How time can change lexical retrieval: The instability of the mental lexicon

(1)

How time can change lexical retrieval:

The instability of the mental lexicon

Kristy Kuik

Rijksuniversiteit Groningen

Applied Linguistics

Supervisor: Wander Lowie

Date: 21 June 2019

Word count:

15.602

(2)

Abstract

The mental lexicon along with its processes seems to be constantly changing over time. Although many linguists have acknowledged that lexical retrieval is affected by language use, current research designs still often fail to include this instability in their methods. Activation levels change over time due to, among others, frequency and recency of occurrence, which affects the mental lexicon and, in turn, the processes involved in word recognition. The present study intended to establish whether the instability of the mental lexicon could also be detected when examining native language (L1) and second language (L2) reaction times to time-specific lexical items two months apart (during two different seasons). The lexical items were categorized as words related to either winter, spring, or to no particular season at all. Two lexical decision tasks were conducted in two homogenous groups of Dutch native speakers who are advanced speakers of English: once during the winter and once during the spring. One group performed both tasks in the L1 and the other group in the L2. The results hint at a facilitatory effect of the variable season (hereafter Season) on the processing of words related to that season, likely due to higher activation levels. For both groups, reaction times (RTs) to the winter words were significantly faster during the winter than during the spring. However, only the L2 group demonstrated significantly quicker latencies for spring-related words during the spring than during the winter. It can be concluded that time has an effect on the mental lexicon. Although some trends have emerged from the data, more research is needed.

(4)

How time can change lexical retrieval: The instability of the

mental lexicon

Introduction

Language processing can be deemed one of the most complex phenomena that human minds are capable of. Many scholars have attempted to explain such processes and, throughout the years, the possible explanations have greatly evolved. Numerous models and metaphors have come into existence to help us understand the functioning of our minds. Initially, a spatial metaphor in which languages are expected to be stored separately in the mind gained popularity (Simpson & Krueger, 1991) but more recently, the activation metaphor in which languages are stored together and are connected in one large network is widely applied (Dijkstra & van Heuven, 2002). The latest and most complete model, which will be discussed in more detail later in this paper, is the Multilink model (Dijkstra et al., 2018). This computational model was designed to capture the vital and essential mechanisms and characteristics of language perception and production for both mono- and multilinguals. The activation metaphor lies at its core, which explains that lexical items and their orthographic, phonological, and semantic representations are connected in a huge network and that activation flows bidirectionally through it. Connections can constantly fluctuate in strength depending on the levels of activation which, in turn, can be affected by factors such as frequency or recency of use or occurrence. The observation that frequently provided information in learning tasks is memorized better or, in other language processing tasks, is responded to faster, is called the frequency effect (Balota, Yap, & Cortese, 2006; Diessel, 2007; Ellis, 2016). A type of frequency effect is the recency effect, where more recently provided information is memorized better (Pfänder & Behrens, 2016; Takashi, 2009; Talmi & Goshen-Gottstein, 2006). With these principles in mind, processing lexical items in a first language (L1) can be expected to be faster than second language (L2) processing, as these words are likely to be accompanied with higher activation levels due to more frequent and often also more recent use. Because frequency and recency are important factors that influence activation levels, language proficiency can be expected to play a role as well, since proficiency is usually closely associated with frequency and recency of use.

L1 and L2 lexical retrieval are topics that have already been extensively researched, yet there are still many issues to explore. For instance, it has been acknowledged that the mental lexicon can change depending on language use, but many research designs have still adopted static displays and failed to include time as a factor (Kroll & Stewart, 1994). Moreover, although frequency and recency effects were found to greatly influence the activation levels within the mental lexicon, these effects were mostly investigated in stable settings and short-term timescales at the levels of milliseconds, seconds, and minutes (Pfänder & Behrens, 2016; Talmi & Goshen-Gottstein, 2006). Studies into a long-term (i.e. relative) recency effect and the instability of lexical

(5)

representations on a long-term timescale, such as at a level of months, seem to be lacking. An interesting way of investigating the instability of the mental lexicon on a larger timescale would be to observe lexical decision latencies of time-specific lexical items, such as season-related vocabulary, on two different testing moments a few months apart. This way, effects of relatively recent and frequent activation might be detected on a larger timescale. Adding an L1 versus L2 perspective to the equation may contribute greatly to gaining insights into lexical retrieval and its instability over time. Especially examining a possible interaction between the variables time (i.e.

season of testing) and language (hereafter Time, Season, and Language1_{) may be valuable. A}

somewhat special group of participants may be interesting to observe, for instance, Dutch native speakers who are highly advanced learners of English as a second language (ESL) and use English extensively on especially an academic level. Apart from theoretical significance, a study like this may also bear practical importance in terms of the need to reevaluate current research designs as they tend to exclude time as a factor.

The main research question of the present study is: Can an instability of the mental lexicon

also be detected when examining the L1 and L2 reaction times to time-specific lexical items during two different seasons? When examining L1 and L2 lexical decision latencies of time-specific lexical

items, such as the season-related words SNOW (for winter) and EASTER (for spring) during two different seasons, the following expectations may arise: An instability of the mental lexicon can be detected, as the latencies of season-related lexical items should be faster during the corresponding season than during an unrelated season, due to temporarily risen activation levels. Although the effects may vary among different languages someone speaks, the effects are expected to be detectable to at least some extent in both languages. The next section of this paper will discuss the most relevant literature in relation to the mental lexicon and which model can be used best to explain its functioning. After introducing the literary framework and posing the research questions and hypotheses, the research design will be explained. Then, the results will be documented and discussed before drawing conclusions.

Background

1. The computer metaphor

It is a widely held view that the human mind is capable of many complex processes which work in mysterious ways. This, however, does not prevent cognitive scientists, among others, from trying to unravel these mysteries. How are we able to do all kinds of extraordinary actions, such as making decisions or processing language? Having a conversation may seem effortless, but the cognitive processes are quite complex. The mind must convert the perceived sounds into words, and consequently convert the words into sentences. In turn, the mind needs to retrieve the meaning and begin to formulate a reply by choosing the correct words, grammatical structure,

(6)

and phonology. Finally, the mind must signal the tongue and the muscles in the face to actually pronounce the sentence. The mind has to select the right language - should an individual know more than one - in all of these steps. With these and many more processes working at an impressive speed, human cognition can be deemed convoluted. A particularly alluring field within the rhetoric of human cognition science is the process of lexical retrieval and how lexical items are stored in our brain. For many decades scientists have engaged in discussions about what metaphor or approach would be most appropriate to explain the (multilingual) mental lexicon and lexical retrieval specifically.

