Accepted 26 March 2014Available online 19 April 2014

Keywords:GlossInferenceRetrieval

rm-meaning con-ted that glossing isn (2001), glossingproviding accurateprehension. Third,e. Fourth, learnersre more frequently

glossed produced more vocabulary learning. However, Laufer and Hulstijn (2001) did not think that glossing was an effectivemethod of facilitating vocabulary learning because compared with other tasks such as using words in a composition, glossingdid not encourage search or evaluation processes and only induced weak involvement load.

* Corresponding author. Tel.: 886 2 77341817.E-mail addresses: kate91a@yahoo.com.tw (L.-l. Huang), cclin@ntnu.edu.tw (C.-c. Lin).

Contents lists available at ScienceDirect

System

journal homepage: www.elsevier .com/locate/system

System 44 (2014) 127e136from context, Hulstijn, Hollander, and Greidanus (1996) emphasized the importance of establishing fonections through external meaning references such as glosses. Rott, Williams, and Cameron (2002) indicafrequently found to improve both the extent and speed of lexical acquisition (p. 185). According to Natiohas four advantages. First, learners can read difcult texts without simplication or adaptation. Second,meanings prevents learners from guessing incorrectly, which should facilitate vocabulary learning and comglossing does not seriously interrupt the reading process and it is less time-consuming than dictionary usmay focus on glossed words, which may encourage learning. Rott (2007) also showed that words that weTextual input is indispensable to vocabulary growth (e.g. Krashen, 1989) but not all natural contexts ensure successfulinferences of word meaning (Hulstijn, 1992; Laufer, 1997), not to mentionword learning. To maximize the benet of learningVocabulary learning

1. Introductionhttp://dx.doi.org/10.1016/j.system.2014.03.0060346-251X/ 2014 Elsevier Ltd. All rights reserved.hundred and eighteen university students read a text with target words occurring threetimes, either glossed or unglossed, under one of three conditions. In the inference-gloss-gloss condition, participants inferred the word meaning in the rst word encounter andwere provided with glosses for the subsequent two word encounters. In the gloss-retrieval-gloss condition, participants were provided with glosses for the rst and lastword encounters but had to retrieve the word meaning in the second word encounter. Inthe full glossing condition, participants were provided with glosses for all the word en-counters. Immediately after the intervention, all participants completed a comprehensiontest and three vocabulary posttests. They completed delayed vocabulary posttests 2 weekslater. The main results showed that the gloss-retrieval-gloss condition was the mosteffective at improving vocabulary learning. The study concludes that gloss-retrieval-glossis facilitative to learning new words from context.

2014 Elsevier Ltd. All rights reserved.Received 15 April 2013Received in revised form 24 March 2014encourage inferring or word meaning retrieving. This study investigates the effects ofcombining glossing with inferring or meaning retrieval on vocabulary learning. OneThree approaches to glossing and their effects on vocabularylearning

Lo-li Huang, Chih-cheng Lin*

National Taiwan Normal University, No. 162, Heping East Road, Section 1, Taipei 106, Taiwan, ROC

a r t i c l e i n f o

Article history:

a b s t r a c t

Glossing facilitates the creation of correct form-meaning connections, but does not

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136128It seems clear that glossing can increase the possibility of learning correct word meanings in context, but leaving wordsunglossed for learners to infer or retrieve their word meanings may increase the involvement load and mental effort, thuscontributing to better retention. Because both glossing and not glossing have advantages and disadvantages, precedingglossingwith inference or following glossingwith retrievalmaycomplement vocabulary learning. However, Rott (2007) foundthat rst glossing words and then retrieving their meanings was not signicantly more effective than only glossing words.Although Rott (2007) explained that repeated processing of glossesmight compensate for the quality of attentional resourcesspent during one word retrieval (p. 189), she allowed learners to refer back to the previously glossed word encounteredduring retrieval, which may only induce visual search for the previous gloss, but not search from memory. In other words,Rotts distinction between retrieval and simply a second glance at glosses might be blurred. To eliminate this concern, theeffectiveness of full glossing (glossing a word in all of its occurrences) and glossing plus retrieval should be reexaminedwithout allowing previous glosses to be viewed during retrieval. Research on inference followed by meaning assistance hasoften presented this assistance in the form of a checklist (Mondria, 2003), but seldom occurs as a gloss along with the word inthe text. Additionally, the effectiveness of inference followed by glossing inword learning has never been compared with thatof glossing plus retrieval or full glossing. Therefore, it is worth investigating these three interventions together.

2. Literature review

Using glosses provides support in unhelpful contexts, but it might deprive the reader of an opportunity to infer, thusreducing the amount of processing (Watanabe, 1997, p. 289) and contradict Laufer and Hulstijns (2001) Involvement LoadHypothesis. The Involvement Load Hypothesis predicts that glosses, although they induce a moderate need imposed by thetask, might not be an effective method of facilitating word learning because they do not trigger readers search or evaluationprocesses. Consistent with the Involvement Load Hypothesis, Rott (2005) found that readers provided with translationglosses only used meta-cognitive strategies, involving noticing of the occurrence of the lexical form and monitoring of wordcomprehension. (without) any meaning making or word inferencing processes (p. 103).

