Strategies Typically Developing Writers Use for Translating Thought into the Next Sentence and Evolving Text: Implications for Assessment and Instruction

Three new approaches to writing assessment are introduced. First, strategies for generating the very next sentence are assessed in reference to the local level as well as the evolving text level of composing in progress. Second, strategies for translating thought into written language are coded with transcription (spelling) skill—low, average, or high—held constant. Third, instead of describing composing skill in reference to a single normed score for age or grade in a standardization sample at a static time in development, translation is studied longitudinally when children are in grades 1, 3, and 5 (ages 6, 8, 10) or grades 3, 5, and 7 (ages 8, 10, 12). Applications of the results are discussed for assessment and instruction grounded in levels and generativity of written language and normal variation in typically developing writers.

contribution of the interdisciplinary approach to assessing written language is the increasing evidence that not only syntax is generative (a set of finite items can be used to create an infinite number of constructions in translation into written language, Myhill, 2009) but also so is text at the genre level during the written translation process (Boscolo, Cisotto, & Cortiana, 2014;Olinghouse & Graham, 2009;Olinghouse & Wilson, 2013).
Another contribution of the interdisciplinary approach is the increasing recognition that translation operates differently in conversational registrar of spoken language than in the academic registrar of written language (Silliman & Scott, 2009), even though both draw on different levels of language-word, sentence, and text/discourse levels-but for different purposes. In the case of oral language exchange, the goal is to create cohesion among words within and across language production during turn-taking, which occurs frequently close in time. In the case of written language, the goal is to create coherence among sentences at the genre-constrained discourse level produced across longer time intervals than in conversation by a writer for a non-present audience unlike conversation when the other participant/s are present (see Halliday & Hasan, 1976;Silliman & Scott, 2009).
Normal variation-individual differences within and across developing writers -Cross-sectional research has shown that developing writers exhibit intra-individual differences in their relative abilities for writing at the word (spelling and vocabulary), sentence, and text levels (Berninger, Mizokawa, Bragg, Cartwright, & Yates, 1994;Whitaker, Berninger, Johnston, & Swanson, 1994) and use a variety of different text level structures in composing that change across development (Berninger, Fuller, & Whitaker, 1996;Fuller, 1995). However, relatively little research has examined the relationships among strategies used to compose the very next sentence (Level 1 translation) and to compose the evolving overall discourse structure (Level II translation). Most research on self-regulated writing strategies has focused on genre-level writing, for example, for narratives and persuasive essays (for review, see Graham, McKeown, Kiuhara, & Harris, 2012;Graham & Perrin, 2007a, 2007b, 2007cHarris, Graham, MacArthur, Reid, & Mason, 2011;MacArthur, Graham, & Fitzgerald, 2006;Olinghouse, & Wilson, 2013;Troia, 2009).
Although some research has focused on combining two sentences to express a single idea (Saddler & Graham, 2005) or the nature of idea expression within syntax (Myhill, 2009), relatively little is known, however, about intra-and inter-individual differences for translation for purpose of constructing the very next sentence, which may occur independently of genre, compared to translating at the evolving text level that characterizes specific genres. Yet, struggling writers often are stymied by how to write the very next sentence and sustain the writing process long enough to compose a text of a specific genre.
Limitations of current assessment approaches for linking results to instruction-Often normed measures of writing for age or grade peers based on a national standardization sample collected at one point in time are used to assess writing skills. Although research is showing that writing development is a dynamically changing process (Olinghouse & Wilson, 2013;Troia, 2009), normed measures are not available that inform assessment of the normal variation within and among individuals over time. Typically single writing skills are assessed, such as composing ability, without controlling for level of development in other writing skills, such as spelling ability, which may constrain the target writing skills being assessed. Likewise, annual tests yoked to national or state standards that use cut-off points do not provide relevant information about normal variation within a grade or age level or across grade or age levels. Such information is relevant to tailoring instruction to individual differences in developing writers.

