The Canadian Modern Language Review / La revue canadienne des langues vivantes

Background

The study took place in the French-speaking province of Quebec. The participants were Grade 6 students enrolled in French elementary schools. All learners were taking part in special intensive ESL programs in which they devoted close to half of their regular school year (approximately 400 hours) to the learning of English. Intensive ESL programs in Quebec differ from traditional immersion models in that they are not content-based: The regular Grade 6 curriculum (science, math, French language arts, etc.) is completed in French in a condensed format that frees up time for ESL (for descriptions of intensive ESL see Collins, Halter, Lightbown, & Spada, 1999; Spada & Lightbown, 1989). The writing samples that formed the corpora for this study were collected as part of a larger longitudinal study designed to assess the relative effectiveness of different distributions of intensive instruction (Collins, White, & Springer, 2005). Student progress was monitored at regular intervals of 100 hours of instruction, using a variety of comprehension and production measures.

The measure of lexical richness used in the study is the lexical frequency profile or Vocabprofile; this computerized analysis technique was developed and tested by Laufer and Nation (1995). Vocabprofile software determines the proportions of running words (tokens) in a submitted text that can be found in each of the following lists: a list of the 1,000 most frequent word families of English, and the 1,001–2,000 most frequent list (both by West, 1953), and Coxhead's (2000) Academic Word List (a list of word families that occur frequently in academic writing). Words that do not occur on any of the previous frequency lists are categorized as 'off-list.' Results of the analysis of Marie-Eve's second story using Cobb's Vocabprofile software (an online version of Vocabprofile is available at www.lextutor.ca) are shown in Figure 1. [End Page 85]

Click for larger view
View full resolution

Figure 1.

Vocabprofile output for Marie-Eve's 400-hour narrative (see Vocabprofile page at www.lextutor.ca for details of these measures)

[End Page 86]

Figures in the bolded rows show that 81% of its 145 words are from the 1,000 most frequent families, 6% are from the 1,001–2,000 list, 1% is from the Academic Word List (AWL), and 13% are off-list. (In Marie-Eve's story, the off-list words were the items soccer and macho and names of characters such as Yannick.) Unlike measures based on type–token ratios that are suited to capturing the extent to which L2 learners use more varied lexis as their proficiency develops, Vocabprofile indicates the extent to which they use more 'difficult' (i.e., less frequent) vocabulary. In other words, rather than focusing on the growing numbers of different words learners use (which would normally be expected to increase over time), the measure is able to tell us something about the kinds of words they use (see Daller, Van Hout, & Treffers-Daller, 2003, for a discussion of this distinction). In addition to the proportions of words at each of four frequency levels, the output of the online version of Vocabprofile offers other details about the kinds of words that feature in a submitted text. As Figure 1 shows, information about the proportions of function, content, Greco-Latin-based and non-Greco-Latin–based words is available, along with counts of types, tokens, and families. However, our initial interest was in tracing learners' lexical development over time using Vocabprofile's frequency list feature.

This method of analysis offers several advantages in addition to providing pedagogically useful insights into the kinds of words learners use. First, since Vocabprofile draws on widely recognized frequency lists, it offers an objective way of defining what 'sophisticated' or 'unusual' lexis is in a way that allows for useful comparisons across a variety of learning and research contexts. Secondly, Vocabprofile has a track record of delineating lexical development in both cross-sectional (Laufer & Nation, 1995; Laufer & Paribakht, 2000); and longitudinal (Laufer, 1994, 1998) studies of learners of English. These studies found that a basic divide between proportions of frequent vocabulary in a learner production (words on lists of the 1,000 and 2,000 most frequent families) on one hand, and less frequent items (academic and off-list words) on the other, proved to be a reasonably reliable indicator of proficiency. The studies show that as learners advance, the proportion of lexis in the academic and off-list categories tends to increase.

Most of the studies have been conducted with adult university students at intermediate levels and beyond, but Laufer and Nation (1995) suggest that Vocabprofile may also be a useful way of assessing beginning learner productions. Vermeer's (2004) research provides support for the effectiveness of using lexical frequency profiling to assess young learner productions: Of the four measures of lexical richness he tested, a measure based on frequency lists derived from a [End Page 87] corpus of Dutch classroom input proved to be the most reliable in distinguishing between native and non-native oral productions. He argues that frequency-based approaches are particularly well suited to investigating child lexical development, given research that has found a relationship between the frequency of words in language input and vocabulary learning in children.