A traditional framework that has been widely applied is the computational approach (Newell, 1994), which uses a computer as a metaphor for the functioning of the mind. As summarized by Van Gelder (1995), the computational approach claims that “the mind is a special kind of computer, and cognitive processes are the rule-governed manipulation of internal symbolic representations” (p. 345). Computers are organized around essentially sequential modules. This is to say, cognitive processes such as lexical retrieval follow sequential, identifiable steps from one point to another, leading to the recognition or retrieval of a certain internal representation. With the mind as a computer, it can manage many different basic operations and combine them into complex tasks (Newell, 1994). It should be noted that equating the human mind with a computer also has its limitations. A point of criticism can be that a modular approach is less likely to account for factors such as change or time compared to continuous processing models. The activation metaphor, for example, is a more powerful and general model than the computer metaphor, as it allows continuous changes over time to take place in the mind. A strong point of the activation metaphor is that it resembles neural activation in the brain. Mental entries can differ in their degree of activation, with activation levels increasing after a certain event or decreasing after a lack of certain events (McClelland & Rumelhart, 1981). For instance, after coming across a certain lexical item the activation level rises, whereas not seeing or hearing an item would decrease the activation level over time. By this logic, the lexicon cannot be deemed static as activation levels are constantly changing (De Bot & Lowie, 2010). More elaborate insights into the mental lexicon and the processes and effects of activation will be discussed at a later stage in this paper.

Despite the criticism, the computational approach has served for decades as an insightful and influential framework to cognition research. Throughout the years many models and metaphors have emerged within the lines of the computational framework, yet with slight adaptations to fit a more contemporary view as to language processing. For example, one of the most important adaptations was that the slightly outdated belief that language processing takes place in sequential steps (serial processing) has evolved into the assumption of parallel processing. Discussing all computer-based models and their progressions in detail would, however, go beyond the scope of the current study. Therefore, only a few of the most important

(7)

models with the computational approach at their core will be referred to in the next section, before discussing one of the most contemporary and accepted language processing models in more detail.

Within the lines of the computational framework many models have emerged, each with different focuses or slightly different starting points. To begin with, these computational models could differ from each other in their fields of focus. Some researchers chose to develop a monolingual model to describe language processing, such as The Speaking Blueprint (Levelt, 1989, 1993), whereas others geared towards a bilingual one (De Bot, 1992). The specific area of language processing that researchers focused on could also differ: some were language production oriented, such as De Bot’s adaption of the Speaking Blueprint or Kroll and Stewart’s Revised Hierarchical Model (De Bot, 1992; Kroll & Stewart, 1994), whereas others focused on language

perception, such as the Interactive Activation (IA) and the Bilingual Interactive Activation

(BIA/BIA+) model (Dijkstra & Van Heuven, 2002; McClelland & Rumelhart, 1981). Furthermore, some researchers had opposing views as to how different languages are stored in our brains. On the one hand there are researchers like Simpson and Krueger (1991) who were in favor of the

selective access hypothesis. They adhere to the view that different languages are stored separately

in the brain: only the target language will be selected and activated for use. The metaphor often used to clarify this hypothesis is the spatial metaphor, in which lexicons or parts of lexicons are presumed to be situated in separate locations in the brain. On the other hand, more contemporary research found evidence for the non-selective access hypothesis, which posits that lexical storage is language-independent (De Groot, Delmaar, & Lupker, 2000). Hypotheses and models like this one assume that mental entries are part of a large network and are connected to one or many other entries. In this section it has been explained that a computer-based framework has shaped many models regarding language processing. The chapter that follows will discuss the latest accepted model that is unique in the sense that it includes both multilingual language production and perception: the Multilink model (Dijkstra et al., 2018).

2. Multilink

With many models capturing only a small fragment of language processing, the need for a more complete model that would capture the vital fundamental mechanisms and characteristics of language processing came into existence. Therefore, Dijkstra and his colleagues (2018) proposed Multilink: “a computational model designed to provide a general account of monolingual and bilingual word retrieval in comprehension and production” (p. 16). As Multilink merges both language production and perception processes into one theoretical paradigm, it has become an auspicious model for language processing. The activation metaphor lies at the core of Multilink: the model was designed as an interactive network of orthographic, semantic, and phonological representations, with these variables interacting in an intricate manner. Multilink allows simulations of the interactions between these and numerous other features.

(8)

Figure 1.”Standard network architecture of Multilink. Note: Input is indicated by blue underscore, orthographic (O)

representations by green underscore, semantics (s) by ‘concept’, phonological (P) representations by slashes. EN = English, NL = Dutch. Output is task dependent. Here slashes indicate a phonological output in the same or a different language (for word naming or translation)”. Retrieved from Dijkstra et al. (2018, p. 6).

Figure 1 shows the core structure of Multilink, which is a symbolic lexical network through which activation flows and spreads. As the authors explain: “a written input word activates various lexical-orthographic representations, which in turn activate their semantic and phonological counterparts, as well as associated language membership representations (English or Dutch)” (Dijkstra et al., 2018, p. 6). By way of illustration, for visual word recognition this would mean that when people read an input word (e.g. HOOD), many language-independent lexical items which are similar in orthography will be co-activated, such as the English words FOOD and HOLD, and the Dutch words HOOS and HOND. As a consequence, their connected semantic representations or concepts will be activated in a language nonselective way. If Multilink were a unidirectional model, the activation flow would only move forward from the concept level to the phonology level and thus activate /hʊd/. However, since Multilink is bidirectional, the flow of activation may also go back a level and activate other orthographic representations that are associated with the activated concept. For the input word HOOD it would signify that after the activation of its semantic concept, the Dutch orthographic representation KAP will also be activated, because HOOD and KAP share the same concept. In turn, the associated phonological representations /hʊd/ and /kap/ will become activated. Note that when FOOD is also triggered on a lexical-orthographic level for the input word HOOD, the Dutch variant VOEDSEL will also be activated, along with its phonological representation /vutsel/. Finally, depending on the task at hand, the mind will have to select the most appropriate representation from all activated ones.

(9)

Let us assume a Dutch-English bilingual performing a lexical decision task with English as the target language. The word HOOD should be recognized as an English word, as the orthography, semantics, and phonology are all correctly familiar in the target language. The input word HOND, however, should be rejected as an English word: although the orthography, semantics, and phonology are all familiar to the participant, these activations do not belong to the target language of the lexical decision task.