Research Aim
For these reasons, the research aim of the current study was to evaluate whether a coding scheme could be developed that accounted for all observed Level I translating strategies for generating the next sentence and for Level II translating strategies for tying sentences together in the evolving discourse in typically developing writers during elementary and first two grades of middle school (ages 6 to 12). For this research aim, genre was kept constant across specific grade levels in this longitudinal study so that development of translating strategies could be compared for genres kept constant across grade levels as explained in the methods.

Participants
Through a data sharing plan, compositions were analyzed from a previously published study of developmental change in spelling and related language skills among 60 children who differed as to whether they were reliably good (10 girls, 10 boys), average (10 girls, 10 boys), or poor (10 girls, 10 boys) spellers during a five-year longitudinal study of two cohorts that overlapped in grades three to five: cohort 1 had been assessed annually at the university from grades one to five and cohort 2 had been assessed annually at the university from grades three to seven. The mean ages of children in their first year of the study were 6.8 years for cohort 1 and 8.8 years for cohort 2.
Translation strategies were assessed in narrative composing in grades one, three, and five or in expository in grades two and four in cohort 1; or narrative in grades three, five, and seven or expository in grades four and six in cohort 2. Standard scores for the groups were at or below 95 for poor spellers, between 105 and 112 for average spellers, and 122 or above for good spellers on a nationally normed measure of spelling dictation. For additional details about sample ascertainment, test measures, and research design for studying spelling development and related linguistic skills, see Garcia et al. (2010); that study did not analyze compositions as in the current study and did not examine development of Level 1 or Level 2 Translation Strategies.

Procedures
In each year of the study, children produced writing in specific genres. Narrative prompts were "One day at school a surprising or funny thing happened" (years 1, 3, 5) and "One weekend at home a funny or surprising thing happened" (years 3, 5). Expository prompts were "Explain what a computer is and what it does to someone who has never seen or used one" and "Explain what a robot is and can do to someone who has never seen one or used one" (years 2, 4). Because the nine coded translation strategies did not differ as a function of pen or keyboard production, writing samples were pooled across these modes for this study of translation strategies. However, because of scheduling conflicts and other difficulties five cohort 1 and six cohort 2 children were each missing one writing sample.
The coding units were defined as sentences or main clauses with coordinating clauses (e.g. marked by coordinating conjunctions such as and, but, so, or) or subordinate clauses (e.g. marked by subordinate conjunctions such as after, even though, although, before, because, before, if). Children's writings often lacked ending punctuations and sometimes only conjuctions such as then or and served as markers at the beginning of main clauses.
Although a single code often characterized a sentence, in some cases multiple codes applied. For example, consider IA12 state a wish and IA16 If … then statement in this coded next sentence: Even if I can't have my dream job I hope I can at least have a job that would keep me on my feet.
Two coders developed and practiced applying the coding scheme to 30 writing samples. Each independently coded 10 narrative samples and 10 expository samples. Interrater reliability agreement ranged from .80 to .86 (average .83). The remaining writing samples in the longitudinal sample were then coded independently by both raters and codes compared. Overall, there was 68% agreement across coders for final combined Level I and Level II coding. For the slightly less than a third on which there was disagreement, all were discussed until agreement was reached. In the process it became increasingly apparent that developing writers differed in the extent to which they 1) use more than one translation strategy in generating the very next sentence; and 2) use translation strategies at both Level I and Level II as they write the next sentence. Interrater differences were largely related to coding the use of multiple strategies in parallel. All initial disagreements were readily resolved. See Appendix for final Level I and Level II Coding Schemes that accounted for all observed translation strategies in this study. Overall, all next sentences showed evidence of at least one Level I translation category, but some next sentences showed evidence of use of more than one Level I code or both a Level I and Level II code.

Results
For this mixed methods study, both descriptive statistics and inferential statistics are reported. The descriptive results specific for grade level and spelling transcription ability indicate which translation strategies students used and can inform instructional strategies that are grade-and genre-yoked that teachers might use in their instructional program for composing. The inferential statistics support generalization of results related to translation strategies across the participants in the current sample.

