MA Thesis
Master’s in General Linguistics
University of Amsterdam
June 2018
33
1
Abstract ... 3
1. Introduction ... 4
2. Theoretical background ... 5
2.1 Embedding words in long-term memory: the formation of lexical units ... 5
2.2 Filtering and anchoring new memories: the role of long-term memory ... 6
2.3 Words’ level of activation ... 7
2.4 The spreading nature of levels of activation ... 7
2.5 Keeping words in long-term memory: reinforcing the encoding of words ... 9
2.5.1 The interaction effect... 9
2.5.2 The multimedia effect ... 9
2.5.3 The propositional format of memories ... 10
2.5.4 The testing effect ... 10
2.5.5 Depth of processing ... 10
2.5.6 Associations ... 11
2.5.7 Frequency of presentation ... 11
2.6 Connections and the statistical understanding of cues ... 11
2.6.1 Available, reliable, valid: how do cues need to be? ... 11
2.6.2 The flexible understanding of language users ... 12
2.6.3 The conservatism of language users ... 13
2.7 The current study ... 14
2.7.1 Theoretical background ... 14
2.7.2 Research question and expectations ... 17
2.7.3 Design of the study ... 17
3. Methodology ... 19
3.1 Participants ... 19
3.2 Training conditions ... 19
3.3 Materials ... 19
3.3.1 Training session materials ... 19
3.3.2 Test task materials ... 21
3.4 General procedure ... 21
2
3.6 Training session ... 22
3.6.1 Features common to both experimental conditions ... 22
3.6.2 Features of the RoC condition ... 22
3.6.3 Features of the PoC condition ... 22
3.7 Test task ... 23
3.7.1 Design ... 23
3.7.2 Scoring ... 23
4. Results ... 24
4.1 Speed response data (RTs) ... 24
4.2 Accuracy data ... 25
4.3 Summary of results ... 26
5. Discussion ... 27
5.1 Conclusion ... 27
5.2 Limitations of the study ... 28
6. References ... 29
7. Appendixes ... 32
Appendix I: Stimuli – training session ... 32
3
Abstract
Research agrees on the fact that the encoding of novel vocabulary in long-term memory is facilitated by the establishment of connections between target words and known words (e. g. Hilpert and Diessel, 2017; Langacker, 2017; MacWhinney, 2001; Regier, 2005). Part of the literature addresses the topic of possible types of connections between words (e.g., Hilpert and Diessel, 2017). The statistical ability of language users to detect connections is also addressed in the literature. However, no statement is made about the role of the number of connections. Departing from the concept of lexical units of entrenchment theories (e.g. Goldberg, 2003; Schmid, 2017) and from Anderson’s theory on spreading activation (2010), this study explored the relationship between the size of lexical units in the mental lexicon and the ability to recall the words residing in them in the context of a second language. In a psycholinguistic experiment, participants learnt novel vocabulary. The two training conditions differed in the size of the lexical units created during the training session. Ability to recall target words was tested. Contrary to expectations, results showed no significant advantage in favour of either experi-mental condition.
4
1. Introduction
The factors facilitating the integration of novel vocabulary in long-term memory for a second language have been widely debated (e. g., Anderson, Bjork, and Bjork, 2000; Craik and Lockhart, 1972; Tokowicz and MacWhinney, 2005; Hilpert and Diessel, 2017; Langacker, 2017; Schmid, 2017). One point on which research broadly agrees is that the connections between novel words and other words (and in particular to words residing in long-term memory) are crucial for the integration of novel words in the long-term memory system. However, the very definition of connection remains controversial, as there is no consensus on which specific properties of the connections yield the most significant results in integrating novel words in the long-term memory system. Consequently, it is not clear which condi-tions yield the best results for the recall of target words.
Some accounts (e. g., Mahon, Costa, Peterson, Vargas and Caramazza, 2007; Starreveld and La Heij, 1996; Ellis, 2001) postulate the importance of the nature of the connections that novel vocabulary establishes. Other accounts (e. g., Langacker, 2017; Goldberg, 2003), on the other hand, attribute a crucial role to the simple belonging to a lexical unit (i.e., a unit of words in long-term memory, which are connected to one another), irrespective of its specific characteristics and its size.
The study presented here aims at comparing two different conditions of novel vocabulary learning in the context of second language learning. These conditions differ only with respect to the number of connections established during the training session. The overall goal of the current study is to contrib-ute to the understanding of the mechanisms through which second language vocabulary is integrated in the memory system. More specifically, the current study sets out to gather evidence on the influ-ence that the number of connections of novel words to long-term memory words have on the anchor-ing of novel vocabulary in memory. In the followanchor-ing I am goanchor-ing to assume that the number of connec-tions of a novel word to words in long-term memory relates to the size of the lexical unit to which it belongs in the mental lexicon.
In chapter 2 I am going to give an account of the different views on vocabulary learning and I am going to discuss the theoretical issues which motivated the research question at the basis of the current study. In chapter 3 I am going to describe the methodology used for the current study, which consist-ed in a psycholinguistic experiment for the acquisition of novel vocabulary. In chapter 4 I am going to present the findings of the experiment in relation to accuracy and rapidity data. In chapter 5 I will interpret the results in light of the theoretical frame discussed in chapter 2. In chapter 6 I am going to address the limitations of the study.
5
2. Theoretical background
In the present chapter I am going to present the theoretical frame that constitute the departing point of the current study. The topics that I am going to discuss are the formation of lexical memories (sec-tion 2.1), the role of long-term memory for vocabulary learning (sec(sec-tion 2.2), the strength of encoding of words (section 2.3) and the behaviour of words in the mental lexicon as regards reciprocal influence (section 2.4). Furthermore, I am going to address the topic of explicit manipulation of strength of en-coding (section 2.5). In section 2.6 I am going to discuss the statistical properties of the input that impact on the ability of vocabulary learning. In section 2.7 I am going to explain the positioning of the current study in this theoretical frame and to outline the research question that has generated out of this theoretical frame.