A previous study of 10 intensive ESL classrooms in Quebec used Vocabprofile to examine the characteristics of the classroom input that learners were exposed to (Meara, Lightbown, & Halter, 1997). The study determined that the overwhelming proportion of word types in transcripts of classroom input (about 96%) could be found on the list of the 2,000 most frequent word families of English (West, 1953); these results were highly consistent across the 10 classrooms investigated. Initially, the researchers interpreted the findings as evidence that the input the learners received was lexically impoverished; later they revised this view, speculating that input consisting largely of items at the 2,000 frequency level could be considered lexically rich, if, as they surmised, the young Quebec learners' vocabulary knowledge was very basic. Our investigation builds on this work by specifying more closely what such learners know, and how this lexical knowledge develops. More specifically, we focus on the extent to which learners in intensive classrooms are able to transform the large amounts of 2,000-level vocabulary they are exposed to into productive knowledge. In this, we took up Laufer and Nation's (1995) suggestion that the divide between 1,000-level vocabulary and words not in this category is likely to be a useful indicator of development in beginning learners' written productions. Thus, a text produced by a learner after 100 hours of instruction might contain a very high proportion of words from the list of the 1,000 most frequent word families and few from the 1,001–2,000 most frequent, AWL, and off-list categories, but the proportion of items in the later categories could be expected to increase over the course of 400 hours of intensive instruction. Our initial research question was as follows: Do texts show evidence of increased lexical richness (as indicated by decreasing proportions of 1,000-level words) over 400 hours of instruction?

Methodology

Analyses and results

To answer the question of whether the apparent fluency increases also entailed differences in ability to use more infrequent words over time, we turn to the results of the lexical frequency analysis.

Lexical richness of the narratives

Click for larger view
View full resolution

Table 1.

Corpus size in numbers of words at four measurement points (210 writers)

Click for larger view
View full resolution

Table 2.

Percentages of words in four frequency levels in learner narrative corpora produced at intervals of 100 hours of instruction

The Vocabprofile output for the four corpora is shown in Table 2. The first row shows percentages of words from the list of the 1,000 most frequent families (K1). Here we see that in contrast to what was hypothesized, the proportions of words in this most basic frequency category did not decrease over time. In fact, there is an increase from 85% coverage at 100 hours to 91% at 200 hours; after this point, proportions of words in this category appear to stabilize at around 90%. It is rather surprising that proportions of words from the 1,001–2,000 [End Page 91] level (K2) – the zone where growth might be expected for these beginning learners – proportions decrease over time (with a minor disruption to the pattern at 300 hours). The off-list proportions (the category where infrequent items like macho can be found) shown in the last row remain stable at about 3% throughout. The proportion of items on the AWL is negligible in all four corpora. In sum, there is no indication that the narratives used increasingly more lexis from the lower-frequency levels over time; in fact, it appears that proportions decreased (K2) or remained unchanged (off-list).

To determine whether the differences over time in proportions of 1,000-level words were significantly different, we used the following procedure: Each corpus was divided into 42 shorter segments, each consisting of narratives produced by five randomly selected learners. Each segment was then submitted separately to Vocabprofile for analysis; the result was sets of coverage percentages for which means could be calculated and compared. The five-part segments were matched across the four corpora so that segment 1 of each corpus was made up of narratives produced by writers A, B, C, D, and E, segment 2 by writers F, G, H, I, and J, and so on. Each segment was then submitted separately to Vocabprofile for analysis. The rationale for dividing the corpora into segments produced by five writers was based on the need for longer texts. Laufer and Nation (1995) have determined that 200 words is the minimum length needed for reliable Vocabprofile measurement, but almost all of the individual narratives were well under this length, so treating the narratives individually was not an option. As Table 2 shows, the shortest corpus contained narratives produced after 100 hours of instruction. In this corpus, the five-part segments ranged in length from 243 to 543 words; mean length was 358 words (SD = 75); thus all segments in the four corpora were well over the 200-word criterion. Lengths in the subsequent corpora were consistently larger (see Table 3). Experimentation with samples of varying lengths revealed that profiles of compiled segments were reasonably stable as long as they were 200 words or longer, so comparisons of texts of varying sizes seemed justified. ³

Click for larger view
View full resolution

Table 3.