As has been partially implied by the section above, there are several assumptions that underly the current Multilink model: 1) There is little evidence for separate lexicons; 2) There is language non-selective access and parallel activation of word from neighbors; 3) Translation is not done via word association but only by conceptual mediation by connecting word forms from different languages only via their semantics; and 4) The connections between L2 words and their meanings are stronger than proposed in RHM (Dijkstra et al., 2018, pp. 4-5). However, the authors also mentioned a fifth assumption that has not been implemented in their model yet: 5) It may be necessary to make a distinction between language-dependent and language-independent semantic features. With these assumptions in mind, the authors invented a model that supports monolingual and bilingual word processing, including lexical items that differ in frequency of usage, item length, and cross-linguistic similarity. Moreover, the model can differentiate various psycholinguistic tasks when simulating word processing, as it includes a task/decision system. What is also promising is Multilink’s ability to adapt to both high and low L2-profciency bilinguals in these tasks.

Despite having delivered a solid basis for a new language processing model, Dijkstra and his colleagues have also offered constructive suggestions as to future steps that need to be taken to improve Multilink. Firstly, the task/decision system in Multilink should be extended in order to wholly consider task-dependence in lexical processing. Secondly, Multilink is quite beta-oriented, therefore, this theoretical paradigm still needs to be tested in profuse, dedicated empirical studies (Dijkstra et al., 2018). Documentation and analyzation of real language use should also be included, as the current model is strongly based on experimental settings rather than on real language use. Although the model is quite flexible and allows activation to constantly flow bidirectionally, another point of criticism of the present Multilink model is that semantic representations are deemed invariant and stable, whereas they might not be. The current version of Multilink (2018) accepts that “semantic representations are [assumed] simple holistic units” and “semantic spreading of activation between associated representations is left unconsidered” (p. 6). Semantic representations may not be simple units after all, as De Bot and Lowie (2010) have argued that representations are quite variable due to differences in activation. For example, when the lexical item FOOD is triggered, on a semantic level the concepts ‘apple’ or ‘Chinese’ may also be activated. Which semantic concepts will be activated, may depend on whether someone has come across them frequently, recently, or has a certain emotional attachment to them. Even

(10)

within a semantic concept such as ‘apple’ there may be changes over time: at some point someone might prefer red apples and this concept will be activated, whereas at another point it might be green apples. Hence, semantic activation can be deemed more dynamic than the Multilink model illustrates.

3. Activation

As explained by means of the Multilink model, activation plays an important role throughout the lexical network in our mind. The activation metaphor explains that “all lexical items are part of the same network [and] as lexical items are connected through this network, the activation of one lexical item may interactively affect the level of activation of all the lexical items it is attached to” (Bot et al., 2005, p. 46). Activation levels are influenced by many interacting factors, such as the recency of encounter or usage, the frequency of encounter or usage, perceptual salience, or emotional attachment to certain representations, which may cause activation levels to continuously change over time (Pfänder & Behrens, 2016). Thus, the mental lexicon is unstable as “each candidate’s activation pattern changes over time depending on input from various sources” (Dijkstra et al., 2018, p. 9). The activation levels of lexical items and their semantic representations are affected by many factors, with the frequency and recency of activation as exceedingly significant influencers. Both will be explained in more detail in the next section.

The frequency effect is the phenomenon where frequently provided information in learning tasks is memorized better or, in other language processing tasks, is responded to faster (Balota et al., 2006; Diessel, 2007; Ellis, 2016). For word recognition this would indicate that high-frequency words are accompanied with faster recognition speeds compared to low-high-frequency words, because the higher frequency of those items leads to higher activation levels as can easily be accounted for by the Multilink model. Perea, Rosa, and Gómez (2005) investigated the frequency effect for pseudowords in a lexical decision task and in two of their experiments they replaced only one internal letter of a baseword to create a nonword. The results of the experiments show that nonwords which differed only one letter from a low-frequency baseword are categorized faster than high-frequency ones. Lexical decision for high-frequency pseudowords can be deemed more difficult than for low-frequency pseudowords, because these nonwords are noticeably similar to real lexical items. As these existing lexical items are greatly familiar to participants due to their general high frequency of occurrence, it takes more time to differentiate between the actual word and the nonword. However, there are other studies that yield the opposite result or that found no baseword frequency effect at all (Yap, Sibley, Balota, Ratcliff, & Rueckl, 2015). Yap and his colleagues stated that they found no significant correlation between nonword response time and baseword frequency in the lexical decision task, yet the effect seemed to have a greater influence for participants with higher vocabulary knowledge (Yap et al., 2015). Although there are controversial findings into frequency effects for pseudowords, most researchers will agree that, for real words, high-frequency words are processed faster than

(11)

low-frequency words, because higher activation levels facilitate lexical processing. In this section. the frequency effect was discussed from a general word frequency perspective. The next section will discuss the frequency effect from a more individual perspective.

Apart from the general word frequency, the frequency effect can also apply more locally to lexical items that are encountered more often than others by individuals, and will therefore differ among people (Dijkstra et al., 2018). Frequency of occurrence is not a static concept, rather it can vary as a result of the “strength and nature of our experience” (Pfänder & Behrens, 2016, p. 6). For instance, simply seeing a word repeatedly on a few billboards may result in lower activation levels compared to that same word repeated in an important context that an individual may care about. Thus far it has been established that the activation levels of lexical items within individuals may vary continuously depending on frequency of occurrence and the strength of the experience (Pfänder & Behrens, 2016). When representations are no longer activated, they gradually decay towards their resting level activation (RLA). In Multilink, the RLA of lexical items is determined by the overall word frequency distribution and the actual reaction time distribution for comprehension of the items. More frequent words are expected to have a higher resting level activation because they have been activated regularly, which may result in faster recognition or selection. Among other explanations, the frequency effect may also account for highly regular phenomena such as code-switching. A person might use a word from another language than the target language, due to a much higher activation value for that particular concept resulting from its higher frequency (De Bot et al., 2005).

Sometimes the frequency effect interacts with other processing aspects such as the recency effect. The recency effect is the tendency for individuals to remember more recently provided information better. Two of the most important factors that characterize the recency effect were addressed by Pfänder and Behrens (2016). Firstly, the activation is stronger when the amount of time that has elapsed since the last occurrence of an item is short. Secondly, the activation will be stronger depending on the more often an item has occurred recently. As a consequence, the speed of word processing will be facilitated by recent and frequent activation. The recency effect has been abundantly found in recall tasks (Pfänder & Behrens, 2016; Szmrecsanyi, 2006). Research has shown that when the number of items to be recollected is larger than the memory capacity, the contents of the visual short-term memory are severely affected and a strong recency effect arises (Kool, Conway, & Turk-Browne, 2014). In other words, when the set size to be memorized was within the capacity of the short-term memory, the entire sequence was equally likely to be stored, but with larger set sizes the more recent items were recalled with higher probability. Additionally, Takashi (2009) found evidence for a stronger recency effect among participants who have a large working memory capacity in a recall task. He explained that people with a great working memory capacity were sensitive to recollection because “the working memory capacity has a positive relation with the recollection process” (Takashi, 2009, p. 548).