2.1 Embedding words in long-term memory: the formation of lexical units
Construction grammar defines language knowledge as a network of symbolic units (Croft, 2001) or, in the words of Goldberg, as a large “network of constructions” (Goldberg, 2003, p. 219). The learning mechanism through which language knowledge is attained and reinforced is “entrenchment”, a con-stant cognitive reorganizational process which operates by varying the strength of linguistic represen-tations (Schmid, 2017). In the specific case of vocabulary, the acquistion of linguistic knowledge is synonymous with the integration of novel vocabulary into the existing mental lexicon – that is, in long-term memory. Indeed, as long as novel vocabulary is not integrated in long-long-term memory, it is not able to interact with its lexical neighbours. Integration is achieved by means of the formation of connec-tions. Connections can link a linguistic form and its corresponding meaning: in the words of Hilpert and Diessel (2017, p. 58), they are a “repository of form-meaning pairings”. Furthermore, they can also link form-meaning pairings to other form-meaning parings. The result of the formation of these connections are lexical units. Once words are part of a unit, they have a potential for recurring in this same pattern to the extent that they can be described as an “event waiting to happen” (Langacker, 2017, p. 39). Reinforcing linguistic knowledge consists in the continuous adaptation of units according to their frequency of activation. In sum, we can say that the acquisition of vocabulary consists in the formation of connections between already known words and novel words and in the subsequent for-mation of lexical units, and that the more the connections between words, the bigger the correspond-ing lexical unit.
Another contribution to the topic of units is MacWhinney’s “unified model” (2008). In the unified model lexical items are represented as form-function associations, which are in turn connected to each other and form units. MacWhinney argues that connections between lexical items are the prod-uct of the statistical analysis condprod-ucted by the brain on the input. This process is automatic and implic-it; MacWhinney states that in spite of this, connections can also be established explicitly. He describes this phenomenon as the creation of “resonance” between lexical items (MacWhinney, 2008, p. 367). Resonance can be understood as the trace that words leave in long-term memory and as the subse-quent activation of target items (which is stronger after each repetition trial). MacWhinney stresses the effectiveness of learning vocabulary by establishing connections. Connections can involve the phonetic, lexical or syntactic level. In order to exemplify the process of explicit connection formation, he cites the example of the German word Wasser (water), which on the phonological level can be associated to the sound of water itself. On the lexical level, it can be contextualised in the sentence Wasser trinken (to drink water) and can therefore be semantically associated with the action of drink-ing. This semantic association has the advantage of activating other semantically related words, for example Bier (Bier trinken) (to drink beer), hence reinforcing the association link. In MacWhinney’s
6 words: “[…] these resonant associations between collocations, sounds and other words help to link the German word Wasser into the developing German lexicon.” (MacWhinney, 2008, p. 361). The current study was designed in order to create such lexical units. This was the aim of the training phase. In the test task the influence of their size for the recall of new vocabulary was tested.
2.2 Filtering and anchoring new memories: the role of long-term memory
As we have seen in the previous section, learning new vocabulary is possible thanks to the integration process of the novel target words in long-term memory. Long-term memory does not only act as an anchor for novel vocabulary, but it also has a paramount role in the understanding of new vocabulary. Ellis explained this phenomenon by saying that “perception [...] is automatically filtered and patterned by our existing long-term memory schema […]” (Ellis, 2001, p. 35). Elements in short-term memory, that is, novel vocabulary, are analysed by comparing them to lexical items stored in long-term memory and finding analogies and differences with them. Gass and Selinker (2008) also attribute a selective function to long-term memory. They argue that the first step towards intake is ‘apperception’, which is the process whereby the learner makes sense of the input by using her previous knowledge. Apper-ception can be described as the recognition of the specific part of the input (or of some specific prop-erties of it) which are meaningful and can be meaningfully sorted in long-term memory. According to Gass and Selinker, long-term memory elements that facilitate apperception can be of various nature, from phonological knowledge to previous lexical knowledge in the second language (from now on: L2) or in the first-language (from now on: L1), from world knowledge to language universals. Appercep-tion also explains how new informaAppercep-tion can be integrated in our knowledge system. We can conclude that only new information which resonates with previous knowledge (that is, information that can be understood by means of using the already encoded one) can be integrated in the linguistic system. This could be seen as a kind of limitation: real “knowledge jumps” seem not to be a viable option. However, this feature of knowledge acquisition is precisely the one that guarantees that information is retained better. This is another piece of evidence suggesting that connections between novel words and words residing in long-term memory are of vital importance.
In an experiment in which they compared memory for faces and for snow crystals, Goldstein and Chance (1970) found that recollection of faces was significantly better than that for snow crystals (74% correct recalls for faces and 30% for snow crystals). The results cannot be explained by using perceptual properties of the stimuli, since the snow crystals were easily distinguishable from each other, as were faces. However, the results were explained by arguing that memory for targets reflects how well the material can be encoded. Successful encoding requires the attribution of some sort of meaning to the input. Since stored knowledge about human faces is vast and much more diversified than that on geometrical forms, the encoding of new faces was more successful than the one of new geometrical forms. Evidence in favour of this argument comes also from the study by Tokowicz and MacWhinney (2005). This study shows that subjects in very initial stages of second language acquisi-tion (from now on: SLA) were sensitive to violaacquisi-tions of L2 grammar, therefore demonstrating that implicit learning was taking place. Significantly, sensitivity to violation was limited to features of the input which were relevant in the L1 of subjects (violations in tense-marking were perceived as such, since L1 and L2 behaved similarly in this respect; violations in the use of determiner number agree-ment, on the other hand, were mostly not perceived as such since they were treated differently in L2 and L1). The described behaviour was assessed for L2 grammar learning: this demonstrates that pre-vious knowledge is generally crucial in the understanding of new input and not only for vocabulary learning.
Further evidence in favour of the role of long-term memory for the encoding process of novel linguis-tic knowledge comes from a study by Majerus and D’Argembeau (2011). They tested memory recall for lists of words. Two variables were manipulated in the training session: item recollection and order recollection. Their finding is that item recollection is strongly influenced by previous semantic
7 knowledge, whilst order recollection is not. So, again, it seems that knowledge stored in long-term memory provides a basis which serves the storing of novel words.
Departing from these considerations on the role of long-term memory, I have designed the current study so that novel words were presented alongside known words. Novel words were apperceived on a phonological level (the phonemes that constituted them were known). Furthermore, they were presented as being visually associated with long-term words. This associated presentation was aimed at developing new lexical units containing both novel and long-term vocabulary.
2.3 Words’ level of activation
In this section I am going to analyse a specific characteristic of words, that is, their level of activation in long-term memory.
As every L2-learner and L1-speaker knows, some words are easier to recall than others. The reason why we experience ease of retrieval is that some words are more strongly encoded in long-term memory than others. This makes them more available to recall, hence reducing the effort required to retrieve them. These are the findings of Anderson and colleagues (M.C. Anderson et al., 2000), who observed that the reaction time (from now on: RT) necessary for an action to occur upon presentation of a specific target word (producing a word, carrying out a judgement task or a translation tasks only to mention a few examples) is shorter for some words than for others. On the basis of their data, the researchers formulated the hypothesis that words have different degrees of availability, or, in their terminology, different levels of activation. A “level of activation” is the amount of input necessary to retrieve a target word and make it available for further tasks. The smaller the input necessary to achieve this, the shorter the related RTs when the learner is presented with the input that will allow a particular action to occur. RTs are considered to be a reliable measure of words’ level of activation, since words with higher levels of activation have a faster access to consciousness and can therefore be used more promptly.
