6/29/00
Last Updated 2/9/12
 
Dr. Ed Vavra's KISS Approach to Sentence Structure
Return to Essay # 9 Definitions of Grammatical Terms
Definitions of the "T-unit"
Detail from
Madonna with
saints and 
members of the
Pesaro family
1519-26 
TITIAN
(approx. 1495-1576)

     In 1965, Kellogg Hunt's Grammatical Structures Written at Three Grade Levels established the "T-unit" as the fundamental yardstick for measuring natural syntactic development. Researchers had long been seeking some way of accurately measuring the obvious increase in length and complexity of children's sentences as they grow older, but all attempts had failed. The apparently obvious method -- counting the average number of words per sentence -- fails because very young children write very long sentences composed of several compounded main clauses. In hindsight, Hunt's solution appears simple -- he counted the average number of words per main clause. Because of differences in the definition of "main clause," he called his basic unit of measurement a "T-Unit," which stood for "minimally terminable unit." 
     A "T-Unit," for Hunt,  is essentially a main clause defined as including all subordinate clauses and other constructions that go with it. Hunt's work was so convincing that almost every research project that followed it has used his concept of  the "T-unit." Far too often, however, researchers have whimsically redefined the T-unit such that the yardstick is not a good yardstick. As a result, not only is comparison between studies almost impossible, but the studies themselves are often incomprehensible. Starting from Hunt's original study, this essay will examine the redefinitions in some of the major studies. The differences in definitions should explain why any serious study should be accompanied by complete transcripts of the students' writing.

Studies in Natural Syntactic Development

1965
Kellogg Hunt,  Grammatical Structures Written at Three Grade Levels

     Hunt's explanation of the "T-unit" is so crucial that it needs to be given in full:

3-8. A More Promising Index 
      The increase in mean clause length and the increase in the number of subordinate clauses add up to something worth noting. If it is true that (1) the average main clause written by successively older students has more subordinate clauses attached to it, and if, in addition, (2) those clauses are longer, then the total length of such a unit would of course increase as a result of the two subsidiary kinds of lengthening. This unit whose total length is being discussed contains, to repeat, one main clause with all the subordinate clauses attached to it. The number of subordinate clauses can, of course, be none.
      The length of such a unit might turn out to be a good index of maturity. It might turn out to be an even better index than the two subsidiary factors because of the fact that an individual who was high in subordination index but low in clause length (or the reverse) would have those opposite tendencies moderated by this combining index. 
      A whole piece of writing could be sliced up into units of this sort, just as a rib pork roast is sliced off into chops. The person slicing need only be careful to cut where the joints come instead of cutting into a chunk of solid bone. There should be no trouble deciding whether an expression, if it is intelligible at all, goes with the preceding main clause or the following. An and between two main clauses would always go with the second clause, beginning it just as coordinating conjunctions so often begin the sentences of mature writers. A student's failure to put in periods where he should would not interfere with the slicing process unless the passage already was an unintelligible garble. 
      Here is a sample to be sliced up. It is printed just as the fourth grader wrote it, except that the spelling is corrected. It is a whole theme, punctuated as one sentence, 68 words long. This one fourth grade sentence is four times as long as the average twelfth grade sentence.
 I like the movie we saw about Moby Dick the white whale the captain said if you can kill the white whale Moby Dick I will give this gold to the one that can do it and it is worth sixteen dollars they tried and tried but while they were trying they killed a whale and used the oil for the lamps they almost caught the white whale
      That same theme sliced off into these unnamed units appears below. A capital letter now begins each unit and a period ends each one. A slant line indicates the beginning of each new clause. One unit begins with an and, and another with a but. Each unit is grammatically capable of being considered a sentence. In fact, these units are the shortest grammatically allowable sentences into which the theme could be segmented. If it were segmented into units any shorter, some fragment would be created. 
1. I like the movie / we saw about Moby Dick, the white whale. 
2. The captain said / if you can kill the white whale, Moby Dick, / I will give this gold to the one / that can do it. 
3. And it is worth sixteen dollars. 
4. They tried and tried. 
5.  But / while they were trying / they killed a whale and used the oil for the lamps. 
6. They almost caught the white whale. 
     As segmented above, several units contain only a single clause -- a main clause, of course -- like a simple sentence. There are several multi-clause units like complex sentences. In fact, the second unit is rather intricate, for within the main clause is embedded a noun clause and within it is both an adverbial if clause and an adjective clause. There are no units like compound sentences or compound-complex sentences, for such units must be cut into two or more parts so that each will contain only one main clause. 
      These units need a name. It would be simplest to call them "minimal sentences." However, the word "sentence" already has so many different meanings that misunderstanding would be certain to result. The word "sentence" has troubles enough already. A fresh, neutral sounding name would be better. These units might be christened "minimal terminable units," since they would be minimal as to length, and each would be grammatically capable of being terminated with a capital letter and a period. For short, the "minimal terminable unit" might be nicknamed a "T-unit." One would hesitate to use both initials and let it be nicknamed an "M T unit." So "T-unit" will be the name used for it in this investigation. 
      As a potential index of maturity, the unit has the advantage of preserving all the subordination achieved by a student, and all of his coordination between words and phrases and subordinate clauses. 
      Slicing a theme into these units does destroy the student's coordination between main clauses, or, more accurately, between T-units, as in numbers 3 and 5 in the theme above. But to do so is a gain rather than a loss, if one is searching for an index of maturity. For it has already been shown that certain fourth graders destroy the significance of sentence length by their tendency to string T-units together endlessly with and after and, forgetting to use periods. In fact, coordination with and between T-units is an index of immaturity significant for grade at the .01 level. So such destruction is merciful.
      In anticipation of actually testing out this index, we can see that it holds promise.
      Any two grammarians should be able to agree on an analysis, whether their denominations be transformational or structural or traditional. The investigators used in this study agreed without exception so long as they were confronted by well-formed sentences either declarative or interrogative. (20-22)
     Several things need to be pointed out about this explanation. First, a "T-unit" is a main clause with all its subordinate clauses. Although he does not explain it, Hunt's reference to the T-unit as being more "accurate" than "main clause" probably results from the confusion about the definition of a main clause. A T-unit can be punctuated such as to stand by itself as a sentence. "Compound" and "compound-complex" sentences are to be sliced into "simple" or "complex" sentences. Any other slicing would cut "into a chunk of solid bone." Hunt is, moreover, probably correct when he states that, given this definition, "Any two grammarians should be able to agree on an analysis, whether their denominations be transformational or structural or traditional."
     Hunt did not even attempt to provide any linguistic or psychological explanation for why the T-unit is the base of natural syntactic growth. Given the preceding decades of failure to  find a method of measurement, one can't blame him for that. He was content to demonstrate that if we count certain very specific constructions in specific ways, we will get a graph showing increasing complexity with age and grade level. But his failure to provide a theoretical explanation resulted in a minor failure in his procedures and also left his well-defined "T-unit" open to revision and abuse.
     The minor failure in his procedures involves how he counted words -- he excluded "garbles." According to Hunt:
     "Before the writings could be analyzed, a small amount of extraneous matter had to be excluded. A piece of this extraneous matter, called a garble, was any group of words that could not be understood by the investigators. Here is a passage from the fourth grader who committed the most garbles. The garble is italicized. Where the investigators felt sure that a word was merely a wrong inflection, the correctly inflected form appears in parentheses.
     The man (men) burned the whales to make oil for the lamps in the town. And the man in the little boats and the white whale eat (ate) the boats up and the white whale went down and came up and eat (ate) The other up too and the rest came back to the ship.
. . .
     Two eighth graders and two twelfth graders each allowed one garble to stand unrevised and undeleted. Fourth graders committed more. One distinguished himself with fourteen garbles totaling 68 words. Another committed three totaling 10 words. Two others committed one each. All these fourth graders were boys. The nine girls committed no garbles.  (6)
Although Hunt provides a table with statistics for the number of, and words in, garbles for each student, there is no indication, for example, of how the 68 garbled words of the fourth grader affect his count of words per T-unit. For Hunt's purposes, this may not have been a flaw, but if we look at words per T-unit as a reflection of what is going on in the writer's brain, such an omission is very important.
     The T-unit (main clause) is the base of the KISS psycholinguistic model of how the brain processes language. That model, in other words, provides both a linguistic and a psychological explanation as to why the T-unit is the base of natural syntactic growth. Readers -- and writers -- use short term (working) memory to chunk words into phrases, subordinate clauses, etc., and, in turn, chunk these to a main subject / verb pattern, i.e., a main clause. At the end of a main clause, readers -- and writers -- dump the contents of STM into long-term memory, clearing STM for the next main clause. Although garbles may be unintelligible to the researchers, they clearly represent words that were in the writer's STM. Simply eliminating them from consideration not only, distorts the statistics, it may also throw the baby out with the bath water. The garbles, perhaps, should be the babies for our attention. In Hunt's single example, the individual words all make sense --"And the man in the little boats ..." This garble, in other words, reflects a problem in syntactic processing. 
     One wonders if Hunt would have discarded the following from the writing of a fourth grader:
The worst day I spent was when my momy told me my aunt died and had to go to school. 
Such sentences pose real problems for the statistical researcher -- does this sentence contain four clauses? Or five? But it was, in all probability, the very complexity of the clause structure that caused the writer's problem. Words in garbles should not only be counted in the statistics, but also be examined very closely. The problem of garbles illustrates why copies of all the students' writing should be made available in future research projects. 