In contrast, inferencing, which involves using learnersworld and linguistic knowledge to guess wordmeanings in context,may involve more mental effort and result in higher lexical retention (Hulstijn, 1992). However, Mondria (2003) found thatthe meaning-inferred method (i.e., learners inferred word meanings from context before memorizing them), though yieldinga similar amount of retention to the meaning-given method (i.e., learners were provided with word meaning), was lessefcient because of its time-consuming nature. The meaning-inferred method also has the following limitations. Learnersmay infer from insufcient contextual clue, need to unlearn incorrect inferences that they have made, and not be skilledenough at inferring (Hulstijn, 1992).

To retain the cognitive benets of inferring word meanings from context, Hulstijn (1992) introduced multiple-choiceglosses, which required some mental effort to select from target word meaning options based on text information. In aqualitative investigation, Rott andWilliams (2003) found that multiple-choice glosses triggered a search for lexical meaningand an interaction between the various gloss options and the context provided by the passage. led readers from a tentativeto a stable form-meaning mapping (p. 53) and were more likely to prevent readers from skipping target words than the no-gloss condition. Similarly, multiple-choice glosses, compared with single translation glosses, helped readers to retain morevocabulary and stimulated them to use more semantic-elaborative processing strategies, including active evaluation of theinitial meaning assignment, further search for meaning clues, and reevaluation of form-meaning connections in subsequentencounters (Rott, 2005). Such elaborative processing with rehearsal is conducive to retention (Hulstijn, 2001). However, Rott(2005) did not exclude the concern that more time was spent on multiple-choice glosses. Additionally, irrespective of howmuch the multiple-choice procedure reduced erroneous guesses, it was inevitably and inherently error prone (Hulstijn,1992, p. 123). Even when the multiple-choice procedure only included two alternatives, some students still guessed incor-rectly, making the multiple-choice condition generally, although not statistically, inferior to the single gloss condition(Watanabe, 1997). These incorrect guesses can be transferred and strengthened in subsequent encounters when they makesense in reader discourse models (Rott, 2005). Rott et al. (2002) also found that the superior immediate vocabulary learningproduced by the multiple-choice gloss condition was not retained better than that produced by the no-gloss condition 5weeks later. This might suggest that the mental effort exerted through processing multiple-choice glosses did not lead to thesufciently deep level of processing that is claimed to promote long-term retention of formemeaning connections.(andthat) the search and evaluation demands of usingmultiple-choice gloss conditions were too low to be effective (p. 207). Thesedisadvantages mean that multiple-choice glosses can be improved.

Another major process involved in committing a word to memory is retrieval. Retrieval involves rst forming a perceptionand then retrievingmeaning frommemory (Nation, 2001). Each retrieval may strengthen the link between form andmeaningor create an additional retrieval route, making it easier to retrieve the same item again if the interval between two retrievals isnot so long that the previous memory fades (Carrier & Pashler, 1992; Nation, 2001; Roediger & Butler, 2011). When learnersretrieve an encoded meaning in their mental lexicon, this may foster meaning retention (Rott, 2007) or attenuate the rate atwhichmeaning is forgotten (Wheeler, Ewers, & Buonanno, 2003). Rott (2007) showed that wordmeaning retrieval resulted inmore vocabulary learning than visual enhancement by using bold text. In her study, participants read three texts. Targetwords appeared four times in three different conditions: the four-time gloss (4G) condition, the gloss-retrieval-bolded-bolded (GR) condition, and the gloss-bolded-bolded-bolded (GB) condition. The results showed that the GR conditionyielded signicantly more vocabulary learning and retention than the GB condition, suggesting that retrieval consolidated

form-meaning connections better than bolding did. However, no signicant difference existed between the 4G condition andthe GR condition. This surprising result may indicate that allowing learners to check previous glosses during retrieval isproblematic or suggest that the retrieval effect was as strong as the effect of repeatedly processing glosses.

This review shows that the literature is uncertain whether more mental effort inuences long-term word retention.Therefore, it is unclear whether glossing preceded by inferring or followed by retrieval, which is assumed to require moreprocessing effort than full glossing, facilitates word learning. Thus, it is worth investigatingwhether these interventions affectvocabulary learning.

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136 1293. The study

This study addressed the following research questions:

1. Do glossing preceded by inference, glossing followed by retrieval, and full glossing interventions affect vocabulary learningdifferently?

2. Does response accuracy of inference and retrieval predict vocabulary learning differently?

3.1. Participants

We recruited 124 freshmen at the high-intermediate level1 from a university in Taiwan for the study, but six of themwereeliminated from the analyses because they had learned one of the target words or did not nish the delayed posttests. Theremaining 118 participants, including 88 females and 30 males, were randomly assigned to one of the three interventionconditions, with 39e40 students in each condition. The average participant score for the reading section of an Englishplacement test2 administered by the university was approximately the same in each condition, ranging from 75.45 to 78.51,F(2, 115) 1.29, p .28.

3.2. Target words and material

The target words included eight low-frequency nouns, 7e11 letters long. They were cognoscente, pentimento, sinistral,dextral, sfumato, pigment, provenance, and craquelure. Thesewords were judged to be difcult for university freshmen. Most ofthem were technical art terms and all of them were essential to the main ideas in the passage.