Importantly, levels of activation are not invariable in time, but can manipulated. Anderson and col-leagues (Anderson, Bjork and Bjork, 2000) mention a condition in which the activation level of a word can even decreased. This is the case when a certain word has to be suppressed. This is also the case if a word has been activated in the same moment as another word but does not constitute the required response, e. g., naming an object. This suppression process takes place unconsciously and its aim is to clear the way for the right word to surface to consciousness. As is argued by Anderson and colleagues, this will happen by means of suppression of the non-target words, which will subsequently have a lower degree of activation as a result of the suppression mechanism. Decrease of activation does not play a role in the current study. However, this phenomenon provides further evidence in favour of the concept of interconnection of words in the mental lexicon.
Higher levels of activation can be attained by actively manipulating them. For example, they can be increased by means of additional presentation trials of the target words. This is a direct technique. However, Anderson and colleagues (Anderson, Bjork and Bjork, 2000) made the observation that it is also possible to achieve this in an indirect way. In the next section I am going to illustrate the mecha-nism that makes indirect influence possible, namely, the spreading of words’ levels of activation.
2.4 The spreading nature of levels of activation
The level of activation of a given word increases whenever the word is directly accessed, be it actively (e. g. through retrieval) or passively (e. g. through presentation), i. e., when the word is encountered in the input. However, it is possible to increase words’ levels of activation indirectly, that is, by acting on
8 one or more of the words to which the target word is connected in long-term memory. This is possible due to the fact that word activation has a spreading nature, as Collins and Loftus (1975) and Anderson (1983) hypothesised. This means that, once a word of the mental lexicon is accessed, the words con-nected to it are also going to be accessed. Evidence in favour of this mechanism is consists in smaller RTs for target words, if the words connected to them have been accessed previous to the test. As we have already seen, access means also a subsequent greater level of activation. For this reason, it is possible to argue that activation spreads to the elements which form part of a common lexical unit, i. e., to elements which are connected to each other. If one word is directly accessed (for example, by means of written presentation), another word belonging to the same lexical unit is also going to be activated and therefore receive some input, albeit in a smaller quantity. This is due to the fact that the quantity of the input decreases as it spreads. Upon direct access to the target word, its subsequent increased level of activation in long-term memory can be observed. Encountering the word cat, for example, will increase its level of activation by a certain degree. If the word dog is accessed, the level of activation of the word cat will also be increased, since the two words are semantically connected to each other. This will however be the case to a smaller extent. According to the same principle, the word barking will also increase the activation level of cat, but to an even smaller extent. The increase in levels of activation can be assessed by measuring the RTs necessary to complete an action.
Activation spreads in all directions in which connections between words exist. This implies, for exam-ple, that it takes places also across languages: if an L2 word defines a concept or category which exists also in the L1, it will automatically have a higher level of activation than L2 words which define con-cepts that do not exist in the L1 or which alternatively bring categories together which in the L1 are expressed by means of two different words. Spreading activation, however, is present not only at the semantic level: phonological similarity also determines connections between words and therefore spreading activation at this level, too. Phonology-based connections also work across languages, so that cognate words in an L2 can receive secondary activation through the activation of a similar L1 word. Cognates receive activation also on a semantic level and it has been shown that they receive a greater activation as compared to non-cognate translations (De Groot and Keijzer, 2000).
These observations brought Collins and Loftus (1975) to formulate two hypotheses which describe the nature and functioning of long-term memory: a) long-term memory is not a static system and b) words residing in long-term memory do not only react to external influence, but also influence each other internally. This mechanism has been explained by pointing out that when a certain concept is used, related concepts are more likely to be accessed than semantically unrelated ones.
Indirect activation can be elicited in different modalities (for example: written text, images represent-ing words and orally spoken words). Activation can be inter-modal or across-modal. In their work on interference and facilitation effects in the picture-word paradigm, Mahon and colleagues (Mahon, Costa, Peterson, Vargas, and Caramazza, 2007) and Starreveld and La Heij (1996) found that the presentation of pictures related to target words facilitates their production.
As illustrated in the first section of this chapter, constructionist approaches describe language knowledge as being organised in units. These units have a crucial function in anchoring words effec-tively in long-term memory. Moreover, the theory of spreading levels of activation described in this section, is crucial to the design of this study. This is because I have assumed spreading activation to be the mechanism which allows that words belonging to lexical units of different size also have different levels of activation. This is the case because words of one lexical unit are connected to each other and are therefore anchored in the mental lexicon. If we assume spreading activation to be at work in lexi-cal units, we can hypothesise that every time that a word of the unit is accessed, the others are, too. Let us assume that all words belonging to two lexical units have the same frequency in the input. This means they each of them is activated as often as all the others. Let us also assume that these two lexical units are not equally big. In the bigger lexical unit, words will be activated more often than in the smaller one, since they will receive indirect activation for every other word of the unit that is di-rectly activated.
9
2.5 Keeping words in long-term memory: reinforcing the encoding of words
As we have seen in the previous sections, in order to be encoded in long-term memory, novel vocabu-lary has to be connected to known words. Furthermore, we have seen that each word has its own specific level of activation. Activation levels are not fixed in time and can be purposefully manipulated. In what follows I am going to present an overview of specific techniques which have shown to be ef-fective in order to enhance words’ level of activation and therefore in order to make words more read-ily available for retrieval. I am going to argue that all these techniques have a common denominator that is responsible for their success as vocabulary learning techniques. I am going to argue that they all facilitate the emersion of new connections between target words and other words in long-term memory and that, ultimately, they cause the enlargement of the lexical units to which the target words belong.
In the following sections I am going to provide an overview of these techniques. For each I am going to interpret the mechanism responsible for enhancing words’ level of activation in the light of the size of lexical units. The last technique mentioned in this section is frequency of presentation. Frequency of presentation is different from the other techniques: it consists in repeated presentations of the target word and does not produce the enlargement of the target lexical unit. Indeed, per se it does not yield the formation of new connections, but only the reinforcement of the already existing ones.
I am going to present each technique separately and to analyse it in the light of the number of con-nections.
