in this experiment. He studied 169 lists of 13 meaningless syllables until he could reproduce all of them correctly. After periods varying between 20 minutes and a month, he relearned a list he had learned previously and additionally learned a similar, new list of nonsense syllables (De Bot et al. 2004; Ebbinghaus 1964; Van der Hoeven & De Bot 2012). It turned out that Ebbinghaus performed better on the lists that were relearned than the lists that were newly learned. In other words, he experienced an advantage of relearning the old lists over learning the new lists. This relearning effect is also visualized in Figure 1.
Figure 1. Relearning effect (adopted from De Bot et al. (2004: 375)).
Figure 1 shows that the retention of acquired information can be distinguished into two different levels of memory (see the recall threshold and the recognition threshold). While one may not recall information, it is possible that one recognizes that same information. The difference between this recall retention and recognize retention can be explained on the basis of the different levels of activations that are needed to reach the recall threshold or the recognition threshold. Concerning
9 recall, the ability to produce a word, a higher level of activation is required than for recognition. If the activation level decreases further than is needed for recognition, the information cannot be retrieved anymore. Figure 1 shows that relearning makes information retrievable again for at least recognition, because the previously known information is reactivated (De Bot et al. 2004). Newly acquired information that was previously unknown, cannot reach that level after a short period of learning.
Additionally, Figure 1 shows another observation by Ebbinghaus. Ebbinghaus mentioned that his forgetting took place rather quickly in the beginning, but his rate of forgetting decreased as a function of time. This observation is well known as “the forgetting curve” (De Bot et al. 2004). Figure 2 visualizes this forgetting curve. Based on this curve, it is assumed that, once information is stored in the long-term memory, it is never completely forgotten (Van der Hoeven & De Bot 2012).
Figure 2. The forgetting curve.
Another important study in the field of the savings paradigm is conducted by Nelson (1978). Nelson (1971) was the first who referred to the relearning effect as the savings effect. Nelson (1978) showed that information will be retained to some extent even if that information is not recognized. The author conducted an experiment to investigate if participants had residual
10 knowledge of the information that they seemed to be forgotten. In this experiment, 24 participants learned a list of twenty number-words pairs, such as ‘48-PARTY’. Four weeks later, the participants returned for a recall retention test, a recognition retention test and a relearning test. Nelson (1978) assumed that the participants lost certain items when these items could not be recalled during the first test. During the recognition retention test, the various number-words pairs were presented to the participants at the same time. Only one pair was included in the list that the participants had acquired, the other pairs were distractors. The participants were asked to select the number-word pair that they had acquired. The items of the next test, the relearning test, were divided into three pools: (i) incorrectly recalled and incorrectly recognized; (ii) incorrectly recalled, but correctly recognized; (iii) correctly recalled. For each pool, half of the items remained the same as during the learning exercise (old items), whereas the other half were changed by items that were not learned before (new items). This test compared the relearning of an originally learned number-word number pair with that of an original number paired with a different response word (for example, ‘48-HORSE’ instead of ‘48-PARTY’). The results of this latter test showed that there is a relearning advantage of old items over new items. Nelson (1971) referred to this relearning advantage as “savings”. Another important finding of Nelson (1978) was that there was a relearning effect even for those items that were not recognized on the recognition retention test. He concluded that acquired information retains even if that information seems to be lost because it cannot be recalled or recognized.
If the savings paradigm is applied to language, the savings paradigm relates to the assumption that language is asymptotic: even after a long period of time, there will be residual knowledge of the items that seem to be lost (De Bot et al. 2004). Moreover, the savings paradigm assumes that this residual knowledge would make it easier to relearn lost items than to learn items that are unknown before (De Bot et al. 2004; Van der Hoeven & De Bot 2012; Schmid & Mehotcheva 2012). Only decades after Nelson’s (1978) research, the savings paradigm was applied to research on language loss. De Bot & Stoessel (2000) performed the first study on the savings effect in vocabulary relearning of languages acquired under natural exposure. They tested to what extent two adult native German participants had knowledge of Dutch, a language that they acquired as a L2 during childhood but that they had not used for many years. One of the experiments they conducted was a relearning task in which De Bot & Stoessel (2000) compared the scores of the German participants to the scores of control participants who had no previous
11 knowledge of Dutch. The task included both Dutch words of which was assumed that the German participants knew them in the past because of the high-frequency of these words and Dutch low-frequency words of which it was assumed that the participants never knew them before. De Bot & Stoessel (2000) reported that the German participants scored better on this relearning task than the control group. The German participants had a relearning advantage of the high-frequent words over the low-frequent words. De Bot & Stoessel (2000) concluded that residual lexical knowledge of the Dutch words that seemed to be forgotten but were known in the past was still in memory.