Thematerial was a text on art authentication adapted frommultiple sources, including three online articles and two articlesfrom CNN Interactive EnglishMagazine. The text included 806words and eight paragraphs at a Flesch-Kincaid Grade Level of 12.Two instructors teaching high-intermediate freshman English at the university judged its difculty appropriate for the targetparticipants. Each paragraph appeared on one page. The shortest and longest paragraphs included 80 and 122 words,respectively. The eight target words were embedded and bolded, and each target word occurred three times in the wholepassage. When a target word appeared, it was either glossed in Chinese in the margin or unglossed. When it was unglossed,participants were required to infer or retrieve the words meaning in Chinese and provide their answers in the blank spaces,which were recorded on a computer. To ensure that participants could not see the target gloss during inferring or retrieval, atargetworddidnotappear twice in the sameparagraphandonceparticipantsmovedon to thenext paragraphon thenext page,they could not return to the previous one. Three types of glossedeunglossed combinations were created for the target words:the inference-gloss-gloss condition, the gloss-retrieval-gloss condition, and the full glossing condition. In the inference-gloss-gloss condition, the targetwordswereglossed in all but therst occurrence,whereparticipantswere required to infer thewordmeaning. In the gloss-retrieval-gloss condition, the target words were glossed in all but the second occurrence, where par-ticipants were required to retrieve the word meaning. In the full glossing condition, the target words were glossed in everyoccurrence. When a target word rst occurred, its contextual clues were carefully designed to ensure that its meaning wasinferable in the inference-gloss-gloss condition. However, the design did not ensure that all the clues were equally inferable.

Following the reading, participants were asked to nish a reading comprehension test. The reading comprehension testwas a four-item choice test. The rst two items tested the global knowledge of the test, including one asking about the textcategory and the other about themain idea. The last two itemsmeasured detailed information related to the knowledge of thetarget words (see Appendix B). In each item, there were ve options, including one correct answer, three distracters, and anoption of I dont know to minimize the guessing effect. One professor and a doctoral student experienced in item designingwere invited to check the test validity.

1 Based on the results of a placement test, the university assigned students to freshman English classes at four different levels: advanced, high-intermediate, low-intermediate, and low. All the high-intermediate participants scores on the placement test were over 121, which translated into 71on TOEFL (IBT), 527 on TOEFL (PBT), 197 on TOEFL (CBT), and 750 on TOEIC.2 The difculty of the placement test falls between the intermediate and high-intermediate levels of the General English Prociency Test (GEPT), a widely

adopted English prociency test developed by Language Training and Testing Center (LTTC) in Taiwan. The test consists of a reading section and a listeningsection. The listening section covers tasks of answering questions and comprehending conversations and short talks. The reading section of the test iscomposed of vocabulary and structure tests, cloze tests, and reading comprehension tests.

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e1361303.3. Instruments

3.3.1. Vocabulary pretestThe pretest was a vocabulary checklist. Participants were required to check every word to indicate whether they knew the

word, and if so, provide its Chinese translation. The checklist included 25 randomly arranged lexical items, including the eighttargetwords and 17 distracters (see Appendix A). Nine of the distracterswere selected from the rst to the third levels of the six-level reference word list published by the College Entrance Examination Center (Cheng, Chang, Cheng, & Gu, 2002) in Taiwan.Thesewordsare familiar to seniorhigh school studentsandthus college freshmenareexpected toknowthem.The familiarwordswere inserted into the test items to ensure that the participants did not skim through the list andmark everyword as unknown.The other distracters were low-frequency words, making the low-frequency target words less noticeable. The distracters wereeither verbs or adjectives, to reduce the possibility of participants learning that all the target words were nouns after taking thepretest. Data from participants who showed any knowledge of the target words were eliminated from the analyses.

3.3.2. Vocabulary posttestsThe immediate and delayed vocabulary posttests consisted of a form-recall test, a meaning-recall test, and a meaning-

recognition test (see Appendix C). The form-recall test required participants to translate Chinese words into the corre-sponding target words. The meaning-recall test required participants to translate the target words into Chinese. Themeaning-recognition test was a denition-selection test, which used target words as prompts. Each item included nineoptions: eight Chinese glosses for the target words, and an I dont know option. Only one option was correct. Each testconsisted of eight items. The content of the immediate posttest and that of the delayed posttest were the same. However, thetest items in both posttests were in different orders.

3.4. Procedures

Participants signed a consent form, completed a vocabulary pretest, and then received the computer reading instructions.Each group of participants was then asked to read the text containing the target words in the designated condition. Par-ticipants were instructed to read for comprehension. After the intervention, participants completed the reading compre-hension test and then the immediate vocabulary posttests. The whole intervention session required approximately 30 min.Two weeks after the immediate vocabulary posttests, delayed vocabulary posttests were conducted without any advancenotice. The vocabulary posttests (both immediate and delayed) were administered in the following order: form-recall test,meaning-recall test and meaning-recognition test. Once participants had completed a test, they moved on to the next testwithout being able to see the previous test.