Sizes of five-narrative segments in numbers of words

[End Page 92]

The proportions of 1,000-level words in the productions across the four time intervals were then compared using a one-way repeated measures ANOVA. The results are displayed in Table 4. Means for proportions of 1,000-level words in the 42 segments (see the first column in the lower half of Table 4) closely resemble the proportions determined by analyzing the corpora in their entirety (see the first row of Table 2). There was a significant difference in the 1,000-level proportions over time, and post-hoc Tukey's HSD for related samples revealed that the difference was between the initial writing time (after 100 hours of instruction) and the subsequent three testing times. ⁴ Other differences also proved to be statistically significant (200 vs. 300 hours, and 300 vs. 400 hours), but the largest differences over time are between earlier (100-hour) and later (200-, 300-, and 400-hour) performance.

Discussion of initial results

These findings show that young, beginning learners of English do not incorporate increasing amounts of less frequent vocabulary into their written productions over the course of 400 hours of intensive instruction.

Click for larger view
View full resolution

Table 4.

Repeated measures ANOVA: K1 coverage

However, we were hardly prepared to conclude that no lexical development had occurred over the 400 hours. Therefore we decided to re-examine the corpus data from a different perspective, looking more carefully at changes within use of frequent words, using tools available at the online Vocabprofile site (www.lextutor.ca). The decision to do so [End Page 93] was based on a number of characteristics of the writing that were observed during the transcription of the texts. First we noticed that the learners relied on L1 to fill lexical gaps in the L2, especially in the earlier corpora. The second indicator that we considered was the extent to which these young francophones relied on words that are cognate with French in their English productions. English contains thousands of words of Romance language origin, and Marie-Eve uses a number of them (e.g., surprised and stomach) appropriately in her story of a schoolyard fight. But there are also instances where French cognates are less suited to the register of simple stories told by 12-year-olds. For instance, one participant wrote, 'He is thinking about the sensation that the boy can have in his heart the next day.' A 12-year-old native speaker of English telling this story would likely speak of the boy's feelings instead. Such cases of using infrequent Romance-based words like sensation (an off-list word) in places where more frequent Anglo-Saxon words like feelings (1,000-level) seem more appropriate recurred in the data: Characters often respond rather than answer, supervise rather than watch over, and reject rather than say no.

Third, we realized the limitations of an approach that measures lexical development in terms of increasing use of words beyond the 1,000-most-frequent level. Changes in the ways these early beginner writers used very frequent English words might also indicate development. For instance, if the data showed that the learners incorporated an increasing variety of 1,000-level lexis into their narratives as the intensive course progressed, this would also be a clear evidence of lexical growth (i.e., increase in productive vocabulary size). We were also interested in the extent to which learners used morphological forms of frequent words in early and late productions. Here the focus is not on greater numbers of words but on ability to use known words in more complex ways. We had the impression that the later corpora contained more instances of inflected forms (e.g., played and playing vs. play) and also more derived forms (punishment vs. punish), so we set out to investigate this systematically across the staged corpora.

In sum, in our reappraisal of the data, we investigated the following characteristics of the staged corpora: reliance on French words, reliance on Romance-based lexis, variety of word families within the 1,000 level, and morphological variety of items within word families. We predicted that over the course of instruction, use of French itself as well as the use of French cognates would decrease. We expected that learners would also use a wider range of basic vocabulary and that there would be increased morphological variation. [End Page 94]

Re-examining the staged corpora

To evaluate the extent to which the writers used French words to fill gaps in narratives, we returned to earlier versions of the corpora from which French words and proper nouns had not yet been removed. We then relied on the list function of the Vocabprofile software. In addition to calculating the frequency percentages shown in the sample output in Figure 1, online Vocabprofile also provides lists of all tokens found in a particular frequency category. All French words that had been tagged were automatically placed in the list of off-list tokens (along with English words not found on the frequency lists). A sample segment of output from the 400-hour corpus is shown below:

charles cherchefrench cherchefrench chew chicanaitfrench chicanefrench chicanefrench chicanefrench chicanefrench chili china chinese chinese choquefrench chum cicanefrench cindy cindy cindy cindy circuitfrench classroom classroom coeurfrench colerefrench colletfrench consolerfrench contest coufrench coupfrench coupfrench coupfrench

Our research team printed out the lists of off-list tokens and counted numbers of tokens tagged as French for each of the eight corpora. These totals were then used to calculate proportions of the corpora that were accounted for by French words.