2.5.1 The interaction effect
Establishing connections between words is beneficial to strength of encoding. This applies even when both words reside in short-term memory. Wollen and colleagues (Wollen, Weber and Lowrya, 1972) conducted an experiment in which subjects were presented with the drawings of two objects at a time and were asked to remember them for a later recall test. In one condition the two objects were pre-sented separately; in a second condition their degree of interaction was manipulated, so that the two objects (e. g., the drawing of a cigar and a piano) were presented as laying on each other. Results showed that interaction had a positive effect on recall probability, so that being able to recall one object determined higher chances or recall for the second object, too. This experiment shows that two semantically unrelated representations can boost each other’s recall if an arbitrary connection between them is established.
I am now going to attempt an interpretation of the interaction effect in the light of the concept of number of connections. It can be argued that by letting words interact with each other, a new connec-tion originates, namely the one putting them in relaconnec-tion. Therefore, we can hypothesise that this technique is successful because it increases the number of connections of target words in long-term memory and consequently the size of the lexical unit of the target word.
2.5.2 The multimedia effect
According to Schwanenflugel (1991), the greater the variety of processing modes that can be related to a target word, the greater is the possibility of later recall. Schwanenflugel’s hypothesis is supported by Chun and Plass’ findings (1996). In their experiment, they presented participants with a text con-taining unknown words. In the control group, explanations of unknown words were presented as writ-ten text, whereas in the experimental group they were presented as images and videos. A multimedia effect was found, showing that a target word is easier to recall if it is connected to memories generat-ed with different processing modes.
10 Chun and Plass’ findings can be ascribed to the fact that target words have developed an additional connection to a specific modality of processing. In the light of the aforementioned studies, I would like to conclude that the positive results of the multimedia effect could be a consequence of establishing new connections between words and processing modalities.
2.5.3 The propositional format of memories
Research agrees that short-term memory processes novel elements which are presented together by grouping them as opposed to treating them as isolate elements. This grouping strategy results in the propositional format of memories (Anderson, 2000; Tomasello, 1992): words are grouped together and constitute the arguments of a larger unit, called proposition. The strategy of assigning memories to propositional units has been called chunk-mode. Evidence supports the chunk-mode strategy: in-deed, it has been found that if one of the constituting arguments of a proposition can be recalled, chances that the other arguments are also going to be recalled increase significantly (Anderson, 2000). The important function of the chunk-mode has also been highlighted by Tomasello (1992), according to whom binding words together into more complex structures is a constituting part of the learning process.
Again, we can consider the positive effect of the propositional format of memories in the light of the number of connections. We can argue that the decisive factor of this technique is the establishment of connections between words. Semantic relations are not necessary for this technique to work. This is obvious if we consider that the advantage of the chunk-mode has been found also for the recall of number sequences.
2.5.4 The testing effect
The testing effect describes the impact that a test session has on the ability of later recall of vocabu-lary, if it is used as a study session. Indeed, testing sessions have been found to yield better results than additional study sessions. Testing practices have been compared to additional study sessions and results show that additional study practice is not as effective as testing is (Anderson, 2000). The main difference between testing practices and study sessions is that testing requires an active effort in or-der to retrieve the target word. This effort could be decisive in embedding the target word more effec-tively in long-term memory.
Roediger and Karpicke (2006) tested this hypothesis in an experiment. After a first presentation of target words, participants underwent either a further session of study or a test. The experimenters were able to assess retrieval-induced facilitation as a consequence of the testing effect. Carpenter and DeLosh (2006) also analysed the testing effect and were able to establish the positive effects of testing practice on later recall.
2.5.5 Depth of processing
Depth of processing refers to the idea first proposed by Craik and Lockhart (1972) that the more re-sources are devoted to processing target words, the more strongly these will be encoded in long-term memory. According to Craik and Lockhart, the first and most simple processing is phonological pro-cessing. Semantic processing requires more resources and involves more levels of processing and ensures therefore a stronger encoding in long-term memory. The act of relating personal experiences to target words is an even deeper level of processing which therefore ensures even more stable en-coding in memory. De Groot and Keijzer (2000) conducted an experiment in which they presented participants with a list of translation pairs from their L1 (Dutch) to a pseudo language. The word pairs were manipulated along the measure of concreteness, so that some word pairs depicted abstract concepts and other depicted concrete objects. Concrete objects are more likely than abstract words to have personal memories attached to them, as well as sensory features such as colour, smell, size,
11 shape etc. In the test task, recall for concrete objects was significantly better than for abstract nouns. The experimenters explained the result by arguing that words symbolising concrete objects were as-signed to long-term memory together with personal memories.
If we analyse depth of processing in the light of the number of connections, we can argue that deep processing consists in linking memories belonging to different areas of language (phonetic, semantic and so forth) to one target word. This technique can therefore also be argued to result in a bigger lexical unit than previous to its application.
2.5.6 Associations
Word associations can be defined as “the links that connect or relate words in some manner in a per-son’s mind.” (Schmitt and Meara, 1997, p. 19). Associations can also be created ad hoc and in the training phase of the current study I have used them purposefully in order to establish connec-tions between words which were not associated prior to the experiment. Associaconnec-tions derived from joint presentation. Since target words were non-words and were shown without explanation and out of context, the connections established were not of semantical nature, but one which only involved the form of words.
2.5.7 Frequency of presentation
Frequency of the input is a fundamental concept in entrenchment theories. In the case of vocabulary, it refers to the occurrences of a target word in the analysed speech. Frequency influences the way linguistic stimuli are apperceived and categorised. Psycholinguistic experiments on lexical frequency have clearly shown that frequency of exposure highly impacts on the ability to recognise, retrieve and process words (e. g. Knobel, Finkbeiner and Caramazza, 2008; Blumenthal-Dramé, 2012). Psycholin-guistics measures such as speed of response and eye movements have provided evidence for the assumption that frequent items require less time and less effort in order to be processed as compared to rare lexical elements. This has been proven to be the case for single words, as well as compound words and sentences. Results have been found to apply to first as well as second language (e. g. Chil-ders and Tomasello, 2001; Kidd, Lieven, and Tomasello, 2010).
Connections between words are also influenced by the frequency with which they are used. This phe-nomenon is explained by Hilpert and Diessel (2017). They argue that not only words, but also types of connections between words, have a specific level of activation. As words, connection types with a higher level of activation are more readily available for use. Hilpert and Diessel refer for example to the inheritance connection between words. This type of connection describes a relation that goes from a more general word to a more specified one. If this connection is often used to relate words to each other, the tendency will be to use it also in future situations.