Descriptive Statistics
Two tables reporting descriptive statistics for frequencies of each coded category in each genre for all participants within a spelling ability group at each grade level are available from the authors. Because of space limitations, only a summary of the main patterns are reported here, first organized by spelling ability and grade levels in which the writing genre was the same (first narrative and then expository). Level 1 translation strategies are presented first followed by Level II translation strategies employing the same organizational structure (spelling ability by grade by genre).
Poor spellers-To summarize for narrative genre, for Level I translation strategies for the very next written sentences, with the rare exception of one first grader on two occasions, translation resulted in a syntactically complete sentence; and all translation reflected 4 strategies-stating an opinion or belief, describing observable behavior, telling the next event, or defining what something is. By third grade, 17 translation categories were observed, but the most frequent were describing observable behavior and telling the next event. In fifth grade, 18 translation strategies were observed (3 new ones-telling next step in procedure, making a comparison, and making an editorial comment-and 3 used in third grade were not used in fifth grade-stating a wish, a goal/plan, or if … then conditions). Again, in fifth grade, the most frequently used translation strategies were describe observable behavior and tell next event. In seventh grade, 15 translation strategies were observed, all of which had appeared in earlier grades, except pose question for reader. By seventh grade, of the 15 translation strategies observed, all had been used in earlier grades except pose question for reader; the most frequently used translation codes included not only describe observable behavior or tell next event, as in earlier grades, but also describe by painting a picture with words or describing a state of mind or feelings. Indirect dialogue was also used more frequently than in earlier grades. Overall, the number of Level II translation strategies within and across levels of language increased from first to third grade and thereafter. However, in general, the poor spellers did not tend to apply multiple translation strategies at either Level I or Level II.
To summarize for expository genre, in second grade of the 7 translation strategies observed in poor spellers, the most frequent was state an opinion or belief. In fourth grade 17 translation strategies were observed. Most frequent were state an opinion or belief, describe observable behavior, describe by painting a picture in words, and describe function or use. In sixth grade, 14 translation strategies were observed. Most frequent were describe function or use, state an opinion or belief, and describe by painting a picture with words. The most frequent Level I strategy for within level connections that was used across grade levels was use of pronouns. Few Level II connections across sentences were used for expository genre.
Average spellers-To summarize for narrative genre, in addition to one presyntactic written language production, average spellers in first grade showed evidence of using 10 Level I translation strategies: stating an opinion or belief, describe by painting picture with words, describe state of mind/feelings, describe observable behavior, tell next event, state a goal or plan, state an outcome, provide an explanation, qualify a prior statement, and create indirect dialogue. However, the most frequently used translation category described an observable behavior. In addition to these, other translation strategies observed in average spellers in third grade included state a fact, state an opinion or belief, tell prior event, give example, make a comparison, make a statement about time or space, repeat part of prior text with substitution, paraphrase prior text, create indirect dialogue, and issue direct or indirect command for reader; but state a goal or plan, as had been used by first graders, was not used. At the third grade level, the most frequently used translation categories were describing by painting a picture with words, describing observable behavior, and telling next event. At fifth grade all the translation codes used by third graders, except give an example, make a comparison, repeat part of text with substitution, create dialogue among characters, and issue command for readers, were observed; but new ones also occurred-state a goal or plan, make a prediction, state if … then conditions, and make editorial comment for reader. At fifth grade, the most frequent translation strategies were describe observable behavior and tell next step or procedure, qualify a prior statement, and describe state of mind or feelings.
At seventh grade, seventeen translation strategies were observed, all of which had been observed in third and/or fifth grade, except define what something is and state a wish. The most frequent translation strategies in seventh grade included describe observable behavior and tell next event. Overall, some translation strategies were used frequently across grade levels and some emerged at specific grade levels.
To summarize for expository genre, in second grade average spellers used 6 translation strategies. Most frequent were state opinion or belief and describe function or use. In fourth grade 18 were observed. In fourth grade fourteen translation strategies were observed with the most frequent being state an opinion or belief, describe function or use, and describe by painting a picture in words. In sixth grade 9 translation strategies were observed. Most frequent were describe function or use, state an opinion or belief, and describe by painting a picture with words. As had been the case with poor spellers, the most frequent Level I within-level connections were use of pronouns across grade levels. Overall, average spellers used few Level II connections across sentences for expository genre.
Superior spellers-To summarize for narrative genre, in first grade 12 translation strategies were observed and no presynaptic written language productions. The most frequently observed translation strategy was tell the next event. In third grade 18 translation strategies were observed. Of these, describe observable behavior and tell next event were most frequent. In fifth grade 22 translation strategies were observed. All had been observed in third grade except give examples, state a wish, state if … then conditions, pretend or imagine, and pose question for reader. Again the most frequent translation strategies included describe observable behavior and tell next event. In seventh grade 17 translation strategies were observed of which the most frequent were describe observable behavior and tell next event. None emerged that had not been observed at an earlier grade. Overall, both Level I and Level II translation strategies were used.
To summarize for expository genre, in second grade superior spellers used 13 translation strategies. Most frequent were state an opinion or belief, describe a function or use, and Jones and Berninger Page 6 Open J Mod Linguist. Author manuscript; available in PMC 2017 January 24. describe by painting a picture with words. In fourth grade they used 15 translation strategies.
The same three were most frequent as had been in second grade. In sixth grade they used 11 translation strategies; describe observable behavior and describe state of mind or feelings were most frequent. Overall, a variety of Level II strategies within and across sentences were observed.