(12)

The recency effect has been profoundly observed in short-term time frames: the effect seemed to disappear when the time frame expanded. The time frames mostly used in previous studies expanded from immediate recall to slightly delayed recall, so, more specifically, they expanded from a millisecond-timescale to a second-timescale respectively (Pfänder & Behrens, 2016; Takashi, 2009). In contrast to earlier findings, Talmi and Goshen-Gottstein (2006) found evidence that a long-term recency effect in recognition memory may exist. Their experimental conditions included “an immediate test, a delayed test after a filled interval, and a continuous distractor paradigm in which the same filled delay preceded the first word and followed every study word” (p.424). They found that “the long-term recency effect in continuous-distractor recognition was equivalent to the recency effect in immediate recognition” (p. 424). They concluded that the absence of a recency effect in the delayed test could not be attributed to the use of an alleged short-term memory store. In their study they adhered to the dual-store model which proposes the existence of a highly accessible short-term memory store and a long-term memory store. Which is to say, whenever a certain amount of time had passed and an item to memorize could no longer be found in the short-term memory store, it moved to the long-term store. Talmi and Goshen-Gottstein (2006) explored both the short-term and long-term timescales: In the immediate recall condition the timescale explored was at a level of milliseconds and thus, short-term; In the delayed condition the timescale used was at a level of seconds, for the distractor task at the end of the item sequence was 15 seconds; In the continuous-distractor task the timescale explored was at a level of minutes, because after each item there was a distractor task of 15 seconds, which delayed the eventual recall with a total of several minutes (Talmi & Goshen-Gottstein, 2006). Thus, they found that the recency effect not only exists at a level of milliseconds (i.e. a short-term recency effect) but also at a level of minutes (what they call a long-term recency effect). Considering the Multilink model, the recency effect appears to be accounted for as the model allows activation levels to constantly fluctuate, which may result in varying resting levels of activation (RLA’s) for items and differences in their processing speed.

Noted should be that the term recency effect is typically associated with the phenomenon that the last items in a list are recalled best: accordingly, it is associated with recall tasks and not so much with other tasks such as lexical decision. It would, therefore, be somewhat inappropriate to use the term recency effect in for instance a lexical decision paradigm. Additionally, despite the reported distinction between short-term and long-term effects (Talmi & Goshen-Gottstein, 2006), the time frame associated with the recency effect can still be considered particularly short (i.e. a timescale at a level of minutes). When discussing a longer time frame, such as a period of hours, days, weeks, or months, a different term might also be more suitable. Thus, when discussing recency effects in other settings than short-term recall, the descriptions relative recency effect and

(13)

All in all, the level of activation seems to be an important influence on the increase or delay in word recognition speed. As has been illustrated by Multilink, activation spreads throughout the mental lexicon with its values being far from static, for the activation levels can constantly change depending on factors such as frequency and recency. As recency can be considered a type of frequency effect, recency interacts with frequency. This is to say, when an occurrence happened recently, the frequency level of that occurrence also increased. However, this does not have to be the case vice versa, as higher frequency levels do not necessarily imply that the items were activated recently. Nevertheless, time seems to be a significant factor that influences activation levels, as well as the embodiments of representations.

4. First vs. second language processing

As has been established in previous sections, lexical processing is influenced by varying activation levels which are affected by, among other factors, frequency and recency. These factors may also account for the differences found between the processing of the native language and other languages. Research has shown that the processes involved in lexical retrieval are language-independent (Dijkstra & van Heuven, 2002; Dijkstra et al., 2018), yet the speed of processing seemed to be faster for the first language (L1) compared to the second language (L2) even for fluent, nearly balanced bilinguals (De Bot & Lowie, 2010; De Groot, Borgwaldt, Bos, & Van Den Eijnden, 2002; Kroll & Stewart, 1994; Plat, Lowie, & De Bot, 2018; Potter, So, Von Eckardt, & Feldman, 1984). Apart from factors such as age of acquisition, also language dominance and history of language use can be deemed important factors in the facilitation or inhibition of language processing in bi- or multilinguals.

As mentioned before, throughout many investigations the native language has shown a processing advantage over second or other languages, but these processes are not stable. One might argue that variability within these processes, especially in certain conditions, might not (only) be dependent on the targeted language, but may also be dependent on language exposure. To illustrate, the mental lexicon may change depending on what season it is, with typical winter vocabulary being highly activated during the winter but not during the other seasons. Variability in language processing may be much more present in the language that is predominantly used during such a period. If that language is for example the native language, variation across seasons might be more visible in the L1 than in the L2. However, more recent research suggested that the first language reports more stable results across experiments. According to De Bot and Lowie (2010), “the subsets of lexical items associated with a language that is not someone’s native language can be expected to be less stable … assuming that more frequent use and more frequent co-occurrence lead to less variable connections in the network” (De Bot & Lowie, 2010, p. 119). Faster processes in the L1 may be attributed to automatic spelling-to-sound conversion which is less available in the L2 (Plat et al., 2018). Plat and her colleagues found that, at a level of milliseconds, participants showed less variability in the L1 than in the L2. In other words,

(14)

participants showed more stable language use in their L1 compared to their L2, because the L2 is not as automatized. An explanation may be that “the decisions made in relation to the second language are subject to more uncertainty” (Lowie, Plat, & de Bot, 2014, p. 220). Even though it is assumed that there are no separate lexicons in the mind, differences between L1 and L2 lexical items can emerge due to L2 items generally being less fully developed and less strongly embedded than L1 items. Noted should be that differences like these do not have to be language-dependent, but may also emerge when comparing lexical items from different registers within languages. Although the effects of a more stable L1 than L2 are likely to diminish with increasing proficiency in the second language, the effects were still found even in highly proficient students of English.