Scores for each test were tallied separately. For the meaning-recognition tests, a correct answer received a score of 1 andan incorrect answer received a score of 0. Themeaning-recall test was scored in the samemanner, except that partially correctanswers, which were semantically related but incomplete translations of a target word, were awarded half a point. Tomeasure complete and partial vocabulary learning on the form-recall test, this study adopted a scoring procedure revisedfrom Barcrofts (2002) lexical production scoring protocol. Although Barcrofts scoring protocol aimed to measure completeand partial word learning, its scoring was neither accurate nor consistent. A score covered a range of correct proportions (e.g.0.5 points for an answer that was 25%e49.9% correct) and could represent the percentage of correct letters, the percentage ofpresent letters, or the correct number of syllables. To be more precise and consistent, this study determined scores for eachitem based on the percentage of correct and half correct letters in a word. Correct letters mean that the correct letters of aword are written in the correct order and half correct letters mean that the correct letters are written in the wrong order.When letters that could be both correct and half correct occurred, a judgment that awarded participants the most points wasmade. The percentage of correct and half correct letters in awordwas calculated by dividing the number of correct letters andhalf the number of half correct letters by the total number of letters in the target word. When more letters were written thanwere in the target word, the number of the extra letters was subtracted from the number of correct letters before division.There were eight items in each vocabulary test, so the maximal score was eight for each vocabulary test.

Scoring the meaning-recognition, and form-recall tests was objective; therefore, it was completed by one of the re-searchers. Scoring the meaning-recall tests involved certain subjective judgments on partially correct answers; therefore, anMA graduate was asked to complete the scoring with one of the researchers. The inter-rater reliability reached .98 (p < .01)and differences were resolved through discussions.

To determine if inference or retrieval accuracy predicted participant vocabulary learning and retention, participant re-sponses during the inferring stage of the inference-gloss-gloss condition or the retrieval stage of the gloss-retrieval-glosscondition were scored by two independent raters. The responses were scored using a similar method to that used toassess the answers to the meaning-recall test, except for the semantically related but incomplete responses. Because it wasmore difcult for inference-gloss-gloss learners to produce accurate responses during the inferring stage, semanticallyrelated but incomplete responses were considered correct. Inter-rater reliability analyses showed that agreement reached .80(p .00) in the inference-gloss-gloss condition, .95 (p .00) in the gloss-retrieval-gloss condition, and .88 (p .00) in theinference-gloss-gloss and gloss-retrieval-gloss conditions. Analyses of participant responses during inference or retrievalshowed that the rate of correct responses was higher during retrieval than inferring, with a rate of 53.85% during retrieval anda rate of 34.38% during inferring.

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136 1313.5. Data analysis

To address the rst research question on the effects of the intervention on vocabulary learning, a two-way repeated-measures MANOVA was run, with a 3 (intervention condition) 2 (time) design. The independent variables were theintervention condition (inference-gloss-gloss, gloss-retrieval-gloss, and full glossing), which was a between-subject factor,and time (immediate and delayed vocabulary tests), which was a within-subject factor. The dependent variables were thescores of the three types of vocabulary tests: the form-recall test, the meaning-recall test, and the meaning-recognition test.To compare the degree of attrition among the three interventions, forgetting rate analyses between immediate and delayedvocabulary performances were also conducted.

To address the second research question, linear regression analyses were employed to determine if the accuracy of theresponses made during the inferring and retrieval tasks predicted vocabulary learning.

4. Results

4.1. Effects of the intervention on vocabulary learning

The mean scores in Table 1 (see Appendix D) show that the gloss-retrieval-gloss group yielded the most vocabularylearning and retention on all the tests (immediate form recall, 3.57; immediate meaning recall, 6.06; immediate meaningrecognition, 6.44; delayed form recall, 1.52; delayedmeaning recall, 4.65; delayedmeaning recognition, 5.33), followed by theinference-gloss-gloss (immediate form recall, 3.52; immediate meaning recall, 5.50; immediate meaning recognition, 5.59;delayed form recall, 1.25; delayed meaning recall, 3.59; delayed meaning recognition, 4.60), and full glossing groups (im-mediate form recall, 2.71; immediate meaning recall, 4.46; immediate meaning recognition, 5.28; delayed form recall, 1.20;delayed meaning recall, 3.13; delayed meaning recognition, 4.28). Two-way repeated measures MANOVA showed that sig-nicant main effects were present for the intervention conditions, F(6, 226) 2.94, p .01, and time, F(3, 113) 61.43, p .00,but no signicant interaction existed between the two factors, F(6, 226) .90, p .50. The effect of the intervention con-ditions was further examined using tests of between-subject effects. They showed that signicant differences were producedby the meaning recall test, F(2, 115) 8.32, p .00, and meaning recognition test, F(2, 115) 4.31, p .02, but not by the formrecall test, F(2, 115) 2.15, p .12. Scheffs multiple comparisons showed that the gloss-retrieval-gloss condition wassignicantly superior to the full glossing condition at meaning recall, p .00, and recognition, p .02, but the inference-gloss-gloss condition was not signicantly superior to the gloss-retrieval-gloss and full glossing conditions. The effect of time wasfurther examined using tests of within-subjects contrasts. The tests showed that the amount of vocabulary learning variedsignicantly by time in the form recall test, F(1, 115) 165.08, p .00, the meaning recall test, F(1, 115) 66.49, p .00, andthe meaning recognition test, F(1, 115) 36.18, p .00.