To assess levels of Romance cognate use, we used the cognate index available on the Vocabprofile site (based on wordlists by Bauer & Cobb, 2000) to evaluate the corpora from which French words and proper nouns had been removed. The cognate index measures the proportions of Romance-based families on the 1,000, 2,000, and AWL lists that a submitted text contains. In Figure 1, which shows the output for Marie-Eve's story, this information appears in the last line, where it can be seen that 11% of the lexis this learner used was of Romance-language (usually Greco-Latin) origin. To assess the numbers of different 1,000-level word families used in the four corpora (French and proper names removed), we consulted Vocabprofile output for families (see column 2 in Figure 1). In this instance, we were interested in numbers of families in the K1 (1,000-level) listing (shown in the second row). To determine the number of families, Vocabprofile matches words in a submitted text to lemmatized lists. These are lists made up of headwords (e.g., believe) along with their inflected and simple derived forms (e.g., belief, beliefs, believed, believes, believing, and believer). ⁵ Thus if two writers used the [End Page 95] form believed in a story and three used believing, this would register as a single family in the analysis of these five stories.

We were also interested in the extent to which morphological variants such as belief, believed, believing, etc., occurred in the corpora. Vocabprofile uses the lemmatized lists described above to calculate the mean number of types per family for all 1,000-, 2,000-, and AWL-level words in a submitted text. ⁶ In our hypothetical example of five stories that featured two types (believed and believing) from one family (believe), the software would assign a value of 2 for the family believe. Each of the other words that occurred in the five sentences would be evaluated in a similar fashion. The mean types-per-family figure for Marie-Eve's text can be seen in the fourth line from the bottom in the sample of her Vocabprofile output shown in Figure 1. The figure of 1.13 indicates that few of the 55 word families she used appeared in more than one of its possible inflected or derived forms.

Results of the re-examination

Reliance on French

As expected, the need for French to compensate for lexical gaps in English decreased considerably over time. Results in percentages of the total word count that the French tokens accounted for are shown in Table 5. After 100 hours of instruction nearly 1 word in 16 (~6% of the total corpus) was French; after 400 hours, this proportion has been reduced dramatically to approximately 1 word in 50 (~2%).

Click for larger view
View full resolution

Table 5.

Percentages of French words in learner corpora at four measurement points

The kinds of French words that learners needed also changed over time. Analyses of the nature of the gaps that were filled by L1 words provide further insights into the learners' lexical development. As might be expected, the most commonly supplied French words were relatively low-frequency words needed to describe a particular feature of the picture prompt (e.g., the French translation equivalents of pregnant in a cat with kittens picture, needle in a hospital picture prompt, and punch in a schoolyard fight scene). An interesting exception to the finding that most insertions were topic-specific occurred in the 100-hour narratives: The most frequently supplied French word in the corpus was one with [End Page 96] an important grammatical function, illustrated in the following sentences taken from three different learners' texts:

Amelie is very happy and sa grandmother too

The little girl return in sa house

She kiss sa mother.

Learners needed French in these cases to render the possessive determiner her. ⁷ Possessive determiners in English are an attested challenge for Romance speakers (White, 1998; White, Muñoz, & Collins, in press), and the reliance on L1 to render this relationship clearly signals a grammatical need. In addition, 70% (25 of 36) of the unknown content words rendered in French at 100 hours were verbs; most have English translation equivalents that are high-frequency (1,000-level) words of Germanic origin, such as ask, bring/take, come/go back, give, find, keep, leave, look for, and say.

Reliance on cognates

Click for larger view
View full resolution

Table 6.

Vocabprofile output for three measures at four time intervals for entire corpora and segments (means)

The French cognate indices based on Vocabprofile output are shown in the first row of Table 6. The first line in this row shows the results of the analyses of whole corpora while figures in the second line report means based on analyses of segments. The two sets of figures resemble each other closely. Analysis of the segment means using a one-way repeated measures ANOVA (and post-hoc Tukey's HSD for related samples comparisons) revealed that these changes were all significant (see Table 7). Generally, reliance on French cognates decreased over time (with a minor exception at 300 hours) – as predicted. The most pronounced difference is between the proportion accounted for by words of [End Page 97] Romance language origin at 100 hours and all of the subsequent measures.