Within individuals, differences in language processing between multiple languages may partially be attributed to frequency. The inferior language of bilinguals who do not master both languages equally well may suffer from lower activation levels due to lower frequencies of occurrence for those individuals. As a result of more frequent activation, the language which is mastered best has processing advantages over the inferior language(s) and will therefore have faster responses in lexical retrieval tasks (Mosca & De Bot, 2017). Even for balanced bilinguals, frequency values may differ greatly between languages as the dominance of the language spoken at a certain moment in time may also influence activation levels which in turn influence the processing speed. Lowie and his colleagues (2014) argued that the immediate history of language immersion affects the reaction times in a word-naming task: in a completely L2 immersion the reaction times to L2 words were faster compared to the reaction times to L2 words in an L1 context. As the L2 was used more frequently and recently, processing speed increased. It should be noted that processing differences between the L1 and L2 may also depend on the register needed or preferred in particular situations, as a certain type of vocabulary might be solely used in a certain context. To illustrate, some may be more proficient in the L1 for casual situations and prefer to use the L1 for such moments but prefer to use the L2 in formal or academic situations. With these studies in mind, the conclusion can be drawn that although the L1 is generally processed faster due to automatization, there are still many factors such as language dominance and immersion that influence the processing speed of both the L1 and L2. In other words, the speed of language processing may differ among languages, but the processes involved, no matter the language, are not stable and can constantly change over time to varying degrees.

As many scholars have failed to include time as a factor in their research designs, there is not much research into an interaction between the variables Time and Language. An interaction between Time and Language refers to a state where both the size and the course of each variable’s effect can depend on the value of the other variable. De Bot and Lowie (2010) tested a single participant for several weeks twice a day, once at 1pm and once at 6pm, and they reported a significant interaction between Time and Language in their research. To clarify, they found that the magnitude of the values of the mean response times in the L1 and L2 depended on what time

(15)

the experiment was conducted. This might be explained by the observation that the L1 is usually more stable and less sensitive to variation (De Bot & Lowie, 2010; Plat et al., 2018), whereas the L2 might be influenced more by the time of experiment conduction.

In a setting with a larger timescale, such as a timescale of several months, the principles of frequent and recent activation may suddenly play a role. The effects of time can then be expected to be more pronounced in the L1, because the L1 is expected to be more frequently and often also more recently activated and is therefore expected to have higher activation levels than the L2 (Pfänder & Behrens, 2016). Therefore, a difference between the L1 and L2 in terms of the level of influence that the variable Time has may be observed. However, noted should be that the proficiency level of the L2 can also play an important role here: for highly advanced learners of an L2 the activation levels of the L2 may be a lot more similar to those of the L1 – compared to learners with a lower L2 proficiency – due to more frequent and recent activation and higher resting levels of activation (Dijkstra et al., 2018). Therefore, learners with a lower L2 proficiency can be expected to be more likely to show a significant interaction between the variables Time and Language than highly advanced learners, as L1 and L2 activation levels vary to a greater extent. In other words, for highly advanced learners, the (subconscious) preference for a certain season in which to retrieve typical season-related vocabulary might not depend on whether the target language for retrieval is the L1 or the L2. Whether highly advanced learners of an L2 have faster responses in a certain season may not depend on the language they use, whereas for learners with lower proficiency levels Time may depend on Language. Nevertheless, more research into the interaction between Time and Language is needed, as different timescales can result in different effects of Time on the mental lexicon and its processes.

5. Statement of purpose

Much of our understanding of the mental lexicon has come from static displays. Much research into the mental lexicon has implicitly accepted invariant representations in their designs and failed to include time as a factor (Dijkstra & Van Heuven, 2002; Kroll & Stewart, 1994). It is often assumed that the mental lexicon is a solid unit we can build on, rather than a slippery area that is constantly changing. The processes involved in lexical retrieval and, more specifically, word recognition are actually affected by time and are therefore constantly changing. Recent research has shed some light on the instability of the mental lexicon due to fluctuating activation levels (De Bot & Lowie, 2010; Dijkstra et al., 2018; Spivey, 2007). As has been illustrated by the Multilink model of Dijkstra and his colleagues (2018), activation values are not stable as activation bidirectionally flows through our mental lexical network. Especially frequency and recency effects have been observed to influence activation levels and consequently the speed of language processing.

These effects of frequency and recency were, however, mostly examined in stable settings, disregarding real-time data (Pfänder & Behrens, 2016). Although some studies investigated the

(16)

instability of the mental lexicon (De Bot & Lowie, 2010; Plat et al., 2018), they did not focus on an instability from the perspective of time-specific vocabulary. In addition, research into the recency effect has largely engaged in short-term timescales at a level of milliseconds, seconds, or minutes (Talmi & Goshen-Gottstein, 2006). Research into relatively recent activation and the instability of lexical representations on a long-term timescale, such as at a level of months, seems to be lacking. What is also interesting is to combine the aforementioned insufficiently researched topics – long-term frequency and recency effects and the instability of the mental lexicon viewed from a larger timescale – and examine them from both a native language and second language perspective. Although one connected mental lexicon is assumed, differences in processing speed between the L1 and L2 have continuously been found (De Bot & Lowie, 2010; De Groot et al., 2002). Additionally, investigating a possible interaction between Time and Language may also contribute to our understanding of the instability of the mental lexicon and the effects of time. Along with theoretical importance, the present study may also bear practical significance for the shape of research designs. Different times can influence the mental lexicon and, in turn, word recognition processes. This may imply that current static empirical research designs might not be as reliable as previously assumed and that the results may be (partly) implausible.

Research questions and hypotheses

Research questions

Although research into language processing has progressed immensely over the past decades, there are still some insufficiently researched themes. The mental lexicon seems to be constantly changing, depending on varying activation levels which are influenced by different times and places. Especially frequent and recent activation have been mostly investigated in stable, short-term settings. The present study aims to investigate the instability of the mental lexicon by focusing on the effects of time on the processing speed of time-specific lexical items in a lexical decision paradigm. The timescale explored in the current research is a timescale at a level of months. Additionally, this study aims to find evidence for a relative recency effect at this timescale. The time-specific lexical items chosen for the current investigation were words related to the seasons winter and spring and the two testing moments were logically also during the winter and spring. The current study aims to answer the following research questions:

Can an instability of the mental lexicon also be detected when examining the L1 and L2 reaction times to time-specific lexical items during two different seasons?

1. Do the reaction times to time-specific lexical items differ from the reaction times to neutral baseline words?

(17)

a. Do the reaction times to the L1 and the L2 setting show similar effects of instability over a time frame of several months?

3. In terms of a main effect of time and interactions between the variables, do the reaction times to time-specific lexical items change throughout a time frame of several months?