Participants immediate and delayed vocabulary performances were subjected to forgetting rate analyses. The forgettingrate was calculated by dividing the mean difference between the immediate posttest and the delayed posttest by the meanscore of the immediate posttest and then converting the quotient into a percentage (Groot, 2000). Calculations from the datain Table 1 showed that the greatest rates of attrition in all the interventions occurred on the form recall test (full glossing,55.72%; gloss-retrieval-gloss, 57.42%; inference-gloss-gloss, 64.49%), followed by the meaning recall test (full glossing,29.82%; gloss-retrieval-gloss, 23.27%; inference-gloss-gloss, 34.73%) and the meaning recognition test (full glossing, 18.94%;gloss-retrieval-gloss, 17.24%; inference-gloss-gloss, 22.69%). The highest forgetting rates were observed in the inference-gloss-gloss condition and almost all of the lowest forgetting rates occurred in the gloss-retrieval-gloss condition, exceptfor the form recall test, which produced a rate that was almost 2% higher than its counterpart in the full glossing condition.The smaller rate differences in the gloss-retrieval-gloss and full glossing conditions show that both conditions were lesssusceptible to attrition than the inference-gloss-gloss condition (although the gloss-retrieval-gloss condition generated thehighest vocabulary learning whereas the full glossing generated the least. The gloss-retrieval-gloss conditionmay be superiorto the other two conditions; therefore, its vocabulary learning and retention levels were the highest andmost of its forgettingrates were the lowest. The low forgetting rates of the full glossing condition may be attributed to a oor effect, because itgenerated the least vocabulary learning.

4.2. Relations between response accuracy and vocabulary learning

Table 2 (see Appendix D) shows that most of the mean vocabulary test scores after correct responses were higher thanthose after incorrect responses, with the exception of the immediate meaning recognition test. In the inference-gloss-glosscondition, higher mean scores were observed for correct responses, except for the immediate meaning recognition test andthe delayed meaning recall test. In the gloss-retrieval-gloss condition, correct responses also produced better mean scores,except for the immediate meaning recognition test and the delayed form and meaning recall tests.

Regression analyses showed that response accuracy signicantly predicted vocabulary learning only when the meancorrect response scorewas higher than themean incorrect response score. In other words, correct responses weremore likelyto contribute to word learning than incorrect responses. In the inference-gloss-gloss and gloss-retrieval-gloss conditions,response accuracywas a signicant predictor of immediatemeaning recall performance, R2 .01, F(1, 630) 9.55, p .00, anddelayed meaning recognition performance, R2 .01, F(1, 630) 5.26, p .02. In the inference-gloss-gloss condition, inference

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136132accuracy played a signicant role in delayed form recall performance, R2 .02, F(1, 318) 6.42, p .01, and immediatemeaning recall performance, R2 .03, F(1, 318) 12.29, p .00. In the gloss-retrieval-gloss condition, a signicant differencewas only observed for the immediate form recall test, R2 .01, F(1, 310) 4.00, p .05. Therefore, whether the responsesduring the inferring or the retrieval stages were correct may be related to vocabulary learning and retention, especially in theinference-gloss-gloss condition.

5. Discussion

Participants vocabulary learning performance showed that all of the intervention conditions led to word learning. Of thethree interventions, the gloss-retrieval-gloss condition most beneted vocabulary learning, followed by the inference-gloss-gloss condition and full glossing condition. The difference between the full glossing and gloss-retrieval-gloss conditions wassignicant. The following sections present a detailed discussion on each intervention condition.

5.1. Full glossing condition

The vocabulary learning results produced by the full glossing condition conrm Rotts (2007) contention that reap-pearing glosses could frequently draw learner attention to target words and offer them repeated opportunities to isolatethe glossed word from its surrounding context to rehearse the word meanings in their working memory. However, of thethree interventions, the full glossing condition generated the least vocabulary learning. Additionally, the continual reap-pearance of glosses may not automatically lead to learners repeatedly processing glosses in the margin, let alone increasethe quantity of attentional resources spent on a word, as mentioned by Rott (2007). Therefore, although the involvementload hypothesis predicts that a gloss occurrence results in an involvement load of one because it induces learners need forword meaning, the load of reappearing glosses is not necessarily multiplied by the number of times the glosses occur, asRott assumed.

5.2. Gloss-retrieval-gloss condition

The results show that the gloss-retrieval-gloss conditionwas the best glossing arrangement for aword that occurred threetimes in the input passage. The condition had the best results of form and meaning learning and it also helped to retain themeaning learning more effectively than the other conditions. This nding supports that retrieval can strengthen the linkbetween form and meaning (Nation, 2001), foster retention (Rott, 2007), and attenuate the forgetting rate (Wheeler et al.,2003). As Rott (2007) suggested, the involvement load of a retrieval was assumed to be three because it stimulated theneed to understand a word, search for its meaning in the mental lexicon, and evaluation of whether the searched meaningwas contextually correct.