Click for larger view
View full resolution

Table 7.

Repeated measures ANOVA: Greco-Latin coverage

Types per family

The types-per-family indices assessed morphological variation (i.e., use of forms such as believes, believing, believer, etc.) and are displayed in the bottom row of Table 6. A repeated measures ANOVA (see Table 9) showed that the change over time was significant; the post-hoc Tukey's HSD indicated that the differences were between the first half of the program (the 100- and 200-hour testing times) and the second (the 300-and 400-hour times). In other words, learners' lexical knowledge developed in tandem with their grammatical knowledge, as they demonstrated an increased ability to render different forms of the same word to express different types of grammatical relationships. [End Page 98]

Click for larger view
View full resolution

Table 8.

Repeated measures ANOVA: Number of K1 families

Click for larger view
View full resolution

Table 9.

Repeated measures ANOVA: Type per family

Discussion

The exploration of the staged learner corpora using features of the online version of Vocabprofile identified patterns of lexical development in the young French-speaking learners of English. However, the main pattern was different from what was expected: Results over time showed that instead of ever-greater proportions of less frequent words, the learners used greater proportions of very frequent (1,000-level) [End Page 99] words. This distinction was most pronounced in the comparison between narratives produced before and after 200 hours of intensive ESL instruction. The proportion of 1,000-level words amounted to about 85% after 100 hours of instruction, rose to about 90% after 200 hours, and remained at this approximate level in the subsequent corpora. Not surprisingly, this is a higher proportion than occurs in texts produced by native speakers of English. According to figures reported in Nation (2001), the proportion of 1,000-level words in native speaker literary texts amounts just over 82%. Vocabprofile analysis of a graded reader simplified for intermediate learners of English (Nothing but the Truth, Kershaw, 1999) identified the same figure (82%) for proportions of 1,000-level words.

At first glance, the increased proportions of high-frequency words found in this study suggested a lack of progress. This impression is mistaken, as counts of French words and three other measures available through Vocabprofile revealed. Reliance on L1 vocabulary decreased over time, with the percentage of French tokens reduced to a little over 2% on the final measure. The cognate indices pointed to a reduction in use of Romance-based English words over time, from around 18% on the first measurement to around 13% on the subsequent measures. The pattern of early high use of cognates on the first measurement followed by lower levels from 200 hours onwards is a mirror image of the 1,000-level findings identified by the lexical frequency profiling, and may explain the increase in 1,000-level words over time. Evidence for this interpretation is illustrated by the sentences below, which were taken from the 100-hour corpus:

The father search in the forest but don't have the animal.

The grandmother with the boy search your cat.

Cats are not complicated in the house.

After an eternity of reflection, I called my grandma. She can help me.

All of the italicized cognates (search, complicated, and reflection) appear on the 2,000-level list; their non-cognate equivalents look for, hard, and thinking are all 1,000-level items. Thus use of cognates may account for the relatively high proportions of 2,000-level words on the first measure (and the corresponding low 1,000-level proportions). But as students' learning progressed, their productive lexicons appear to have featured more of the non-cognate 1,000-level words that are more appropriate to the narratives. That is, characters increasingly feel rather than have sensations and answer rather than respond. ⁸ The increased variety of 1,000-level words used in the staged corpora and the increased types-per-family [End Page 100] indices also indicate progress. Not only did the young learners use more word families over time, they also produced a wider range of inflected and derived forms of the words within these families.

The initial finding that lexical richness did not increase over time (as indicated by no apparent increase of 2,000-level+ words over time) raises the question of whether the use of a measure based on frequency lists derived from analyses of adult native-speaker corpora was appropriate in this study of child learners. While such corpora provide a baseline against which child language may be compared, frequency lists based upon a corpus representing child native-speaker language would allow for more fine-grained age-equivalent comparisons. It seems likely that lists based on adult language are an inaccurate reflection of frequent and infrequent lexis in the L2 input the participants were exposed to (Snellings, Van Gelderen, & De Glopper, 2002); the usefulness of the AWL category seems particularly questionable, and it is not surprising that contributions to this zone were negligible (whether an age-appropriate AWL would yield different results merits future consideration). The finding that the participants used a greater variety of 1,000-level words over time hints at increases in lexical richness that more nuanced and appropriate frequency lists might have been able to identify. Evidence in support of this view comes from Vermeer's study (2004) in which he found evidence of increased use of lower-frequency lexis in more proficient child learners in his investigation of the lexical profiles of L1 and L2 Dutch productions. This research relied on frequency lists derived from a corpus of age-appropriate classroom input and distinguished between nine frequency levels.