Hypotheses

Previous research has implicitly or explicitly highlighted that the activation levels of items in our mental lexicon may fluctuate over time (De Bot & Lowie, 2010; Dijkstra et al., 2018). Especially frequency and recency of occurrence and language use have been documented to affect activation levels. Therefore, it would be safe to assume that the processes involved in word recognition may be affected by time as well. To illustrate, winter vocabulary is much more triggered during the winter than it is during the warmer months, as this certain type of vocabulary is not used as abundantly throughout the rest of the year. Similarly, words pertaining to spring should be recognized faster during the spring, due to risen activation levels. These different topics of vocabulary are more recently and frequently activated during certain seasons compared to others, which may result in faster responses. Moreover, the native language should, overall, be accompanied with higher levels of activation and higher resting levels of activation, and thus quicker latencies, as this language is usually more frequently (and often also more recently) used than the second language. In terms of variability, the latencies of the L1 can, on the one hand, be expected to be influenced more by the variable Season than the latencies of the L2, as greater differences in activation levels may occur (Pfänder & Behrens, 2016). This, however, might depend on the level of L2 proficiency as well as L2 language use, as highly proficient learners who use the L2 often, may also be subject to risen activation levels. On the other hand, the L1 has been found to be more stable than the L2, due to automatization (Plat et al., 2018) or high resting levels of activation (Dijkstra et al., 2018) and may therefore be less sensitive to the variable Season. Nevertheless, the interactions between Season and the lexical categories ‘spring’ and ‘winter’ are expected to be significant for both languages, as season-related vocabulary should be activated more quickly during the corresponding season, since frequent and recent activation are temporarily more pronounced. All in all, the expected outcomes and answers to the research questions above can be formulated as follows:

Overall, an instability of the mental lexicon can be detected when examining the L1 and L2 reaction times to time-specific lexical items during two different seasons.

1. The neutral baseline words should be activated rather quickly, as they are activated on a daily basis throughout the year. The lexical items related to the current season, should be activated quickly as well. However, the lexical items related to the season that has passed or that is still to come, should have slower responses.

(18)

2. Overall, the reaction times to the L1 are expected to be faster than the reaction times to the L2, as L1 processing is supposed to be more automatized than L2 processing.

a. Consistent with previous research, the L1 is expected to be more stable and show less variability between the two testing moments. The L2 should show more variability on a timescale at the level of months, because this language is more unstable.

3. The main effect of time for all lexical items and all languages together is expected to be absent, as the RTs to neutral baseline words should not change during the two seasons, winter words are slower during the spring, but spring words are faster during the spring. The reaction times to time-specific lexical items are expected to change over the course of several months depending on the season in which they are tested. Both languages should show significant effects of Season when examining the time-specific lexical categories separately.

Method

In order to investigate whether time influences the mental lexicon and, more specifically, the processing speed of time-specific lexical items, a lexical decision task was to be performed twice: once during the winter season and once during the spring season. The recruited participants were divided into two groups, with one group performing the lexical decision task during the winter and during the spring in the L1, and the other group performing the task twice in the L2. In sum, the same two groups of participants were tested at two moments in time in either one of the language settings. The independent variables that received attention were season of testing (i.e. Season) with the two levels winter and spring, and the language of the experiment (i.e. Language) with the two levels Dutch and English. The dependent variable was the mean reaction times in milliseconds. The lexical decision task contained strings of letters that either formed real words or pseudowords which participants were asked to categorize as quickly as possible, yet focusing to some extent on accuracy as well.

Participants

The sample group recruited for the current investigation consisted of 32 Dutch university students who are studying English as a Bachelor’s degree. As the participants are all studying English on university level, they can be considered advanced English as a second language (ESL) learners who use English (academically) on a daily basis. Most of them were female (F = 24, M = 8) and ages ranged from 18 to 23 (mean = 20.4). None of the participants had experienced any vision problems, hearing impairments, language disabilities, or learning disabilities. All participants can be deemed sequential bilinguals or multilinguals as they had acquired Dutch from birth as their first language, and English at a later stage as their second language. Before the experiment, students received a form in which they could give consent for using their data2_{. After}

(19)

signing, the participants were randomly divided into two groups: one group would continue doing the entire experiment in English, and the other would proceed in Dutch.

Materials

The lexical decision tasks were key to the investigation. Two versions of the same lexical decision task were created: an English and a Dutch version. The experiments were developed with the software OpenSesame (Mathôt, Schreij, & Theeuwes, 2012). In order to convert the offline OpenSesame studies into online versions, the program JATOS (version 3.3.4) was utilized (Lange, Kühn, & Filevich, 2015). In addition, JATOS provided personal links to the experiments which would enable online distribution among the participants. After testing the experiment in multiple web browsers, the lexical decision tasks proved not to be compatible with Microsoft Edge. Therefore, participants were requested to perform the experiment on a computer or laptop with Google Chrome or Safari. The results of their efforts were downloaded from JATOS and uploaded to Excel. As all participants were to perform the lexical decision task twice, Excel was also used to monitor which links the participants had received to avoid confusion as to which results belonged to which participant.

The design of the two versions of the experiment is similar: the two versions are identical except for the languages in which the instructions and the lexical items were presented. Within the experiment, a practice round precedes the actual test in order to get used to the experiment set-up. The experiment contained 180 strings of letters forming either existing words or non-existing words, hereafter nonwords or pseudowords3_{. As has been illustrated by Figure 2, before}

each of the letter strings, a 750-millisecond display of a fixation mark emerged. Each of the strings of letters remained on the screen until the participants had pressed a key, but with a maximum of 2000 milliseconds in order to avoid participants taking abnormally long. The appropriate keys were ‘z’ for an existing word and ‘m’ for a nonword. The fixation dot and the lexical items were displayed in a black, size 28, Mono font on a grey background.

Figure 2. Illustration of the input answer timeline of the lexical decision task.

The lexical items selected for experimental condition were to be related to time-specific themes. As the main focus of the current investigation is on seasonal influences on the mental

(20)

lexicon, the categories deemed most suitable were: winter and spring. Another category that was included was a neutral baseline word category which served as a control condition. Naturally, pseudowords were also included in the lexical decision task. During the lexical decision task, a total of 180 strings of letters sequentially emerged on the participants’ screen. Half the letter-strings selected were actual words and the others were nonwords. The nonwords were all pronounceable and differed only one letter from the existing baseword. The 90 existing words were subdivided into the aforementioned categories (winter, spring, and neutral), with each category containing 30 lexical items. The general word frequencies of the selected English items were adopted from the Corpus of Contemporary American English (COCA) and the Dutch items from the Corpus Hedendaags Nederlands (CHN). Table 1 shows the log frequencies of the words selected for each of the categories and it can be assumed that all items are relatively frequent in both the L1 and the L2. Word length varied between 3 to 8 letter words (mean = 4.99) with 1 or 2 syllables (mean = 1.43) as, unlike the BIA+ model which allowed only 4- and 5-letter word recognition, Dijkstra and his colleagues (2018) allowed words within this 3- to 8-letter range in their Multilink model. Moreover, to confine the selection of lexical items was deemed unnecessary, as the reaction times to the chosen words on one moment in time are not compared to each other, but rather compared to the reaction times to the exact same words, yet on a different testing moment. Thus, the relevant conditions in the current experiment were all within-subject conditions. Although the Dutch and English profiles are not identical, they are considerably similar, so differences between the word categories are not likely to affect the results.