Series of studies on the savings paradigm applied to language loss followed (i.a. De Bot et al. 2004; Hansen 2001: Hansen et al. 2002; Van der Hoeven & De Bot 2012). All of the studies reported results in favour of the savings paradigm. The studies found, similar to De Bot & Stoessel (2000), an advantage of relearning over learning items. For example, De Bot et al. (2004) investigated the existence and accessibility of residual lexical knowledge in foreign languages. Their study included learners of German at universities in the U.S. and learners of French at Dutch universities. The groups participated in different experiments, but the two experiments were quite similar. Both groups were asked to relearn respectively German and French words that were probably known at some point in time but seemed to be lost (old words), and learn words that were assumed to be unknown (new words). The methods to select the old and new items differed for the languages. The old words for testing the participants learning German were selected from vocabulary lists which had been distributed to the students over the course of the semester and therefore it was assumed that the students knew the items. The new words were based on the consideration of a German teacher. This teacher confirmed that the new words would not be known to his or her students. For the experiment in which the learners of French participated, the old words were selected based on standardized vocabulary lists for students French in higher secondary education. The new words were based on their frequency: only low-frequency words were selected. For both groups, the old and new words were pretested. The participants were asked to translate the words in the foreign language to their L1. Based on the outcomes of this pretests, 15 old words and 15 new words for the learning task were chosen. All participants were tested after the learning task. The results showed that the participants were better at relearning. All participants showed a relearning advantage of old words over new words. Only the learners of French were retested four months later. During this second test moment, the learners still performed better on the old words than the new words. De Bot et al. (2004) concluded that the
12 participants were better at both relearning and retaining words they lost compared to the newly acquired words. The findings support the idea that residual lexical knowledge exists.
Another study on the savings paradigm and language loss that is worth mentioning was conducted by Van der Hoeven and De Bot (2012). Their study focused on three different age groups, including older adults. Since the current study focuses on language loss in older adults too, the study of Van der Hoeven and De Bot (2012) will be discussed. The youngest group in their study consisted of participants from 20 to 30 years old. The middle-aged group included participants between 45 and 55 years old. Participants from 70 to 85 years old belonged to the last group. All participants were native speakers of Dutch and learned French for 4 to 6 years in a classroom setting during high school. Van der Hoeven and De Bot (2012) investigated if they could find evidence for an age-related advantage in relearning old over learning new lexical information. The authors tried to find out if and to what extent residual lexical knowledge was available after a prolonged period of disuse, particularly for the oldest participants. Therefore, Van der Hoeven and De Bot (2012) designed an experiment that consisted of two test sessions. The first session was a pretest in which the vocabulary size of the participants was measured. Furthermore, this session consisted of a learning task and a test of the words that had been memorized after learning. Two weeks later, the participants were retested to measure how many words that had been learned or relearned were still memorized. The old words were selected from the same standardized vocabulary list De Bot et al. (2004) used in their study. It is assumed that the words from this list are known to people who learned French during high school. The new words were French pseudo words. Each participant was asked to learn 20 old words and 20 new words. Similar to the other studies that were mentioned in this section, Van der Hoeven and De Bot (2012) found that all age groups performed better on relearning old words than learning new words. The savings effect (that was calculated by the scores of the old minus the new words) was bigger for the oldest group and smallest for the youngest group, because the youngest group learned more newly acquired words compared to the other groups. Like the aforementioned studies, this study found evidence for residual lexical knowledge.
13
3 Present study
Previous studies on the savings paradigm applied to language loss have all been conducted in a L2 context and therefore, evidence for the savings paradigm until now comes from that L2 context only. The savings paradigm was, to our knowledge, not investigated in a L1 context. The present study focusses on language loss in monolingual older adults who were mentally healthy, which means that they did not have any brain damage as far as known at the moment of testing. Similar to the loss of a L2 as a result of cessation using that language, vocabulary loss in monolingual elderly may be due to disuse of certain words. Therefore, it could be suggested that the savings paradigm also applies to language loss in monolingual elderly. However, vocabulary loss in older adults could be driven by other factors, such as aging, as well. Therefore, it could differ from L2 attrition. Although the older adults who were included were mentally healthy, language loss in elderly may be considered as a phenomenon in between of non-pathological and pathological loss (De Bot & Weltens 1991). Therefore, it could be that the savings paradigm is not applicable to vocabulary loss in monolingual elderly.
The aim of this study is to investigate whether residual lexical knowledge exists in monolingual elderly who lost words of their native language. Therefore, it seeks to find out whether words that are no longer used are truly forgotten, or whether residues of this knowledge still exist. The study focusses on the following research question: To what extent does residual lexical knowledge exist in language loss caused by normal aging?