The study made use of a new measure, the types-per-family index. To our knowledge, the index has not been used in previous research – at least not in the computerized format available through online Vocabprofile. ⁹ The finding that the measure detected an increase over time in the variety of morphological forms of words the learners were able to produce inspires confidence in its validity, especially since it occurred in tandem with another indicator (increase in number of 1,000-level families). Nicolae (2003) noted that instruments used in vocabulary research too often neglect the syntactic aspects of words; this measure may prove useful in addressing the imbalance. Further experiments to test its validity and reliability in other contexts are warranted.

It is clear that the other measures used in conjunction with Vocabprofile, including a measure of lexical variation, were needed to provide a more complete picture of the learners' progress. This finding is consistent with the combined approach advocated by Daller et al. in their 2003 study. They used a frequency-based measure in combination [End Page 101] with a measure of lexical variation to assess lexical richness. In the present study, the cognate index – numbers of 1,000-level families and types per family within this level – allowed us to interpret change over time among young low-level L2 learners of English. Overall, the study has provided evidence in support of the importance of the 1,000-level band for understanding L2 vocabulary development among lower-level learners. However, the 1,000-level band may be relevant for explaining aspects of more advanced learners' vocabulary as well. Edwards and Collins (2006) employ mathematical modelling tools (based on Zipf's Law, following Meara, 2005) to demonstrate that analyses of the proportions of words within the 1,000-level band yield productive vocabulary size estimates for L2 groups at a range of vocabulary levels.

Implications for learning and teaching

Results based on a variety of measures showed that intensive instruction contributed to substantial changes in the young ESL learners' productive vocabulary knowledge. We have argued that the increased use of frequent vocabulary is a positive development, if, as we have seen, it goes hand in hand with using vocabulary that is also more register-appropriate, diverse, and morphologically complex. However, lexical development also implies control over more unusual words and raises the following question: How many more hours of instruction might be needed to increase the use of vocabulary beyond the 1,000-most-frequent level? Is there a role for more focused vocabulary instruction to facilitate the process? An investigation of lexical richness by Granger (1996) suggests that some pedagogical intervention may indeed be warranted. Her study of essays produced by advanced adult francophone learners and native speakers of English showed that in comparison to the native speakers, the ESL learners tended to overuse high-frequency English words of largely Germanic origin and to underuse lower-frequency words that are cognate with French. Our findings with beginner-level francophones suggest that there may be an initial stage of overuse (or at least inappropriate use) of these lower-frequency cognates to compensate for lexical gaps; but they are replaced by higher-frequency words (often of Germanic origin). While this change resulted in word choices that were more appropriate for the meanings that the young learners wished to express, Granger's study demonstrates that moving beyond these high-frequency words may present a persistent challenge, even when learners have an apparent cognate advantage.

Investigations of lexical richness among learners from different L1 backgrounds and at different levels of ESL proficiency would allow us [End Page 102] to obtain a more complete picture of the development of frequent, infrequent, and cognate vocabulary knowledge. Longitudinal studies of learners from various L1 backgrounds (Ortega & Iberri-Shea, 2005) would also help chart the developmental path. There is also a need for a better understanding of how L2 learners perceive the semantic overlap between their first and second languages (Kellerman, 1986; Ringbom, 1986), and of the academic words that young L2 learners need to know to be able to read and produce age-appropriate texts. Cummins (1984) has pointed to the difficulties school-aged L2 speakers encounter when they lack sufficient proficiency in the target language to handle academic content. More research along the lines of Gardner's (2004) comparisons of the vocabulary content of different types of reading materials for children could inform the creation of an eventual AWL for young learners. Such research may be able to inform responsive, needs-based instructional interventions so that L2 learning can become more effective. With these goals in mind, we hope that the baselines for beginning-learner productions outlined in this study will serve as a useful basis for comparisons in future investigations and a foundation for improving instruction.