Inferential Statistics
Analyses of variance (ANOVA) were used to identify main effects and interactions with grade and genre for specific coded translation strategies that occurred with some frequency. For F and p values see Table 1 for cohort 1-narrative, Table 2 for cohort 2-narrative, Table 3 for cohort 1 expository, and Table 4 for cohort 2 expository. The means for significant effects are reported in the text that follows. ANOVAs were followed by post hoc analyses with Bonferroni adjustment comparing change across just two grade levels at a time. Only significant results of those post hoc analyses are reported in the text that follows. Cohort 1-Interaction of grade and spelling on Level I translation for narrative genre-Only one time-by-spelling ability interaction for describe by painting a picture in words was significant in cohort 1, but post hoc tests for the simple effect of group also failed to detect any significant differences between spelling ability at any grade level. 20, SD = .42) spellers, but the difference was only significant between the poor and average spellers.

Cohort 2-Main effect of grade on Level I translation for narrative genre-
Results depended on whether the analyses began in grade three rather than grade one and are somewhat different for Level I translation strategies in cohort 2 than for cohort 1. Describe a state of mind or feelings increased from grade three (M = .23, SD = .51) to five (M = .73, SD = .78) to seven (M = .99. SD = 1.06), but was significant only for grade three to seven. Describe observable behavior increased from grades three (M = 1.44, SD = .90) to five (M = 2.49, SD = 1.42), and then decreased slightly by grade seven (M = 2.32, SD = 1.05), but the decrease was not significant; however, increased use was significant between grades three to five and three to seven. Frequency for telling the next event rose from grade three (M = 2.53, SD = 2.49) to five (M = 3.95, SD = 2.91) to seven (M = 4.28, SD = 3.29); but the difference was significant only for grades three to seven.

Cohort 2-Main effect for spelling and interaction with grade on
Level I for narrative genre-For this cohort no main effects occurred for spelling ability or interaction for grade-by-spelling ability for Level I.
001. Use of all translation strategies improved over time. Overall, Level I (next sentence) translation strategies were used more often than Level II (evolving discourse) translation strategies. In general good spellers outperformed average spellers who outperformed poor spellers. There were no significant interactions.