In the present study, a self-designed relearning/learning experiment is conducted in which both older participants from the age of 65 and older and younger participants between 18 to 30 years learned 30 words of two different categories. Half of the included Dutch words occurred in the past and were likely known to the older participant but were unknown to the younger participants because these words do not occur in everyday language anymore (henceforth: existing items). The other half were Dutch pseudo words and were previously unknown to all participants. In order to identify which words might be assumed lost, a preliminary experiment is conducted that will be discussed in detail subsection 4.2.1. Before the learning task, the 30 words were pretested to check which items were still actively known to the participants. After the learning task, two posttests were conducted. The first posttest took place immediately after the learning task and the other posttest was conducted approximately 20 minutes later. During these 20 minutes,
14 the participants performed a non-verbal intelligence test which distracts the participants from the words that they learned.
Bases on previous studies that investigated the savings effect, this study hypothesized that there is a relearning advantage of the existing items over the pseudo items for the older adults, because previously acquired lexical knowledge of the existing items was reactivated during the learning task whereas prior knowledge could not be reactivated for the pseudo items (De Bot et al. 2004; De Bot & Stoessel 2000; Van der Hoeven & De Bot 2012). We predicted that the older participants score would higher on the existing than the pseudo items at the first posttest, which was conducted immediately after the learning task, whereas we did not expect to find such a difference for the younger participants. All younger participants, after all, did not know the existing and pseudo items before the learning task. In other words, we predicted that the difference between the success rates of the existing words and pseudo words would be smaller for younger participants than for the elderly.
Another hypothesis was that the older adults would be more likely to have lost more pseudo words than existing words at the second posttest, while younger adults may lose both types equally. We expected that the difference between relearning and learning may not only would have an effect on what the participants memorize immediately after learning, but also on what they remember after a certain amount of time. The reasoning behind this second hypothesis is that relearning reactivates the words that were known at some point in time to a higher level than previously unknown words can reach (De Bot et al. 2004). If the existing words are reactivated for the older participants, it is expected that they remember these items longer than they can do for the pseudo words. We predicted that the difference between the success rates of relearned words and newly learned words would be smaller for the younger participants than for the elderly when they would be tested after a period of time.
4 Methodology
The present study consisted of two phases. In the first phase, a lexical decision task was conducted to identify potentially lost words that could be selected for the experiment of the next phase. In this second phase, both older and younger adults participated in a self-designed relearning/learning experiment. This experiment concerned a pretest, a learning task and two posttests. The
15 participants who were engaged in the relearning/learning experiment also performed a standardized intelligence test.
Subsection 4.1 discusses the participants of this study. Thereafter, the experimental design will be explained in subsection 4.2. After explaining the research design, the procedure will be explained in subsection 4.3. Finally, subsection 4.4 discusses the data analysis.
4.1 Participants
This study included a total number of 99 participants. In both phases, younger and older adults were involved. Participants of the first phase did not participated in the second phase. The characteristics of the participants who performed the lexical decision task of phase 1 are shown in Table 2 and information about the participants of phase 2 is reported in Table 3.
Table 2. Characteristics of the participants of phase 1
young (n=27) old (n=23) age, y 23.3 ± 2.3 77.7 ± 8.7
age range 18 - 27 65 - 92
male sex, n (%) 10 (44.4) 12 (52.2)
Table 3. Characteristics of the participants of phase 2
young (n=20) old (n=29)
age, y 23.7 ± 3.4 72.6 ± 6.6
age range 18 - 30 65 - 95
male sex, n (%) 6 (30.0) 18 (62.1)
All participants met the following two criteria. First, none of the participants had another native language than or in addition to Dutch. Second, the participants did not have any cognitive deficit, language deficit, or brain damage as far as known. Furthermore, none of the participants who participated in the second phase had a background in linguistics.
The younger participants and some of the older participants were recruited via social contacts of the experimenter. In addition, older participants were found through a note on public
16 blackboard in Amsterdam and Goes, and through the snowballing method, i.e. a participant brings in other participants. A few older adults who participated in phase 1 were recruited via a meeting place in Velserbroek that aims to bring older adults together to maintain social contacts. All elder participants were leading active social lives when they were tested. The vast majority of the older adults who participated in the first phase and all participants of the second phase were still living independently at the moment of testing. Ethical approval was obtained.
4.2 Experimental design
The research methodology consisted of two phases in which the participants of the first phase performed a lexical decision task whereas the participants belonging to the second phases participated in both the relearning/learning experiment and the IQ test. Before the participants performed the tests belonging to their specific phase, all participants filled in a questionnaire that elicited information about the background of the participants. The survey included questions about, for instance, place of birth, acquired languages and reading behaviour. Participants from the age between 18 and 30 years filled in a slightly different survey than the older participants since the younger participants did not answer questions about their reading behaviour in the past. The questionnaires served as screening for potential participants and helped to interpret the data derived from both phases. For example, the questionnaire provided the insight that some words in the lexical decision task were only familiar to participants from a specific region or with a specific dialect. Since we wanted to select words that were likely known to older participants for the relearning/learning experiment, the information from the questionnaire helped us.