In the gloss-retrieval-gloss condition, the initial glossing (as opposed to no initial glossing in the inference-gloss-glosscondition) supports the idea that the connection between form and meaning must be initially established (Nation, 2001),or learner meaningmapping tends to remain shallow and vague (Rott &Williams, 2003). Subsequent retrieval may be the keythat makes the gloss-retrieval-gloss condition superior to the full glossing condition, because repeated opportunities ofretrieval, as well as repetition, contribute to incidental vocabulary learning (Karpicke & Roediger, 2008; Nation, 2001). Rott(2007) argued that subsequent retrieval of a word after initial glossing did not further strengthen the form-meaningconnection because learners may only focus on comprehending a text rather than attend to any aspect of a word. Howev-er, compared with Rotts participants who had the previous glosses for referencewhen retrieving wordmeaning, it is possiblethat learners in this study focused more on the bolded but unglossed word during retrieval because they were required toretrieve word meaning without referring back to the previous glossed word encounter; therefore, their search level mighthave been higher than Rotts. The third word appearance with glossing could serve as a conrmation of the retrieval. Ifforgetting or uncertainty occurs during retrieval, the nal conrmational glossing can help solve these problems, whichprobably explains why response accuracy to unglossed words did not predict the word learning in the gloss-retrieval-glosscondition as much as the inference-gloss-gloss condition.

5.3. Inference-gloss-gloss condition

The inference-gloss-gloss condition was the second best intervention condition found in this study. It was not signif-icantly superior to the full glossing condition, but it generated greater word learning than the full glossing condition bymore than 10% for immediate form and meaning recall and more than 8% for meaning recognition. The superior immediatevocabulary learning in the inference-gloss-gloss condition may be linked to the search and evaluation processes duringinferring that created a higher involvement load (Laufer & Hulstijn, 2001). Although the retention rate in the inference-gloss-gloss condition was higher than that in the full glossing condition, the inference-gloss-gloss forgetting rates werehigher than the full glossing forgetting rates probably because the increased immediate word learning may be morevulnerable to attrition.

Vocabulary learning of inference-gloss-gloss was also comparable to that of the best condition, gloss-retrieval-gloss. Infact, retrieval and inferring may share a common feature. During either retrieval or inferring, learners may rely on context to

cantly better form and meaning recall than incorrect inferences in this condition.

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136 133Hulstijn (1992) stated that when incorrect inferences are made, learners must unlearn these incorrect inferences. Not allincorrect inferences are easily corrected. They may be incompletely corrected, confused with correct glosses, or fossilized inthe memory. Therefore, although inferring can induce a high involvement load, its error-prone and uncertain nature mayreduce its effect on initial form-meaning establishment.

Inferring also exhibits the following disadvantages. Not all contexts are equally inferable (Laufer,1997). Learnersmay focusmore on the association between context and meaning than the connection between form and meaning (Mondria, 2003).When inferences are correct, learners may think that they already know the words and spend less time committing them tomemory or may not try to remember word meanings because they have fullled the comprehension requirement (Hulstijnet al., 1996; Mondria, 2003; Sokmen, 1997). Learners form-meaning mapping may be unstable because they tend to vaguelyunderstand the target words during inferring and are less likely to conrm or disconrm their guesses (Rott & Williams,2003). When an inferred meaning seems plausible in a reader discourse model, readers may be satised with the mean-ing assignment and transfer the inferred meaning to the next word encounter (Rott, 2005). Therefore, although inferringrequires a high involvement load, it does not always result in retention.

6. Conclusion

This study is subject to the following limitations. Participants were recruited from an EFL university at the high-intermediate level; therefore, it is uncertain whether the ndings can be applied to the general public. The study was con-ducted in a computer lab, which did not approximate the participants normal class and reading settings; therefore, theexperimental setting lacked ecological validity. The difculty in inferring each target word might be different, so embeddingthese words in different contexts or choosing different target words might generate different results. Only eight target wordswere included in the 806-word passage; therefore, adopting a text with different lexical coverage may produce dissimilarresults. The two items measuring the knowledge of the target words in the reading comprehension test may have been aconfounding variable in the design because they may have served as a vocabulary learning task, in which the meaning of thetarget items was reviewed. This study adopted a cross-sectional research design, and thus generalization of the ndings torepetitive intervention administration should be treated with caution.

In spite of the above limitations, the results did suggest that gloss-retrieval-gloss was the best intervention conditionfor vocabulary learning. The rst glossed input helps learners establish an initial form-meaning connection. During self-generated retrieval, learners reinforce their initial form-meaning connection and are likely to exercise their inferringskills if their retrieval fails. Whether the retrieval or the inference is correct or incorrect, it can be conrmed or veriedby the third glossed input. Therefore, teachers can explain or provide glosses for unknown words in the rst encounter,have learners retrieve the word meaning in the second encounter, and then help them to conrm the accuracy of theretrieval.

Although this study draws teachers and learners attention to a learning pattern of correct form-meaning establish-ment, self-generated retrieval, and follow-up conrmation, it does not intend to downplay the role of inferring in teachingand learning. As many studies have indicated, an effective approach to vocabulary development must have more than oneorientation (e.g. Nation, 2001). The ndings in this study support the advantages of correct form-meaning establishment,but the goal of this study is not only to emphasize themeaning-givenmethod. Althoughmeaning assistance is crucial to theinitial form-meaning connection and retrieval verication, cultivating learner abilities to infer meaning is still necessary.After all, efcient reading does not result from precise perception and identication of all elements, but from skill inselecting the fewest, most productive cues necessary to produce guesses which are right the rst time (Goodman, 1982, p.34).