Cohort 1-Main Effects across Cohort, Grade, Translation Levels, and
Spelling Ability for Expository-For this Cohort (grades two and four for essay) there were two significant main effects and two significant interactions: grade F (1, 25) = 41.40, p < .001, translation strategies, F (1, 25) = 100.50, p < .001, translation strategies-by-grade, F (1, 25) = 17.37, p < .001, and grade-by-spelling, F (1, 25) = 4.39, p = .023. Overall, the use of each translation code for essay writing increased from second to fourth grade, but in second grade no Level IIB translation strategies (genre-specific) were used. In fourth grade Level IIB translation strategies were used. In second grade, the superior spellers outperformed either the poor or average spellers, but in fourth grade, the superior spellers outperformed the average spellers who outperformed the poor spellers at both Level I (next sentence) and Level II (evolving text).
006. The interaction of level and grade indicated that Level I and Level IIB (genre-specific) translation strategies increased more than Level IIA (within level cohesion). The interaction with spelling ability indicated that those with greater spelling ability increased their use of level II translation strategies more than those with lower spelling ability. For this cohort (grades four and six for essays), there were main effects for grade, F (1, 25) = 7.89, p = .01 and translation strategy F (2, 25) = 107.85, p < .001, and a significant interaction between grade and level of translation strategy, F (2, 50) = 3.99, p = .025. Overall, across these grades, use of all the translation strategies increased, in general Level I translation strategies were used more often than Level II, but the relative use of Level IIB, compared to Level IIA, increased.

Developmental Changes in Translation Strategies
Is cognitive-linguistic translation generative?-A variety of Level I and Level II translation strategies were observed in the narrative and expository compositions, consistent with a generative model of cognitive to linguistic translation (see introduction). However, the nature of the generativity was dynamic, in that the pattern of translation strategies was variable across grade levels, but also constant, in that some comparable translation strategies were used at all or most grade levels. Translation strategies did not develop in a strictly linear fashion.
Which translation strategies show developmental change?-In answering this question, keep in mind that translation strategies are not the same as number of sentences written. The next sentence sometimes reflected both a Level I and Level II translation strategy or multiple Level I translation strategies or multiple Level II translation strategies. There were some significant main effects for increased use of a translation strategy across writing development, but specific increases depended on the starting grade level and genre. For Narrative Level I, cohort 1 showed significant developmental increases from grades one to grade five for describe observable behaviors, tell next event, pretend or imagine what could be, state outcome, explain, and qualify prior statement; and cohort 2 showed significant developmental increases from third to seventh grade for describe a state of mind or feelings, describe observable behavior, and tell next event. For expository, cohort 1 showed significant increases from grades two to four on stating a fact, illustrating with examples and/or counter examples, and pretending or imaging what could be; and cohort 2 showed significant developmental change from grades four to six for describing a function.
For Level II Narrative cohort 1 increased from grade one to five for use of pronouns to tie sentences together and use of single connecting words and -characters, -setting, -plot introduction, -plot in progress, -plot outcome, and -ending statement; and for cohort 2, use increased from grade three to seven for setting, introducing the plot, and plot outcomes. For Level II Expository, for cohort 1, use increased from second to fourth grade in repetition of prior comments and comments that interrupt, use of a topic sentence for a paragraph, and genre-specific strategies for compare and contrast and making an ending statement and for cohort 2, use increased from grades four to six in topic sentence and providing information but decreased in ending statements.

Relationship between Transcription and Translation
How are transcription and translation related?-Transcription (spelling ability) is not the same as translation (transforming thoughts into written language). How transcription (spelling-may interfere with or support translation may depend on grade levels, genres, and Levels (I and/or II) of translation strategies. Overall, however, the findings varied greatly for different translation strategies, genres, grade, and cohort and showed that translation is a generative process in which thought can be translated into written language in varied ways (cf., Boscolo et al., 2015). Translation of those thoughts into language occurs in multidimensional time-the very next written sentence being constructed in linear time and the unfolding sequence of planning, translating, reviewing, and revising in nonlinear time that ebbs and flows like ocean waves as writers transform their thoughts into written language.
What is the nature of translation?-The identified translation strategies in the Appendix are a practical resource for summarizing observed translation strategies typically developing writers used in this process. As such the Appendix provides a reference point from which other researchers can compare translation strategies at Levels I and II observed in the writing samples of the children they study. Much remains to learn about how translation strategies may vary across age levels, genres, and languages. However, the current study shows that while translation may be constrained to some degree by transcription, it is also the case that translation is a separable process from transcription (cf. Alves & Limpo, 2015).