After phase 1 was completed, the questionnaire was changed. A question about linguistic puzzles, such as crossword puzzles, was added. Almost at the end of the first phase, one older participant explained that he knew many old words because he recognized them from the puzzles he frequently solved. It turned out that solving puzzles could explain why some participants knew words that do not occur anymore. We added the question about linguistic puzzles to the questionnaire for the second phase, to ensure that we could control for this in case it was necessary. The questionnaires are included in appendix 1.
17 4.2.1 Lexical decision task
The first phase of this study consisted of a lexical decision task. This lexical decision task was used to select test items for the relearning/learning experiment. The task helped to identify potentially lost words by including older adults. Younger adults were involved too, because from their responses we could trace whether the words were actually unknown to younger people and unused in everyday language. Furthermore, the responses of the younger age group provided information about the transparency of the words with respect to their meaning. We did not want to include words in the relearning/learning task of which the meaning could be easily derived from its form.
Participants who performed the lexical decision task indicated whether they thought that the presented words were existing Dutch words or not. If they thought a presented word existed in Dutch, they were asked to supply the meaning of this word as precisely as possible. The participants had been informed in advance that they could indicate an item as a Dutch word without knowing the meaning.
The lexical decision task contained 100 words in total of which 77 test items and 33 fillers. The test items were words that occurred in the past, but are no longer used in everyday language according to Het nieuwe verdwijnwoordenboek (2015). Het nieuwe verdwijnwoordenboek is a dictionary containing Dutch words that were used in the past, but that have disappeared from the spoken and written language mostly during the twentieth century and none of the words should occur in everyday language anymore. The test items of the lexical decision task were selected from this dictionary.
Since it is important that the older participants knew the words in the past, these words had to appear in the language that the older participants could have heard or read before. Therefore, we introduced the criterium that the words had to be used frequently between 1965 and 1990. Words that did not appear, or only appeared once or a number of times, were excluded. We checked the frequencies with respect to the period of 1965-1990 by using the online tools Delpher and the ngram viewer. Delpher is developed by the National Library of the Netherlands in association with four Dutch universities. Delpher contains a large number of digitalized texts, including newspaper articles. We searched the frequencies of the words in both national and local newspapers who were published during 1965–1990. The ngram viewer is developed by the Digital Library of Dutch Literature (DBNL). The DBNL is a digital collection of primary and secondary information on
18 Dutch language and literature. Furthermore, the frequencies of the words were checked in the database of the ngram viewer. The ngram viewer presented the result of a search in a graph with on the x-axis the dates of the texts containing the searched word and on the y-axis the number of sites for that word in texts from the relevant year. Since Delpher and the ngram viewer did not always provide approximately the same frequency, the counted frequencies of both databases were used.
Furthermore, SUBTLEX-NL (2010) was used to check the frequencies of the words nowadays. SUBTLEX-NL is a database of Dutch word frequencies based on 44 million words from film and television subtitles. Only the words that do not occur in this database are selected for the lexical decision task. By checking the current frequencies, we controlled for the possibility that the items were not disappeared at all but are still used in everyday language.
After searching words in Het nieuwe verdwijnwoordenboek (2015) and checking their frequencies during 1965 and 1990 and the current frequencies, several words seemed suitable for the lexical decision task. However, two additional steps were necessary for the selection of the 77 test items. First, the words were divided into six categories based on their frequencies. The first category contained items that occur 50-100 times in the databases. The second category included items that occur 100-150 times. Items that occurred 150-200 times belonged to the third category. The fourth category contained words with a frequency from 200-250. The fifth category existed of items that showed up 250-300 times. Finally, items that occurred more than 300 times according to Delpher and the ngram viewer belonged to the sixth category. Second, from every category, a couple of words were selected. We deliberately chose not only to include words that were very frequent but also less frequent words in the lexical decision task to avoid the possibility that the older participants knew the meaning of all selected items. The form or meaning of the words and the context in which the words were used, played a role in the selection as well. For example, words that only occurred in a specific jargon or words whose meaning could be directly deduced from the form were not included in the lexical decision task.
The lexical decision task contained 33 fillers. These fillers consisted of 16 Dutch pseudo words and 17 commonly used words. The fillers checked whether the participants were concentrated and to what extent they performed the test seriously. It was expected that none of the participants claimed to know the meaning of the pseudo words since the pseudo words do not exist in Dutch at all. Furthermore, it was expected that all participants knew the meanings of the
19 common words. The results of a participant would be excluded from this study if that participant did not know the correct meaning of more than two common words or claimed to know the meaning of more than three pseudo words. Regarding the pseudo words, less strict criteria were used because the chance that a participant guessed a pseudo word was more likely than the chance that a participant did not know a commonly used word. However, none of the results were excluded because all the participants met the requirements. The lexical decision task, including the test items and fillers, is included in appendix 2.