As we can see, this reinforcement technique constitutes an exception to the other techniques pre-sented so far, since it yields higher activation levels without enlarging the lexical unit of the target word. For the current study, frequency was not manipulated.
In this section we have reviewed several techniques which have proven to increase the levels of acti-vation of target words in long-term memory. Even though these techniques stress the importance of depth and variety of processing, I have argued that the factor underlying their effectiveness is that the number of connections of target words is increased. In other words, I believe the crucial factor to be the quantity of connections and not their quality.
2.6 Connections and the statistical understanding of cues
12 As we have seen in the previous sections, the creation of new connections between target words and other words seems to yield positive results as regards the strength of encoding of target words in long-term memory. In this section I am going to consider the question of whether new connections could be detrimental for the purpose of encoding, differently from what has been assumed so far. The rea-son for this is that I would like to incorporate in this overview possible disadvantages of numerous connections. This would mean that they would hinder encoding in long-term memory, instead of facili-tating it. To this end, I am going to begin with MacWhinney’s (2008) concept of cue validity. This is an account on the characteristics cues should have in order to facilitate the learning process of words. Cues are the lexical elements which appear in the input along with target words. Then, I am going to analyse other accounts making observations about the statistical abilities of the brain in interpreting the input (Knill and Saunders, 2003). To conclude, I am going to present some views maintaining that statistical abilities have clear limitations and that multiplicity of connections can, under certain circum-stances, create instability (Lieven, 2010; Slabakova, 2013).
I would like to begin this review by citing MacWhinney on the Unified Model: “This model relies on a particular version of Construction Grammar that emphasizes the role of storage in lexical maps and the online integration of constructional chunks during both L1 and L2 processing. […]. The model em-phasizes the role of cue availability and reliability in determining the course of acquisition.” (MacWhinney, 2008, p. 363). MacWhinney emphasises the importance of two phenomena, cue avail-ability and cue reliavail-ability. Cue availavail-ability is the measure of cue presence when a target event takes place. Examples from the field of morphology are especially helpful here: let us take the English plural flexion of nouns as the target event. Furthermore, let us assume that the related cue that we set out to observe is the letter -s at word end. Cue availability is the amount of times the cue (-s) is present over the amount of times in which the event (plural) occurs. If the cue is always present when the event takes place, we have to do with a totally available cue. This idea can be elucidated in a simple question: when nouns have a plural flexion, how often do they end in -s?
Now I am going to elucidate the concept of cue reliability. This concept refers to the relationship be-tween the times the cue is present in the input over the times in which the cue has the function of expressing the target event. This idea can be elucidated in a second simple question: when -s occurs in the input at the end of a word, how often does it have the function of expressing plural flexion of a noun?
As we can see, in the example used here, even though the validity of -s for the plural flexion is very high, its reliability for this event is low, since the letter -s at the end of a word can correspond to sev-eral other events, for example to the third person agent in a verb conjugation. For this reason, it is difficult to draw a conclusion about cue and target event. This problem can be solved by resorting to the concept of cue validity. Cue validity equals the product of the two measures of cue availability and cue reliability and has a maximum value of 1. Cue availability and reliability also have a maximum val-ue of 1, implying that cval-ue validity is at its maximum when availability and reliability also are. In the example of the plural flexion for nouns this situation would occur when 1) every time when a plural noun flexion occurs in the input, the -s is also present; 2) every time a word ends in -s, we can be sure that this is the plural flexion of a noun.
MacWhinney’s considerations on cue validity are at odds with the argument of multiplicity of connec-tions reviewed in section 2.5. There, the argument was made that multiple connecconnec-tions facilitate the encoding of words in memory. Here, on the other hand, the argument is made that a simple one-to-one relationship would yield the best results. I am now going to apply the concept of cue validity to the field of vocabulary learning. The logical consequence of this concept would be that it is more ef-fective to learn a target word always in association with the same word, as compared to with several different words.
13 In the current section I am going to present a different view on the statistics understanding of the input. Fine and Jaeger (2011) emphasize the great flexibility of language users throughout their life span. They argue that language users can count on a statistical understanding which is highly adaptive to the input. This implies that they are able to pay attention to different statistical cues according to the validity that they have at a given moment. When the relied-on cue loses validity, language users are able to change their behaviour and attribute more importance to a more valid one.
Adaptivity to the input was assessed also in other cognition fields. Knill and Saunders (2003) were able to observe it as related to human vision. They conducted a study on judgement about slanted surfaces. Their participants disposed of two visual cues in order to make judgements. The experimenter manip-ulated the validity of either cue and could observe that, whenever the validity of one cue was not useful any more in order to understand the visual input, participants shifted to the more reliable cue.
2.6.3 The conservatism of language users
In the preceding sections of this discussion, I have showed arguments in favour of the hypothesis that language users are capable of managing multiplicity of cues in the input, since they possess refined skills of statistical analysis which allow them to use the most reliable cues and always learn from the input to use the implicit rule which governs the input. However, not always does the encoding of the input seem easy to understand. What if language users are overchallenged with the variety they en-counter in the input? Findings in the field of first language acquisition do stress the difficulty posed by mappings which are not one to one. Lieven (2010) found that if one function is mapped by only one form, that is, if the cue is reliable, the target form expressing the function will be learnt more quickly as in the case of functions which are mapped by a multiplicity of forms (that is, in the case in which cues are not reliable). One could hypothesise that learners will rely on other cues in in order to ana-lyse the input, but this does not seem to be the case or at least this mechanism does not kick off im-mediately. In other words, unreliable cues do not immediately cause the adaptive reaction which fos-ters the search for more reliable cues. On the contrary, it is possible that the outcome of such unrelia-ble cues be unclarity. Lieven goes further and explains that the input parameters on which language users rely in their L1 vary significantly according to the developmental phase they are in, so that at some stage cue validity can be preferred over cue reliability. However, this is not the case for the initial stages of development, where learners seem to prefer to be conservative in the understanding of the input. Analogue observations to Lieven’s have also been made in the field of adult second language acquisition. Slabakova (2013) found that variability in the input delayed the emergence of rule-like behaviour in the second language and argues that the amount of variety in the input is directly pro-portional to the time needed to produce a standard grammar. Interestingly, she equals variability to ambiguity, referring to the point in the analytical behaviour where the learner keeps at distance be-cause of their insecurity. Slabakova addresses the conservatism of language learners and points out that the process of L2-learning is guided by cues which are set close to those valid in the native lan-guage. Cues are not fixed at all times, but they tend to approach those that are significant for the L2. Nonetheless, the pattern outlined here clearly points at behaviours which are not open to the multi-plicity and diversity of cues, but rather – even though predominantly in the beginning stages of acqui-sition – rely on familiarity. This view is also reinforced by Regier’s findings (2005). Regier has tried to explain some phenomena in first language acquisition with a model which postulates the same cogni-tive mechanism put forward by Slabakova (2013). The experiments were conducted by simulating a neural network. The conclusions that he drew from the results were that so called second labels cause interference in memory. Second labels can be seen as form-function mappings which are not uniquely defined. This is for example the case mentioned above of one function having more than one form. As we have seen, non-unique form-function mappings constitute unreliable cue in the input. Therefore, second labels can also be defined as input with low cue reliability. As Regier (2005) states, second labels impair the understanding of the input, because they interfere with memory for the already stored first labels. Interestingly, the disadvantage of second labels over first labels remains, even
14 though over time comprehension probability for second cues rises above chance. This persisting dis-advantage can be explained with the mechanism of interference, which relies in turn on the one of spreading levels of activation: when the second label and its referent are presented to the system, the associated nodes will also be activated. This will be the case for the primary label associated to the referent. This mechanism will cause interference, hence impairing the learning process. This will be also true after a long time and after many presentations of the referent with the second label. Once again, we see that multiplicity of cues in the input cannot be readily processed and that, more mark-edly in the initial stages of the learning process, these could constitute an obstacle.