Table 1. The log-10 mean frequencies, standard deviations and the number of words that are not in the top 5000

frequency rank for Dutch and for English. Lexical categories are separated. N is the number of lexical items.

Language Category Mean log frequency

SD N No. of words not in top 5000 Dutch Winter 3.86 0.51 30 13 Spring 4.05 0.54 30 8 Neutral 4.37 0.49 30 2 English Winter 3.83 0.61 30 14 Spring 4.07 0.57 30 7 Neutral 4.39 0.50 30 3

Procedure

The respondents received an e-mail containing information about the experiment they were about to participate in. Attached to this e-mail they could find the document in which they could give their consent. They were to return the signed document before they could receive a link to the experiment. The information provided to the participants included the broad topic of the present study and the procedure. The participants were notified that the experiment consisted of two parts: a lexical decision task once during the winter and once during the spring. The online lexical decision tasks would each take approximately five to ten minutes.

(21)

What was not shared with the participants was the exact area of interest of the current study. The experiment was developed specifically to be able to detect possible differences between the recognition speed of season-related words during two different meteorological seasons: winter and spring. Participants were neither informed that the words of interest were season-related nor that the two testing moments took place in two different seasons. This information was purposely omitted in order to avoid any imaginable influences on the results.

After having received the consent forms, participants were randomly divided into either the English or the Dutch group. Their next e-mail containing a link to the lexical decision task was sent during the winter. Whether they received the English or Dutch version of these two tasks was dependent on in which group they were placed. The results of the lexical decision task emerged in JATOS immediately after completion. During the spring, the participants received the final part of the experiment, which was another e-mail containing a different link to a similar experiment. After completion, the participants were thanked for their time and effort.

Analyses

The results of the first and the second testing moment were all stored in JATOS version 3.3.4 (Lange et al., 2015), but needed to be downloaded and converted into an Excel file in order to increase the accessibility of the otherwise unclear data. The results from JATOS were downloaded as separate text (.txt) files which, in turn, could be imported into Excel. Before performing the statistical analysis, the dataset was cleaned by removing erroneous responses (4.2% of the data) and responses with reaction times deviating more than three times the standard deviation from the mean (0.12% of the data). These particularly fast or slow responses may have been caused by participants temporarily being distracted or by participants accidentally pressing a key too early and should therefore be disregarded. The statistical analyses applied were t-tests and repeated measures ANOVAs and they were performed on solely the correct responses by using IBM SPSS software (version 26 for Windows). Both the main effects and interaction effects of the variables were calculated.

Results

The assumptions for a t-test and repeated measures ANOVA were met, so these statistical tests were used to analyze the reaction times to time-specific lexical items on two different testing moments two months apart. On the next page, Table 2 presents the mean scores (M) and the standard deviations (SD) per group on the two different testing moments and the total average reaction times. The next section describes all statistical outcomes for: 1) the overall performance; 2) the winter-related words solely; 3) the spring-related words solely; 4) the neutral word category solely; and 5) the comparison between the neutral lexical items and the season-related words.

(22)

Table 2. Mean RTs and SDs for the L1 and L2 group during the winter and during the spring and the total average scores.

N is the number of participants.

Word category → All categories combined

Season ↓ Group Mean SD N

Winter L1 (Dutch) 696 90 16 Spring L1 (Dutch) 702 89 16 TOTAL 699 86 16 Winter L2 (English) 678 55 16 Spring L2 (English) 661 43 16 TOTAL 670 46 16

First, the reaction times to all the lexical items taken together for both the L1 and L2 group together were analyzed, in order to determine if there were any differences between the first testing moment (i.e. winter) and the second testing moment (i.e. spring). The mean reaction times from the two testing moments were relatively similar. A paired samples t-test confirmed this observation and revealed that the mean latencies of all words and all languages together did not differ significantly between the first testing moment (M = 687, SD = 74) and second testing moment (M = 681, SD = 72); t(31) = 0.75, p > 0.05. Figure 3 shows a boxplot of all words together and all languages together on the two testing moments. Thus, the main effect of Time (i.e. Season) on the overall reaction times was not significant.

(23)

Figure 4 illustrates the mean reaction times during the two testing moments for the L1 and the L2 group separately. It can be seen that for the L1 group the reaction times during the second testing moment were slightly slower than during the first testing moment, whereas the L2 group performed somewhat faster during the second testing moment compared to the first one. When the two groups were compared to each other, the mean results suggest that there was a small difference between the overall reaction times of the groups, as the L2 group showed slightly quicker latencies than the L1 group. Although the overall mean reaction times of the L2 group were faster than the means of the L1 group, an independent samples t-test exposed that the L2 group (M = 670, SD = 46) did not perform significantly faster than the L1 group (M = 699, SD = 86), as t(30) = 1.19, p > 0.05. A repeated measures ANOVA with Season (i.e. Time) and Language as variables was used to establish whether there was an interaction between Season and Language. The analysis revealed that the overall interaction between Season and Language was also not significant, F(1, 30) = 2.38, p > 0.05, but it had a medium effect size (ηp² = 0.07). After analyzing the results separately per group, the outcomes of the effects of Season remained not significant: the L1 group showed F(1, 15) = 0.25, p > 0.05 and the L2 group showed F(1, 15) = 3.41, p > 0.05. The effect size for the L1 group was small (ηp² = 0.02), but large for the L2 group (ηp² = 0.19).

(24)

Table 3. Mean RTs and SDs to winter-related words of the L1 and L2 group in two seasons. N is the number of participants.

Word category → Winter

Winter L1 (Dutch) 668 85 16

Spring L1 (Dutch) 707 104 16

Winter L2 (English) 666 48 16

Spring L2 (English) 729 95 16

Figure 5. Plot with the mean RTs to the words related to winter for the L1 and L2 group in the two seasons.