Based on the results of the lexical decision task, 15 words were selected for the experiment in the next phase. For this selection, three aspects have been taken into account. First, none of the younger participants knew the right meaning of these 15 selected words in the lexical decision task. Second, the selected words were largely unknown to the older participants who performed this task. Third, of the words that were eligible, these 15 words had the highest calculated frequencies during the period of 1965 until 1990. Appendix 3 provides an overview of the selected words. The selected words from the lexical decision task are used in the relearning/learning experiment.
4.2.2 Relearning/learning experiment
The relearning/learning experiment was a self-designed experiment in which both younger and older adults participated. This experiment included 30 words that could be divided into two different categories: existing and pseudo words. The 15 existing words were selected from the lexical decision task, as is discussed in the previous subsection. These existing words occurred in former language use. The task also included 15 Dutch pseudo words that were previously unknown to all participants. For every existing word, a pseudo word was developed that matched the existing word in number of syllables. The form of the words and their meaning also played a role in the matching. For example, the pseudo word drettig matched the old word luimig since both words ends in the suffix –ig, and oparm matched the existing word keerweer based on a similar semantic transparency. Keerweer contains the words keer ‘turn’ and weer ‘again’. If someone walks in a dead end, then the person has to turn again to go back. For the pseudo word oparm the meaning ‘huge muscle’ was chosen. A muscle could be on top (op) of an arm (arm). An overview of the existing words and their matched pseudo words are given in appendix 3.
20 Although we tried to match the pseudo words as best as possible to the existing words, it was impossible to match precisely. Because of this, the control group of younger adults was very important. For them, there should be no difference between the existing and pseudo words. If the control group performed different on one of the conditions, it is possible that the items of that condition is easier or more difficult to learn.
The relearning/learning experiment contained 4 stages: (i) a pretest in which we controlled whether participants already knew the correct meaning of items before they (re)learned them; (ii) a learning task in which the participants were asked to learn the meaning of the presented items; (iii) a posttest that was conducted immediately after the learning period; (iv) a second posttest took place after the approximately 20 minutes. In the pretest and the posttests, the participants were asked to produce the meaning of the presented items. The items were presented in a randomized order. The experimenter judged the answers of the participants as correct or incorrect, which will be further discussed in subsection 4.4.
At the beginning of the learning task, participants were instructed to learn the meaning of the words that would be presented one by one. It was emphasized that the participants should try to remember the meanings because they would be asked to produce them later. The words and their meanings were offered in 5 rounds of 30 trials in which each word appeared once. Within these sets, the order was randomized. The learning task was self-paced: as soon as the participant pressed the space bar, the following word and its meaning were presented. As a result, all participants saw all words equally often, but they could process the instruction at their own pace. We did not impose time constraints, because this could result in a disadvantage for the older adults, who were expected to need more time to process the words and their meanings.
All participants were tested individually, in person, in a separate room. Only the experimenter was present in same room. All items were presented on a computer screen.
4.2.3 IQ test
The adults who participated in the relearning/learning experiment were asked to solve various non-verbal puzzles for 20 minutes. The puzzles test was the age-normed subtest ‘Matrix Reasoning’ of the third edition of Wechsler Adults Intelligence Scale (WAIS-III), a standardized intelligence task. The number of correct responses offered both a raw score and a score that was corrected for
21 the participant’s age. In general, an older participant was allowed to make more mistakes than a younger participant to get the same normalized score.
4.3 Procedure
After the participants of the first stage filled in the questionnaire, they performed the lexical decision task. Participants of the second phase started with the questionnaire as well. Then, the pretest of the relearning/learning experiment was conducted followed by the learning task. After the learning task, the participants immediately engaged in the first posttest. Then, they performed the IQ test, as is described in the previous subsection. Apart from the fact that the IQ scores would help us to analyse the data, the IQ test had an additional function. By solving different puzzles the focus of the participants diverted from the word learning experiment to something else. This was important because the participants participated in the second posttest after solving the puzzles. By diverting the participant’s attention, it became possible to detect possible differences between the responses on the first and second posttest for both age groups. If a participant solved the puzzles of the IQ test in less than 20 minutes, the participant had to wait the remaining time before the second posttest was conducted. We tried to keep the time between the first and second posttest the same for all participants.
4.4 Data analysis
The responses of the participants during the pretest and both posttests were scored as 1, indicating that the participant produced the correct meaning of the presented item, or 0 if the participant did not know the meaning of the presented item or produced an incorrect meaning.