The views exposed in this section seem to undermine the effectiveness of the design of this study, which foresaw that in the condition of richness of connections, novel words be presented in every trial in combination with a new word. However, these theories are especially relevant for the mechanisms that guide the statistic comprehension of the input of language users. Therefore, these arguments are not going to be incorporated in the conception and design of this study.
2.7 The current study
2.7.1 Theoretical background
The current study sets out to analyse the impact that the size of lexical units has on the ability of recall of target words in a second language.
In section 2.1 I introduced the concept of lexical units of entrenchment theories. In this theoretical frame, lexical units are defined as clusters of words connected to one another in long-term memory. Their interconnectedness is the reason why words belonging to the same unit have a potential for occurring together. Entrenchment theories assume that the fact that a word is connected to other words is paramount for its encoding level in memory. Some accounts presented in section 2.6.3 (e. g., Lieven, 2010, Slabakova, 2013 and Regier, 2005) express the view that additional connections can be detrimental for the encoding of words in memory. However, the reason for this lies on the reliability of connections and not on their number. Therefore, it can generally be held that entrenchment theories do not assume the size of lexical units, i. e., the number of words contained in them, to impact on the ability of recall of the single words. This is the reason why I have set out to explore this aspect of lexi-cal units.
In section 2.5 I introduced a series of approaches on word processing (among others, Craik and Lock-hart, 1972, Anderson, 2000 and Langacker, 2017). The common view of these approaches is that vari-ety of processing and depth of processing are paramount in order to achieve a better encoding of words in memory. However, these approaches do not make a prediction about the number of connec-tions of words in memory either. I assume that the positive effects described by these approaches can also be explained by using the concept of the number of connections. I am going to
In line with the view of this body of scholarship 1) presentation is more effective if words are not pre-sented alone, but interacting with one another. In this case, interpreting the results quantitatively, we can argue that new connections between words are created; 2) processing of words is more effective if it involves several levels of processing. In this case, interpreting the results quantitatively, we can argue that new connections are created between a word and some of its own dimensions, e.g., the semantic one, the one of personal memories etc.; 3) storage is more effective if words are not stored away as single elements, but as parts of bigger units. In this case, interpreting the results quantitatively, we can argue that new connections between words are created.
Therefore, as I have illustrated in the previous sections, they can be understood as quantitative and not qualitative phenomena.
In this study, I argue that the size of lexical units is crucial for the encoding of words in memory and I argue Anderson’s theory of spreading activation (2010) to play a key role in the encoding mechanism.
15 Anderson’s theory defines words’ levels of activation as the input necessary to make a target word available for further action. The greater the level of activation, the more readily available the target word. Furthermore, the theory states that, when a word is accessed and therefore activated, all words connected to it are also accessed and activated. This phenomenon is due to the spreading nature of activation. The consequence is the increase of levels of activation of all words involved in the process. This mechanism is particularly relevant for the current study. Let us assume that a word is activated through direct access (let us call this activation round no. 1), for example through written presentation. All words belonging to the same lexical unit are consequently going to be activated, too (activation round no. 2). The spreading nature of activation determines that the word originally activated is going to be accessed again (activation round no. 3), since it is connected to the words which were accessed in round two. Activation loses strength after each spreading round; however, it has been shown to impact on the ability of vocabulary recall also when indirect (Anderson, 2010). The mechanism by which spreading activation impacts on the level of activation of the single words of lexical units is par-amount in the current study.
I am now going to apply the mechanism described above to two hypothetical lexical units, a bigger and a smaller one. I am going to represent them graphically in the figures below and I am going to describe the effects of spreading levels of activation in the context of lexical units with different sizes. Figure 1 represents a bigger lexical unit, constituted by five words (a, b, c, d and e).
Figure 1
Figure 2 represents a smaller lexical unit, constituted by two words (a’ and b’).
16 If the target word a of the bigger lexical unit is accessed, it gets activated (activation round no. 1), as represented in Figure 3.
Figure 3
The activation of the directly accessed word does not remain by the target word, but spreads to the words connected to it: b, c, d and e (activation round no. 2). In Figure 4 we can observe that each of the words connected to a (b, c, d and e) gets activated in turn and that activation spreads further departing from these words (activation round no. 3).
Figure 4
This activation round involves also a, from which activation originally departed. As a consequence of this spreading mechanism, a gets a certain amount of direct activation at the beginning of the process described (in the activation round no. 1) and a smaller amount of indirect activation further on (in the activation round no. 3). As far as the smaller lexical unit is concerned, the same process applies.
17
Figure 5
2.7.2 Research question and expectations
Contrary to what is assumed by the entrenchment theories exposed here, I assume the size of lexical units to be crucial for the strength of encoding of words in memory. Furthermore, contrary to what assumed by several approaches on vocabulary learning, I assume that depth of processing and variety of processing are not the decisive factors in the encoding process of words in memory. In fact, I as-sume that they can be explained by resorting to the number of connections of words within lexical units. The concept of lexical unit and Anderson’s theory of spreading activation (2010) are the very conceptual pillars of the research question I have formulated in this study:
RQ: Assuming equal frequency of occurrence, do words belonging to lexical units of different sizes show different levels of activation?