Furthermore, the reaction times to the different lexical categories (winter, spring, and neutral lexical items) were also investigated separately. The results from the three categories will be reported sequentially, starting with the lexical items related to winter. Note that, for convenience, a summary of the most important results of all three categories will be presented later in this section in Table 6, but for now the results will be reported category by category. For the lexical items related to winter it can be perceived from the data in Table 3 that, overall, these lexical items had quicker latencies during the winter than during the spring. Figure 5 illustrates these results. A repeated measures ANOVA showed a significant effect of Time (i.e. Season) on winter words, F(1, 30) = 15.45, p < 0.05, with a large effect size of 0.34. Analyses of the reaction times to winter-related words for the groups individually revealed significant results for both: for the L1 group, F(1, 15) = 4.34, p < 0.05, ηp² = 0.22, and for the L2 group, F(1, 15) = 12.20, p < 0.01, ηp² = 0.45. This means that both groups recognized winter-related words significantly faster during the winter than during the spring with large effect sizes. The interaction between Season and Language for winter words was not significant, F(1, 30) = 0.90, p > 0.05 and it had a small effect size (ηp² = 0.03).

(25)

Table 4. Mean RTs and SDs to spring-related words of the L1 and L2 group in two seasons. N is the number of participants.

Word category → Spring

Winter L1 (Dutch) 677 110 16

Spring L1 (Dutch) 673 119 16

Figure 6. Plot with the mean RTs to the words related to spring for the L1 and L2 group in the two seasons.

An overview of the mean reaction times and standard deviations to the spring-related words are presented in Table 4. Closer inspection of the table shows that the spring-related words were on average recognized faster during the spring than during the winter. These results have been illustrated by Figure 6 for both groups separately. Further analysis indicated that for the lexical category ‘spring’, the effect of time for the groups combined was not significant, F(1, 30) = 2.58, p > 0.05, with a medium effect size of 0.08. An analysis of the effect of Season on solely the L1 group showed that the reaction times to spring-related lexical items are unaffected, as a repeated measures ANOVA revealed insignificant results, F(1, 15) = 0.03, p > 0.05, with a small effect size of 0.002. However, participants who completed the experiment in the L2 showed significantly faster reaction times to spring-related words during the spring compared to the reaction times during the winter, F(1, 15) = 6.10, p < 0.05. The calculated effect size for this group was large (ηp² = 0.29). In sum, the L2 group responded significantly faster to spring-related words during the spring than during the winter, whereas the L1 group did not show significantly faster latencies. The interaction between Season and Language for this word category was found not to be significant, F(1, 30) = 1.79, p > 0.05, but with a medium effect size of 0.06.

(26)

Table 5. Mean RTs and SDs to neutral words of the L1 and L2 group in two seasons. N is the number of participants.

Word category → Neutral

Winter L1 (Dutch) 643 71 16

Spring L1 (Dutch) 659 98 16

Figure 7. Mean RTs to the neutral words for the L1 and L2 group in the two seasons.

Moreover, the lexical items that were not related to any specific season, so the words within the lexical category ‘neutral’, were examined as a control condition. The mean reaction times and standard deviations are summarized in Table 5. In order to assess whether there are significant differences between the two testing moments, a repeated measures ANOVA was again used. No significant effect was found for the groups evaluated together, F(1, 30) = 0.79, p > 0.05, and a small effect size of 0.03 was found. A closer look at the groups separately showed insignificant effects for both groups as well: the L1 group, F(1, 15) = 0.52, p > 0.05 and the L2 group, F(1, 15) = 0.31, p > 0.05. The effect sizes for the L1 and the L2 group were both small with scores of 0.03 and 0.02 respectively. The results of solely the neutral baseline words highlighted that participants of both the L1 and the L2 group performed approximately the same in both testing moments, which – although unconventional – has been illustrated by Figure 7. There was also no statistically significant interaction reported between Season and Language within the lexical category ‘neutral’, F(1, 30) = 0.005, p > 0.05, with a small effect size of 0.005. Thus far, the results for all categories have been addressed. As mentioned earlier, a summary of the most important reported p-values is presented on the next page in Table 6, yet with their exact values.

(27)

Table 6. The exact p-values for winter-related, spring-related, and neutral words for the L1, the L2 group, and the combined

total. A green font indicates the outcome is hypothesized and a red font means the outcome is unexpected.

Word category → Winter Spring Neutral

Group ↓ p-values

L1 (Dutch) 0.048* 0.869 0.483

L2 (English) 0.003* 0.026* 0.586

Total 0.000* 0.118 0.381

* = significant.

Green = expected, Red = unexpected.

Lastly, the reaction times to the neutral words were compared to the reaction times to the season-related lexical items. On the one hand, the overall scores of the lexical category ‘neutral’ were significantly faster than the reaction times to the lexical items which were unrelated to the season of testing (i.e. winter-related words during the spring and spring-related words during the winter), F(1, 30) = 25.73, p < 0.01, and showed a large effect size with ηp² = 0.46. To clarify, the mean reaction times to the neutral words were much faster than the mean reaction times to the winter-related lexical items in the spring and the spring-related words in the winter. Even when analyzing the means of the two groups separately, both groups revealed significant values: for the L1 group, F(1, 15) = 5.95, p < 0.05, with a large effect size of 0.28 and for the L2 group, F(1, 15) = 26.92, p < 0.01, with a large effect size of 0.64. On the other hand, the overall latencies of the neutral baseline words did not differ significantly from the reaction times to the words related to the season they were tested in, F(1, 30) = 4.16, p > 0.05, with a medium effect size (ηp² = 0.12). This is to say, the average scores of the neutral baseline words were similar to the scores of winter-related words during the winter and spring-winter-related words during the spring. The results of both groups separately also uncovered insignificant results: for the L1 group, F(1, 15) = 1.90, p > 0.05 with a medium effect size (ηp² = 0.11) and for the L2 group, F(1, 15) = 2.99, p > 0.05 with a large effect size (ηp² = 0.17).

Discussion

In order to answer the main research question “Can an instability of the mental lexicon also

be detected when examining the L1 and L2 reaction times to time-specific lexical items during two different seasons”, the sub-questions require some attention first. Each sub-question and its

hypotheses will be repeated before the results of the lexical decision tasks are analyzed and discussed.

1. Do the reaction times to time-specific lexical items differ from the reaction times to neutral baseline words?

The hypotheses for this research question were: 1) The neutral baseline words are expected to be activated rather quickly, because they are high frequency words and are therefore highly activated throughout the year; 2) The lexical items related to the season in which the experiment took place should be activated quickly as well; and 3) The lexical items related to the