Regarding to the IQ test, the number of correct answers was calculated for each participant and used as a raw score. In the analyses, the raw scores were corrected for the participant’s age based on the age norms provided by the manual of the WAIS-III in order to make the scores comparable between participants. Finally, the scaled IQ scores were used to statistically correct for intelligences differences between in the groups in the relearning/learning experiment.
To test the hypotheses as described in section 3, we conducted a logistic regression analysis using the generalized linear mixed model from the lme4 package (Bates et al. 2015) in R. This
22 statistical model was used because of the repeated measurements during the experiment. Moreover, a generalized linear mixed model allowed us to exclude certain trials from further analysis. A trial is the response of an individual participant for a specific item and test. Age group (old and young), condition (existing or pseudo), the IQ score and their interactions were included as predictors for the outcome response. The binary predictors (age group and condition) were contrast coded. The IQ scores were mean centered by group. We included participant and items as random factors.
5 Results
This section reports the results of this study. Subsection 5.1 discusses the results of the IQ test. The results of the relearning/learning experiment will be discussed in subsection 5.2. This latter subsection discusses respectively the results from the pretest, the first posttest and the second posttest.
5.1 Results of the IQ test
As is discussed in subsection 4.2.3, the older and younger participants performed the task “Matrix Reasoning” of the WAIS-III. For this intelligence task, the participants were asked to solve as many puzzles as possible, with a maximum of 26, in 20 minutes. The scaled scores for IQ were normally distributed according to the Q-Q plot and histogram. The results of the intelligence task for both the older and younger participants are shown in Table 4.
Table 4. Participants characteristics of the participants of phase 2, incl. the scaled IQ score
young (n=20) old (n=29)
age, y 23.7 ± 3.4 72.6 ± 6.6
male sex, n (%) 6 (30.0) 18 (62.1)
23 An unpaired t-test was used to find out if there is a significant difference between the means of the IQ scores of the older and younger group. Although the older group had a slightly higher mean IQ score than the younger group, this difference was not significant (t(36.486) = –0.958, p = 0.34).
5.2 Results of the relearning/learning experiment
This section discusses the results of the older and younger age group on the relearning/learning experiment. The results of 49 participants were used in the analyses of the experiment. As discussed in subsection 4.2.2, a pretest was conducted in which each participant was asked to indicate whether they knew the meaning of the 30 presented words or not. Only six older participants provided in total 11 correct answers on the 1470 trials (30 items * 49 participants = 1470) during the pretest. None of the older and younger participants claimed to know the meaning of the pseudo words. The items that were known and the number of participants who knew the correct meaning before the learning task are shown in Table 5.
Table 5. Items that were known to the older participants during the pretest
Items that were known before learning Number of participants who knew the
meaning correctly before learning
genaken 1 katijf 1 keerweer 1 luimig 2 oorlof 2 snaphaan 3 wildzang 1
The trials in which the items were known already at the pretest were eliminated from the final dataset, meaning that the total number of observations was 1459.
Table 6 and Figure 3 summarises the results of both the first and second posttests. The tables present the mean proportion of correct responses on existing and pseudo items per age group of both the first and second posttest.
24
Table 6. Mean score of correct meanings at the first and second posttest for both older and younger participants
Age group n First posttest Second posttest
Existing words Pseudo words Existing words Pseudo words
Young 20 0.58 0.48 0.53 0.39
Old 29 0.70 0.70 0.67 0.71
Figure 3. Mean score of correct meanings at the first and second posttest for both older and younger participants
Table 6 and Figure 3 show that the proportion of learned items was higher for the younger group in general than for the older group. The older group scored lower at the second posttest when compared to the first one, for both existing and pseudo words. Furthermore, Figure 3 shows that adults generally found pseudo words harder to learn than existing words, while younger participants did not make a difference between existing and pseudo words. If a significant interaction between age ground and condition is found, this pattern of results is in line with the predictions. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
existing pseudo existing pseudo
old young
mean score correct items at first posttest
mean score correct items at second posttest
25 5.2.1 First posttest
Logistic regression analyses using the generalized linear mixed model showed a significant main effect for age, i.e. the younger adults scored significantly higher at the complete first posttest than the older adults (OR = 3.561, 95% CI = [1.312, 9.670], z = 2.492, p = 0.013). However, there was no a significant effect for neither condition (OR = 0.615, 95% CI = [0.181, 2.098], z = –0.775, p = 0.438) nor IQ (OR = 1.137, 95% CI = [0.914, 1.414], z = 1.153, p = 0.249). The hypothesized interaction between group and condition did not reach significance, although the effect size associated with the interaction is fairly large (OR = 1.844, 95% CI = [0.821, 4.141], z = 1.483, p = 0.138). No significant interactions were observed between age group and IQ (OR = 1.187, 95% CI = [0.769, 1.832], z = 0.773, p = 0.440) and between condition and IQ (OR = 1.036, 95% CI = [0.872, 1.230], z = 0.399, p = 0.690). However, a significant interaction between age group, condition and IQ was found (OR = 0.716, 95% CI = [0.514, 0.998], z = –1.974, p = 0.048). This latter interaction is visualized in Figure 4 and 5. The scores on the pseudo words of both the older and younger group at the first posttest as a function of IQ are shown in Figure 4. Figure 5 shows the scores on the existing words of both the older and younger group at the first posttest as a function of IQ.