Applications to Educational Practice
What are the implications of the current findings for assessment?-The Appendix also provides a useful resource for identifying specific translation strategies at specific grade levels for specific genres during clinical or classroom assessment. Educational and psychological professionals should avoid drawing conclusions about a child's writing ability based on a single measure of writing-either for composition or for a transcription skill-or writing assessment at a given grade level. On the one hand, written composing is a generative process that draws on multiple levels of language, may change dynamically across grade levels in terms of translation strategies used, and may be constrained due to intra-and inter-individual differences in transcription processes resulting from normal variation. On the other hand, the descriptive analyses provide clues as to what strategies may be most typical at a given grade level for a specific genre for a specific level of spelling (transcription) ability; analyses of such strategies in the writing samples children produce during clinical assessment may help identify those aspects of translation and/or transcription with which particular students struggle and that should be targeted for instruction.
What are the implications of the current findings for linking assessment to instruction?-Translation strategies can be taught at Level I-for the very next sentence as well as for both Level I (the very next sentence) and Level II (contribution of the very next sentence to the evolving text of varying genres). Both the descriptive analyses in the text and examples in the Appendix provide insight into which translation strategies might be taught at specific grade levels for specific genre for specific spelling (transcription) abilities.

Limitations and Future Research Direction
Only English speaking children from homes in which the parents were for the most part well educated and highly involved in and supportive of their children's literacy development participated in this longitudinal study. Future research might extend the research on translation strategies for writing the next sentence to other written languages, cultures, and socioeconomic groups.
Future research should also explore the most effective ways to teach developing writers how to apply both Level I and Level II strategies to their own writing for different genres. Much research on strategy instruction in writing has focused on text-level strategies for specific genres. However in the authors' teaching, clinical, and research experience, a recurring observation is that many developing writers struggle with what to write next-the very next sentence-and not just genre-specific, text-level composing. Indeed in a related study in which children's meta-cognitions about writing were studied by interviewing them about how they would explain what writing is to a same-age or younger child, awareness of constructing the very next sentence often emerged prior to or in conjunction with text level composing (Berninger, Gesolowitz, & Wallis, In Press).
In addition, writing instruction would benefit from more research on the complexities of writing deriving from the generativity of written as well as of oral language, the multiple levels of language, normal variation in transcription (spelling) abilities, and variety of cognitive-linguistic translation strategies. Such translation science of research into practice is needed to inform valid assessment-instruction links in an era of both 1) standards-based education and high expectations for all students to meet common standards; and a 2) a knowledge explosion about the biological and environmental diversity of individual written language learners. IA20. Make a statement about time (when) and space (where).
Narrative: I woke up on a Sunday morning in December and my roof was white!

B. Provide an explanation
Narrative: The surprising thing that happened was Spenser was not in my class.
Essay: The best part in using a computer is the internet because you can find games and information.

C. Modify Text
IC1. Qualify a prior statement-place limit(s) on it. Essay: This is part of reasons.
IC3. Repeat part of prior text with substitution.
Narrative: It was surprising because I thought I wouldn't get that in the third grade.
That's the surprising thing.
Essay: Robots are a type of machine. They are used for many different things. Some robots are toys and are used to entertain while others are used for different things like helping do everyday things. Robots are made of metal, and can often have similar features to humans. They are useful and can be made and work in many different ways.

D. Create Dialogue
ID1a. Direct dialogue among characters.
Narrative: My brother's pet guinea pig started yelling, "I'm a rebel guinea squirrel!" over and over again.
Narrative: She said that my teacher was gone and she was the substitute.
ID2a. Pose question for reader audience.
Narrative: The end or is it?
Essay: Then, robots would be a species. Wouldn't that be cool?
ID2b. Make editorial comment for the reader audience.
Narrative: I bet that was a good way for him to start off his day.    Open J Mod Linguist. Author manuscript; available in PMC 2017 January 24.