Future studies could explore four research directions. This studymeasured vocabulary learning in isolation. A similar studycould replicate this study by using other contextual test measures. This study only investigated words that occurred threetimes; therefore, it is worth investigating words that occur more than three times to determine whether other interventionpatterns, such as inference-gloss-retrieval-gloss, can generate a greater learning effect. Another similar study can conductqualitative investigations to delve into learner strategy use. Future studies could also incorporate analyses of participantsindividual differences into research on intervention effects.

Acknowledgments

We appreciated the System reviewers insightful comments on an earlier version of this manuscript. We also thanked theMinistry of Science and Technology of the Republic of China, Taiwan for sponsoring this study under Contract Nos. NSC 102-2511-S-003-025 and NSC 101-2511-S-003-057-MY3.evaluate the possible meanings of a word. The main difference between the two conditions is that retrieval is primed with aglossed word encounter and is more likely to lead to a specic correct response, while inferring requires a learner to draw onwider knowledge sources to derive meanings that tend to be vague and error-prone, even after repetitive meaning guessing.This may explain why the correct retrieval rate was higher than the correct inference rate in this study. The low correctinference rate might limit word learning in the inference-gloss-gloss condition because correct inferences yielded signi-

Appendix A. A Vocabulary Checklist

Please check the following words to show whether you know the word or not.

e.g. house B I dont know this word. 1 I know this word, and it means e.g. farce 1 I dont know this word. B I know this word, and it means _______1. emphasize B I dont know this word. B I know this word, and it means _______2. disappear B I dont know this word. B I know this word, and it means _______3. sinistral B I dont know this word. B I know this word, and it means _______4. practice B I dont know this word. B I know this word, and it means _______5. quotidian B I dont know this word. B I know this word, and it means _______6. pigment B I dont know this word. B I know this word, and it means _______7. valuable B I dont know this word. B I know this word, and it means _______8. lachrymose B I dont know this word. B I know this word, and it means _______9. craquelure B I dont know this word. B I know this word, and it means _______10. continue B I dont know this word. B I know this word, and it means _______11. different B I dont know this word. B I know this word, and it means _______12. pentimento B I dont know this word. B I know this word, and it means _______13. maladroit B I dont know this word. B I know this word, and it means _______14. energetic B I dont know this word. B I know this word, and it means _______15. inveterate B I dont know this word. B I know this word, and it means _______16. cognoscente B I dont know this word. B I know this word, and it means _______17. asphyxiate B I dont know this word. B I know this word, and it means _______18. episcopal B I dont know this word. B I know this word, and it means _______19. sfumato B I dont know this word. B I know this word, and it means _______20. southern B I dont know this word. B I know this word, and it means _______21. produce B I dont know this word. B I know this word, and it means _______22. provenance B I dont know this word. B I know this word, and it means _______23. hypodermic B I dont know this word. B I know this word, and it means _______24. dextral B I dont know this word. B I know this word, and it means _______25. gubernatorial B I dont know this word. B I know this word, and it means _______

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136134

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136 135Appendix B. Reading Comprehension Questions

Table 1Means, standard deviations and learning percentage of vocabulary learning measured immediately and 2 weeks later.

Time Intervention conditions

Inference-gloss-gloss(N 40)

Gloss-retrieval-gloss(N 39)

Full glossing(N 39)

FR MR REC FR MR REC FR MR REC

Immediate 3.52 (1.74) 5.50 (2.09) 5.95 (1.87) 3.57 (1.60) 6.06 (1.39) 6.44 (1.31) 2.71 (1.82) 4.46 (2.18) 5.28 (2.59)44.00% 68.75% 74.38% 44.63% 75.75% 80.50% 33.88% 55.75% 66.00%

Delayed 1.25 (1.31) 3.59 (2.15) 4.60 (2.05) 1.52 (1.33) 4.65 (1.95) 5.33 (1.96) 1.20 (1.34) 3.13 (2.03) 4.28 (2.16)15.63% 44.88% 57.50% 19.00% 58.13% 66.63% 15.00% 39.13% 53.50%

Note. Maximum vocabulary learning score was eight in each cell. Assessment tasks: form recall (FR), meaning recall (MR) and meaning recognition (REC).

Table 2Means and standard deviations of vocabulary learning after correct and incorrect responses.

Interventions FR MR REC

IM DE IM DE IM DE

inference-gloss-gloss correct (N 110) .47 (.40) .20 (.38) .80 (.40) .42 (.49) .64 (.48) .64 (.48)inference-gloss-gloss incorrect (N 210) .43 (.65) .11 (.20) .62 (.48) .44 (.50) .72 (.44) .54 (.50)gloss-retrieval-gloss correct (N 168) .48 (.38) .17 (.29) .76 (.42) .57 (.49) .80 (.40) .70 (.46)gloss-retrieval-gloss incorrect (N 144) .40 (.35) .17 (.30) .75 (.43) .60 (.49) .81 (.40) .65 (.48)inference-gloss-gloss gloss-retrieval-gloss correct (N 278) .48 (.39) .18 (.33) .78 (.41) .51 (.50) .74 (.44) .68 (.47)inference-gloss-gloss gloss-retrieval-gloss incorrect (N 354) .42 (.55) .13 (.25) .67 (.47) .51 (.50) .76 (.43) .59 (.49)

Note. Maximum vocabulary learning score was one in each cell.