Frequency of occurrence is constituted by the times a target word has been accessed in the mental lexicon. As I have illustrated in section 2.5.7, there is consensus on the fact that frequency of occur-rence highly influences the levels of activation of words. Therefore, in the current study this measure has been treated as a fixed variable. However, I argue the size of lexical units to also have a great im-pact on levels of activation.
In the following section I am going to briefly outline the design of the study and motivate its structure in relation to the research question posed.
My prediction about the results of the experiment is that the research question will be answered posi-tively. This would mean that the size of lexical units is crucial in determining the level of activation of words. Furthermore, I expect bigger units to yield higher levels of activation and therefore be advan-tageous for the encoding of words in memory.
2.7.3 Design of the study
In order to be able to test the research question, I designed an experiment with two different condi-tions of novel vocabulary presentation. Since I aimed at exploring the role of the size of lexical units, I designed an experiment with two training conditions: in one condition (condition of richness of con-nections), novel words were presented in order to be integrated in bigger lexical units. In the other condition, (condition of poverty of connections), novel words were presented in order to be integrat-ed in a smaller lexical unit. Lexical units were creatintegrat-ed ad hoc during the training session: the connec-tions between words were established by means of repeated presentation in the training session and did not existed before the training session.
18 The output of the two training conditions in terms of lexical units is analogous to the conditions de-scribed in Figure 1 and Figure 2 of the previous section.
In order to create lexical units, target words were presented together with known words, which I called accompanying words (these words will be described in detail in section 3.3 of the next chapter). As I stated in this chapter, I argue depth of processing to not have a decisive impact on the ability of retaining words. I have argued that the truly decisive measure is the number of connections, irrespec-tive of their type. Therefore, I aimed at creating superficial connections between words, i. e., connec-tions based on the surface form of words. I achieved this by excluding from the training session any kind of explicit processing of either the non-words or the real words.
The reason why I decided to connect target words to known words is that the embedding of novel vocabulary in long-term memory has shown to rely on words already residing in long-term memory and use them as anchors for the new ones (for more details about this, see section 2.2 on the role of long-term memory in the learning process).
The frequency of occurrence of target words was the same across the two conditions. Therefore, we can assume that the levels of activation of target words of the two conditions were comparable. The special interest of this study was the part of activation which was determined not by frequency of occurrence, but by the condition of integration in memory.
19
3. Methodology
In the present chapter I am going to describe the methodology of the current experiment. I am going to start by describing the criteria which led to the selection of the participants; then I am going to describe the materials used for the training session of the experiment. Successively, I am going to describe the procedure adopted. Then I am going to illustrate the training session and the different designs of the two training conditions. Finally, I am going to present the test task and the scoring ap-plied.
3.1 Participants
For the current study I recruited as participants 20 native speakers of German with English as their first L2. Participants were university students and postgraduate students and they had an English pro-ficiency level corresponding to the CEFR level of B1-C1. The language propro-ficiency level was either certified by an official language examination or self-assessed based on the CEFR criteria. The differ-ence in proficiency levels did not compromise the validity of the study: indeed, the only linguistic pre-requisite was to be familiar with English sounds and the English morphology. The reason for this is that participants had to learn non-words (the target words of the experiment) in association with accom-panying words which belonged to the A1 CEFR level. Participants were randomly assigned to one of the two conditions of the experiment in equal number. In order to obtain randomised groups, the randomisation formula of Excel was used. In the course of the experiment, the data of one participant had to be excluded; consequently, the analysed data was constituted of nine complete participant data sets for the group of richness of connections and of ten data sets for the group of poverty of connections. Participants did not receive any compensation for their participation in the study.
3.2 Training conditions
The current experiment was designed in order to compare two training conditions: the condition of richness of connections (RoC) and the condition of poverty of connections (PoC). The two conditions differed with regards to the design of the training session. In both conditions, the aim of the test task was the creation of lexical units. These were constituted of target non-words and real words present-ed next to target words. The crucial difference was the size of lexical units: whereas in the RoC condi-tion they contained five words, in the PoC condicondi-tion they contained only two words. These specific differences are going to be illustrated in greater detail in section 3.5. No control group was included in the design.
3.3 Materials
3.3.1 Training session materials
The stimuli of the training session consisted of target words and accompanying words. Recall of words was tested in the test task, whereas accompanying words were not and were only used in the training session. Target words were non-words legal in the English language system, i. e., non-words which respected the phonological and morphological rules of English. The reason for using non-words was that only these make it possible to control for previous knowledge, allowing to set the level of previ-ous knowledge to zero for all participants. This was important in order to be able to test only the
ef-20 fects of the training session, which aimed at making the participants learn the target words in lexical units.
English non-words were taken from the database assembled by Balota and colleagues (Balota, Yap, Cortese, Hutchison, Kessler, Loftis, Neely, Nelson, Simpson and Treiman, 2007) in their English Lexicon Project, which was originally constituted by 40,481 words and 40,481 non-words. The database was filtered by word length: only non-words of 1-3 syllables of length were taken. This length criterion was set so to reproduce the most frequent patterns of word length in real English. Indeed, according to the CMU Pronouncing Dictionary created by the Speech Group at Carnegie Mellon University (CMU), from a total of 133,357 recorded words, 111,934 have a length of 1-3 syllables, amounting to 83,9% of the total. After running this first selection on the pool of non-words, a random selection formula was car-ried out. This was the same Excel formula applied for the selection of participants. The result of this procedure was a pool of 200 non-words. A further selection was applied to this second pool of stimuli in order to be sure to include words with different endings. This step was taken in order to avoid that the repetition of patterns could skew the results of the training session by positively influencing memory retention for rhyming words. A further selection of the stimuli was aimed at ensuring varia-tion for grammatical categories, so that the main categories of noun/verb, adjective, and modal ad-verb were represented. The final pool of stimuli was constituted of 20 non-words. Target non-words were presented in the training session, each for four times. The total presentation trials amounted therefore to 80. Target non-words were presented together with real words (accompanying words). Accompanying words amounted to 80 units and were randomly selected from a word list created by the Language Assessment Department of Cambridge English for the project “Key English Test” (Cam-bridge English, 2012). To the Key English Test project belonged words that had been assessed as be-longing to the productive vocabulary of L2 speakers with a proficiency level corresponding to the A2 of the CEFR. In this experiment, the reason for selecting accompanying words generally used at a lower proficiency level than the proficiency levels of participants, was that I wanted to increase chances that all participants were familiar with all accompanying words. This was crucial, since accompanying words constituted the anchors in long-term memory to which novel non-words were connected by means of associated presentation. As for accompanying words, these were not filtered according to word length.