26 Figure 4. Scores on the pseudo words of both the older and younger group at the first posttest as a
function of IQ
Figure 5. Scores on the existing words of both the older and younger group at the second firsttest as a function of IQ
27 Figure 4 shows that the chance on a higher score on the pseudo words increases as a function of IQ for both the older and younger adults. In other words, a higher IQ score is associated with an increased chance to produce a correct meaning for pseudo words. As we can see in Figure 4, the difference between older and younger adults with respect to the chance of a correct response decreased as the scaled IQ score increased. Similar to Figure 4, Figure 5 shows that the score on the existing words of the younger group increased as function of IQ. However, for the older group the relation between performance on existing words and IQ score was not observed. In other words, the role of IQ was different for older and younger participants when focusing on existing words.
5.2.2 Second posttest
Investigating the scores at the second posttest, the statistical analysis revealed a significant main effect for age. This means that the younger adults scored significantly higher on the complete second posttest than the older adults (OR = 6.676, 95% CI = [2.257, 19.749], z = 3.431, p < 0.001). The model did not find a significant effect for condition (OR = 0.665, 95% CI = [0.174, 2.549], z = –0.595, p = 0.552) nor IQ (OR = 1.143, 95% CI = [0.910, 1.437], z = 1.151, p = 0.250). Furthermore, a significant interaction was found between age group and condition (OR = 3.381, 95% CI = [1.311, 8.720], z = 2.521, p = 0.012), i.e. the difference between the older and younger group was larger for pseudo words than for existing words. However, there were no significant interactions between age group and IQ (OR = 1.196, 95% CI = [0.756, 1.892], z = 0764, p = 0.445), between condition and IQ (OR = 0.987, 95% CI = 0.849, 1.160], z = –0.158, p = 0.874), and between age group, condition and IQ (OR = 0.733, 95% CI = [0.523, 1.023], z = –1.817, p = 0.069).
5.2.3 Lost items
Moreover, we were interested in whether existing words were more likely to be retained at the second posttest. Therefore, only items that were learned on the first posttest were analyzed to see if they were still known on the second posttest. Items on the second posttest were marked as 0 if they were retained and 1 if they were lost. Figure 6 shows the mean proportion of lost items, specified for both age groups and both conditions. The term “lost items” refers to items that were known at the first posttest, but not at the second posttest.
28 Figure 6. Mean proportion of lost items for both the older and younger age group
Figure 6 shows that the older age group lost more items than the younger group in general. It also shows that the older group lost more pseudo words than existing words. The younger group of participants, on the other hand, lost more existing words. We expected that older adults would be more likely to lose more pseudo words than existing words at the second posttest, while younger adults would lost both types equally. With respect to the older group, the descriptive analysis of the lost items is in line with our hypothesis. However, for the younger group we did not expect a difference between their scores on the existing and pseudo words. A logistic regression analysis using the generalized linear mixed model should give a decisive answer to the question whether there is a significant interaction between age group and condition regarding the lost items.
The results from the statistical analysis showed a significant main effect on age group (OR = 0.142, 95% CI = [0.046, 0.437], z = –3.401, p < 0.001). However, there was neither a significant main effect for condition (OR = 1.039, 95% CI = [0.347, 3.112], z = 0.068, p = 0.946) nor for IQ (OR = 0.979, 95% CI = [0.780, 1.200], z = –0.201, p = 0.841). Furthermore, the model revealed a significant interaction between age group and condition (OR = 0.222, 95% CI = [0.053, 0.930], z = –2.059, p = 0.040). This interaction indicated that elderly participants were more likely to lose pseudo words and retain exiting words, while this pattern was reversed for the young participants. No significant interactions were found between age group and IQ (OR = 0.983, 95% CI = [0.667, 1.449], z = –0.085, p = 0.932), condition and IQ (OR = 1.072, 95% CI = [0.838, 1.373], z = 0.555,
0 0.05 0.1 0.15 0.2 0.25
existing pseudo existing pseudo
29 p = 0.579), or age group, condition and IQ (OR = 1.207, 95% CI = [0.771, 1.888], z = 0.823, p = 0.410).