Appendix C. Vocabulary Posttests

Appendix D. Tables

References

Barcroft, J. (2002). Semantic and structural elaboration in L2 lexical acquisition. Language Learning, 52, 323e363.Carrier, M., & Pashler, H. (1992). The inuence of retrieval on retention. Memory & Cognition, 20, 633e642.Cheng, H. H., Chang, H. H., Cheng, Y. S., & Gu, Y. S. (2002). An English reference word list for the department required English test. College Entrance Examination

Center.Goodman, K. S. (1982). Reading: a psycholinguistic guessing game. In F. V. Gollasch (Ed.), Language and literacy: The selected writings of K. S. Goodman (Vol. 1,

pp. 33e44) Boston, MA: Routledge & Kegan Paul.Groot, P. J. M. (2000). Computer assisted second language vocabulary acquisition. Language Learning and Technology, 4(1), 60e81.Hulstijn, J. H. (1992). Retention of inferred and given word meanings: experiments in incidental vocabulary learning. In P. J. L. Arnaud, & H. Bejoint (Eds.),

Vocabulary and applied linguistics (pp. 113e125). London: Macmillan.Hulstijn, J. (2001). Intentional and incidental second language vocabulary learning: a reappraisal of elaboration, rehearsal and automaticity. In P. Robinson

(Ed.), Cognition and second language instruction (pp. 258e286). Cambridge: Cambridge University Press.Hulstijn, J. H., Hollander, M., & Greidanus, T. (1996). Incidental vocabulary learning by advanced foreign language students: the inuence of marginal

glosses, dictionary use, and reoccurrence of unknown words. Modern Language Journal, 80, 327e339.Karpicke, J. D., & Roediger, H. L., III (2008). The critical importance of retrieval for learning. Science, 319, 966e968.Krashen, S. (1989). We acquire vocabulary and spelling by reading: additional evidence for the input hypothesis. Modern Language Journal, 73, 440e464.Laufer, B. (1997). The lexical plight in second language reading: words you dont know, words you think you know, and words you cant guess. In J. Coady, &

T. Huckin (Eds.), Second language vocabulary acquisition: A rationale for pedagogy (pp. 20e34). Cambridge: Cambridge University Press.Laufer, B., & Hulstijn, J. H. (2001). Incidental vocabulary acquisition in a second language: the construct of task-induced involvement. Applied Linguistics, 22,

1e26.

L.-l. Huang, C.-c. Lin / System 44 (2014) 127e136136Mondria, J. A. (2003). The effects of inferring, verifying and memorizing on the retention of L2 word meaning: an experimental comparison of themeaning-inferred method and the meaning-given method. Studies in Second Language Acquisition, 25, 473e499.

Nation, I. S. P. (2001). Learning vocabulary in another language. Cambridge, England: Cambridge University Press.Roediger, H. L., III, & Butler, A. C. (2011). The critical role of retrieval practice in long-term retention. Trends in Cognitive Science, 15(1), 20e27.Rott, S. (2005). Processing glosses: a qualitative exploration of how form-meaning connections are established and strengthened. Reading in a Foreign

Language, 17, 95e124.Rott, S. (2007). The effect of frequency of input-enhancements on word learning and text comprehension. Language Learning, 57, 165e199.Rott, S., & Williams, J. (2003). Making form-meaning connections while reading: a qualitative analysis of the effect of input and output tasks on word

learning. Reading in a Foreign Language, 15, 45e74.Rott, S., Williams, J., & Cameron, R. (2002). The effect of multiple-choice L1 glosses and input-output cycles on lexical acquisition and retention. Language

Teaching Research, 6, 183e222.Sokmen, A. J. (1997). Current trends in teaching second language vocabulary. In N. Schmitt, & M. McCarthy (Eds.), Vocabulary: Description, acquisition, and

pedagogy (pp. 237e257). Cambridge: Cambridge University Press.Watanabe, Y. (1997). Input, intake, and retention: effects of increased processing on incidental learning of foreign language vocabulary. Studies in Second

Language Acquisition, 19, 287e307.Wheeler, M. A., Ewers, M., & Buonanno, J. F. (2003). Different rates of forgetting following study versus test trials. Memory, 11, 571e580.

Three approaches to glossing and their effects on vocabulary learning1 Introduction2 Literature review3 The study3.1 Participants3.2 Target words and material3.3 Instruments3.3.1 Vocabulary pretest3.3.2 Vocabulary posttests

3.4 Procedures3.5 Data analysis

4 Results4.1 Effects of the intervention on vocabulary learning4.2 Relations between response accuracy and vocabulary learning

5 Discussion5.1 Full glossing condition5.2 Gloss-retrieval-gloss condition5.3 Inference-gloss-gloss condition

6 ConclusionAcknowledgmentsAppendix A A Vocabulary ChecklistAppendix B Reading Comprehension QuestionsAppendix C Vocabulary PosttestsAppendix D TablesReferences