In the following tables I am listing some examples of target non-words and of some accompanying words. The complete set of target non-words and accompanying words is listed in appendix 1.
Table 1
Target non-words
Examples of target non-words dawyer earnong apabian melps rudaism Table 2
Accompanying real words
Examples of accompanying real words cut
morning air centre people
21
3.3.2 Test task materials
In the test task, participants were tested for their recall ability of stimuli presented in the training ses-sion. The stimuli used to this end were 20 non-words. Half of them had been presented in the training session. The other half was constituted by a slight modified version of the non-words of the training session. The modification consisted in the substitution of one letter in each non-word. The results were also non-words that of course respected the English phonetic and morphology. Importantly, in the test task this set of modified non-words required a different response as the known non-words: the correct answer for known non-words was affirmative (stimulus is known), whereas the correct answer for modified non-words was negative (stimulus is unknown). During the familiarisation proce-dure, participants had been informed that they had to respond affirmatively only when the stimulus presented was exactly like the one learnt in the training session.
3.4 General procedure
Participants completed both the training session and the test task by using the experimenter’s laptop computer. Both the training session and the test task took place in the same quiet room and were administered with a break of one hour separating them. Previous to the experiment, participants were made familiar with the procedure by means of a trial in which both training session and test task were simulated. The familiarisation phase lasted approximately five minutes. The experimenter was present in the room during the trial, in order to be able to answer questions about the procedure. During the training session, the experimenter was outside of the room. Participants were made aware of this before the beginning of the session. Upon completion of the training session, participants had been instructed to call the experimenter. The length of the training session was set by the experimenter to 8 minutes and could not be speeded up or slowed down by participants. As the training session was concluded, the participants left the experiment room and came back one hour later for the test task. During the test task the experimenter was also outside the room. The test task was self-paced and was completed in 2-3 minutes, depending on the participant.
3.5 Familiarisation phase
In this phase, the experimenter repeated the basic information about the experiment, including in-formation about its approximate duration, its phases and the administration mode (visual presenta-tion of the text on a computer screen). This informapresenta-tion had already been communicated in the re-cruitment phase. The experimenter did not disclose any information on the scope of the experiment. Then the experimenter illustrated the structure of the training session: she said that participants would read words on the computer screen, always two at a time, and that they had to read both. After this introduction, the trial of the training phase was initiated and participants saw it on the computer. Then, the experimenter illustrated the structure of the test task: she said that participants would be shown words on the computer screen, this time one at a time; moreover, they would need to make a decision about each word and press the corresponding key on the keypad. The association between decision and key was the following:
1) M-key: the word is known from the training phase 2) Z-key: the word is not known from the training phase
Participants were informed about the fact that a word that was only similar to one presented in the training session, but not exactly the same, would not qualify for the m-key answer.
Following this explanation, participants went through a trial test task and questions about the proce-dure were answered.
22
3.6 Training session
All participants underwent a single training session during which they were showed the stimuli in form of written text on the experimenter’s laptop computer. The training sessions of the two experimental conditions were differently designed. In the following, I am first going to describe the features of the training session common to both conditions; then I am going to outline the distinctive features of each condition.
3.6.1 Features common to both experimental conditions
The stimuli were presented in written form by means of a PPT-presentation. The PPT-presentation was shown in full screen modus. The background was black and the text font was white. The target non-words were the same in both conditions and amounted to 20 units. Each non-word was shown in a separate PPT-slide and was presented repeatedly for a total of four times. In each presentation trial the target non-word was not presented alone on the slide, but alongside an accompanying word. Accompanying words were real words. Target non-words were presented at the centre of the slide, whereas accompanying words were presented at the bottom of the slide. Participants were informed about the fact that they had to read both words. The display time of each slide had been automatically set at six seconds. This relatively long exposure time was motivated by the fact that non-words require a longer reading time than real words. The total duration of the training session amounted to eight minutes. The four repeated presentations of each target non-word were randomly distributed along the overall number of total presentations. This order was determined by using the Excel randomisa-tion formula. Massed presentarandomisa-tions, that is, subsequent repeated presentarandomisa-tions of the same target non-word, were avoided. The reason for this choice is that this method has been proven to be less effective than spaced presentations for subsequent retention (Ellis, 1995). The presentation order of the target non-words was the same for all participants of both conditions, since the order was ran-domised only once and then used for both conditions and all participants.
The indication given by the experimenter regarding the behaviour to adopt during the training session was to read both words presented on the screen mentally or aloud. The training sessions of all partici-pants took place without interruptions except for one. The participant in question was excluded from the study and did not proceed to the test task.
In the following paragraphs I am going to illustrate the difference between the two training conditions.
3.6.2 Features of the RoC condition
In the RoC condition, target non-words were presented in all four presentation trials with a different accompanying word. Table 3 below exemplifies this presentation pattern. Target non-words are shown between asterisks, next to them the accompanying words.
Table 3
Presentation trial RoC condition
1 *dawyer* – cut
2 *dawyer* – arm
3 *dawyer* – day
4 *dawyer* – eat
3.6.3 Features of the PoC condition
In the PoC condition, target non-words were presented in all four presentation trials with the same accompanying word. Table 4 below exemplifies this presentation pattern. Target non-words are shown between asterisks, next to them the accompanying words.
23 Table 4
Presentation trial PoC condition
1 *dawyer* – cut 2 *dawyer* – cut 3 *dawyer* – cut 4 *dawyer* – cut
3.7 Test task
3.7.1 Design
The test task consisted in a lexical decision task: participants were shown a total of 20 single words. These were displayed one at a time on the computer screen, in white font on a black background. Ten were the non-words that they had learned in the training session and another ten were slightly modi-fied non-words (for more details about modimodi-fied non-words, see section 3.6.2). Participants were instructed to decide whether they recognised these words from the training session in this exact form or not. The decision was made by pressing on one of two keys on the keyboard: the M-key for a posi-tive answer and the Z-key for a negaposi-tive answer. The test task was self-paced: the following word was shown only after that one of the two pre-set keys was pressed. This led to the consequence that in some rare cases (7 in total, for all decision trials and all participants) reaction times were too long to be included in the statistical analysis. The test task was run by using the software E-prime 2.0.
3.7.2 Scoring
For the test task two measures were registered and analysed: accuracy of response and response speed (RTs). Accuracy was computed by summing up all correct responses (1 point for correct answers, 0 points for incorrect answers). Response speed was computed as the reaction time between stimuli presentation and key pad response (the hitting of one of the two allowed response buttons). The test task was programmed with the software E-Prime 2.0.