5.2.4 Gained items
In addition, we wanted to know whether existing words were more likely to be gained at the second posttest if they were not known at the first posttest. Therefore, only items that were scored as incorrect during the first posttest were analyzed to see if they were known on the second posttest. Items on the second posttest were marked as 0 if they were unknown and 1 if they were known to the participants. The mean proportion of the items that were gained at the second posttest are shown in Figure 7.
Figure 7. Mean proportion of gained items for both the older and younger age group.
From Figure 7, we see that the participants of the younger group gained more items at the second posttest than the older group. Figure 7 also shows that the older participants knew more existing words than pseudo words at this second posttest when they did not produce the correct meaning of these items at the first posttest. In contrast to the older age group, the younger group produced more often the correct meaning of pseudo words at the second posttest when they did not produce the correct meaning at the first posttest. If there is a significant interaction between group and condition, the results are in line with the predictions as discussed in section 3.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
existing pseudo existing pseudo
30 A logistic regression analysis using the generalized lineair mixed model showed that there is a main effect for age group (OR = 6.41, 95% CI = [1.678, 24.500], z = 2.717, p = 0.007). No main effect was found for condition (OR = 0.418, 95% CI = [0.117, 1.486], z = –1.348, p = 0.178) or IQ (OR = 0.910, 95% CI = [0.684, 1.210], z = –0.651, p = 0.515). Furthermore, there were no significant interactions between age group and condition (OR = 4.981, 95% CI = [0.825, 30.065], z = 1.751, p = 0.080), age group and IQ (OR = 1.561, 95% CI = [0.872, 2.794], z = 1.499, p = 0.134), condition and IQ (OR = 0.884, 95% CI = [0.562, 1.392], z = –0.531, p = 0.596) and age group, condition and IQ (OR = 0.681, 95% CI = [0.274, 1.692], z = –0.827, p = 0.408).
6 Discussion
The aim of this study was to answer the question whether residual lexical knowledge exists in monolingual Dutch elderly who lost words of their native language. Therefore, the study found out whether words that are no longer used are truly forgotten, or residues of this knowledge still exist. To this end, the relearning/learning experiment was carried out on a test group consisted of older adults from 65 years and older and the control group of younger adults between 18 and 30 years old.
The first hypothesis of this study was that elderly would have a relearning advantage of the existing items over the pseudo items. It was predicted the older participants would score higher on the existing words than the pseudo items at the first posttest, whereas we did not expect to find such a difference for the younger age group. However, no significant interaction between age group and condition was found. An interesting result of this study is that there was a significant interaction between age group, condition and IQ at the first posttest, indicating that a higher IQ score is associated with an increased chance to produce a correct meaning for the words that were learned during the learning task. However, the results suggested that this is only the case for words that were previously unknown to the participants. The IQ scores of the older adults were not associated with a higher score on the existing words at the first posttest. It seemed that a different factor is involved in learning the existing words for the elderly, which could be residual lexical knowledge.
The significant interaction that was found between age group and condition at the second posttest is in line with the first hypothesis. The elderly scored higher on the existing words than
31 the pseudo words at the second posttest, whereas the younger participants performed slightly better on the pseudo words. The results of the younger adults suggested that the pseudo words might be easier to learn than the existing words, although we have matched the pseudo items to the existing ones. Interestingly, even if the pseudo words were easier to learn, the older participants performed much better on the existing words at the second posttest. The relearning advantage for existing words over pseudo items for the older adults only confirmed that residual lexical knowledge is still present.
The other hypothesis was that the older adults would be more likely to lose more pseudo words than existing words at the second posttest, while younger adults would lose both types equally. The results of this study were in accordance with this hypothesis. The results showed that existing words were more likely to be retained by the older age group at the second posttest than pseudo words. We observed a significant interaction between age group and condition by investigating the lost items, indicating that the older age group was more likely to lose pseudo words and retain existing words, while this pattern was reversed for the younger group. The results with respect to the older participants can be explained on the basis of the savings effect: relearning reactivates the words that were known at some point in time to a higher level than previously unknown words can reach. Older adults reactivated more existing words than pseudo words. Therefore, the existing items were also memorized for a longer period than the pseudo words.
Finally, a borderline, but not significant interaction was found between age group and condition with respect to the gained items. This could mean that there is no interaction between the included variables or that there is too little power. Possible reasons might be that there were too few participants or too few items included in the test, creating the possibility that significant interaction is found when power improves.
The results are in line with the findings discussed in De Bot et al. (2004), De Bot & Stoessel (2000), Hansen (2001), Hansen et al. (2002) and Van der Hoeven & De Bot (2012). These authors found evidence for the savings paradigm by confirming the relearning advantage for old over new information. However, the previous studies were limited to language loss in a bilingual context, while the present study focused to language loss in monolingual elderly. Interestingly, similar results were found.
This study has a number of limitations which should be taken into account. First and foremost, this study is based on the assumption that the older participants knew the existing words