    As noted previously Hunt's lack of a linguistic or psychological foundation for the "T-unit" led to its abuse. Researchers felt free to redefine it, often with little or no justification. But Hunt's work should be first base for anyone interested in natural syntactic development in school children.


1967
Roy O'Donnell, William J. Griffin, and Raymond C. Norris. Syntax of Kindergarten and Elementary School Children: A Transformational Analysis

     This study is more complicated than Hunt's in that it covers oral as well as written language. The researchers, however, adopted Hunt's definition of "T-unit" almost exactly. (33-34) Like Hunt, they also eliminated garbles from the word count,

but two special rules were adopted to make the count more uniform and meaningful. Contractions such as he'd and isn't were regarded as two words, and compound nouns (whether written solid or hyphenated in dictionaries) were given the count indicated by the number of bases involved. Thus "snowball" would be counted as two words. (33)
Why "snowball" is more meaningful when counted as two words rather than one is not explained, and, without access to transcripts of the students' writing (and, in this case, speech), it is impossible to determine the effects of eliminating the garbles or of the "refinement" in word count. Like Hunt, O'Donnell defined (in transformational terms) and counted other constructions, but these go beyond the scope of the T-unit. For our purpose here, the important point is that O'Donnell's T-unit was Hunt's T-unit with only minor revisions.

1976
Walter Loban's Language Development: Kindergarten through Grade Twelve

     One of the purposes of Loban's study was "[t]o examine the language development of the same group of children from age five until age eighteen." Covering both oral and written language, it is the most comprehensive, and probably the most sophisticated study that has ever been done on natural language development. It runs into problems with its "Elaboration Index Weights." Beyond the T-unit, Loban attempts to measure syntactic maturity by assigning point values to "language variables." An adjective is worth 1/2 point; parentheticals,, nominative absolutes,  prepositional phrases, modals, participles, gerunds, and infinitives are worth 2 points each;  objective complements and appositives get 3 each; a "First-order Infinitive Clause" (among others) earns 5 points. (17) These number values raise some very serious questions, but on the fundamental question, Loban's T-unit (which he calls a "communication unit") is Hunt's T-unit. (9)
     Like Hunt and O'Donnell, Loban attempts a statistical description of natural syntactic development, but he does not offer a psycholinguistic explanation of why such development occurs. (See "The Principles of Syntactic Development" in Chapter Four of  TGLA.)  He does, however, offer a very astute set of hypotheses on what else to look for as signs of maturity in syntactic development. (6-7). His study is also extremely valuable in that he divided the students into three groups (high, low, and random) based on a number of factors (21-23), and followed the development of each group.



     The preceding studies all attempted to describe natural syntactic development. Although Hunt was later to call for a complete revision of the way grammar is taught, none of the three studies is primarily concerned with teaching. Their emphasis is on describing natural syntactic development, and all three studies define the "T-unit" almost exactly as Hunt originally proposed it. Other researchers, in their attempts to denigrate the teaching of grammar, revised and/or abused that definition.
 
Comparative Studies on Teaching Grammar

1969
John C. Mellon's Transformational Sentence-Combining: A Method for Enhancing the Development of Syntactic Fluency in English Composition

      Mellon's study was an attempt to show that instruction in sentence-combining, together with instruction in transformational grammar, is more effective than traditional instruction in grammar. It predates publication of Loban's major study, but lists the work of Hunt and O'Donnell in the references. Interestingly, in his section on "Background Research," Mellon focuses on the Bateman-Zidonis study, rather than on the work of Hunt and O'Donnell. The Bateman-Zidonis study, which did not use the T-unit, was an attempt to compare  the teaching of grammar by using the Oregon Curriculum and teaching that used a more traditional approach. In this background section, Mellon's only mention of Hunt is to chide Bateman and Zidonis for not using the T-unit:

Bateman and Zidonis' scheme for computing structural complexity leaves much to be desired. Their use of the orthographic sentence ignores the findings of Hunt (1964), who shows that the independent clause is a more reliable unit. Apparently the experimenters wished to count coordinate conjunctions resulting in compound sentences, although the incidence of this structure is inversely proportionate to maturity (11-12)
One must wonder, however, what it was that Mellon wanted to count. After rebuking Bateman and Zidonis for not using Hunt's T-unit, Mellon significantly revised the definition of it:
His rules for T-unit segmentation are as follows:
 
1. Each independent clause, including all constituent constructions, counts as one T-unit.
2. Clauses of condition, concession, reason, and purpose (although traditionally considered constituents of independent clauses) also count as separate T-units.
3. Independent clauses occurring as directly quoted discourses count as T-units. Speaker tags are discarded.
4. Orthographic sentence fragments count as part of the T-unit to which they belong.
5.  True fragments resulting from the omission of a single word count as T-units with the missing word supplied. Other true fragments are discarded.
6. Unintelligible word strings, vocatives, interjections, and various parenthetical or a-syntactic expressions found in conversational writing are discarded.
7. Independent clauses differing from preceding clauses only in their subject, and thus elliptical beyond their verb auxiliary, are discarded.


He then explains that:

"These procedures are comparable to those employed by Hunt (1964). Item two, however, follows from the experimenter's view that logical conjunctions ('if,' 'although,' 'because,' 'so that,' etc. are T-unit connectors much like the coordinate conjunctions, in that both groups of words join independent clauses. Item seven also differs from Hunt's routine. It represents an attempt to discard clauses with repeating (thus elliptical and vacuous) predicate phrases, such as 'and so did John' or 'but neither could the man,' which otherwise would count as T-units." (42-43)
"Essentially the same as those employed by Hunt?" Counting clauses "of condition, concession, reason, and purpose" as separate T-units is a fundamental difference. It increases the number of T-units in any given passage, and decreases the number of subordinate clauses, thereby affecting both the number of words per T-unit, and the number of subordinate clauses per T-unit. And discarding main clauses such as "so did John" reduces the number of short T-units in a passage, thereby almost automatically increasing the number of words per T-unit. Without copies of the students' work, it is impossible to judge the effect that this change had, but without such copies and with so little explanation, Mellon's adaptations raise major questions of his cooking the books.
     A thoughtful reader might well wonder. Mellon had read Loban's 1963 study, and Hunt and O'Donnell had already noted the blossoming of subordinate clauses, particularly adjectival, in grades seven through nine. #1 Psychologists, moreover, were already aware of the "overproduction" factor in learning -- when we learn something new, we tend to overproduce (as in the child's "I cutted the paper.") Mellon chose to work with seventh graders, i.e., students who are on the beginning edge of this blossoming period, but he gave no reason for doing so. He did however, suggest "a three-year program of more or less similar sentence-combining practice extending through grades seven, eight, and nine," (59) These are precisely the years in which previous research was suggesting that T-units increase most dramatically naturally. To his credit, he also notes that 
In the final analysis, however, it must be left to the reader himself to determine whether the enhanced growth experienced by the experimental subjects was indeed the direct result of transformational sentence-combining, as well as, incidentally, whether this growth represents a phenomenon which might just as accurately be labeled cognitive development. (72)
Was Mellon simply pushing natural development to get short-term gains which would give him the statistics he was looking for, or does sentence-combining have lasting effects? # 2 [For more on this, see also the discussion of O'Hare.]
     Mellon's rules for identifying T-units also suggest that that he is counting to get satisfactory numbers, but that he has no other reasons for counting what he does. [He specifically states that "we have no theory modeling and thereby explaining the process." (80)] Consider, for example, Rule 4: "Orthographic sentence fragments count as part of the T-unit to which they belong." In the absence of examples, I assume this to refer to something such as the following, from a fourth grader's writing:
My family has seven people in it. First my dad then my mom then my oldest sister Cyndee and then me and Crystal,  Tara, and Amanda they're my youngest sisters.
Would Mellon have considered "First my dad then my mom then my oldest sister Cyndee and then me and Crystal,  Tara, and Amanda" as appositives to "people," and thus part of the first T-unit? That would add seven appositives to his statistics, but it would totally ignore the fact that it was written as a fragment because the fourth grader's working memory could not handle long appositives.
     Or consider Rule 5: "True fragments resulting from the omission of a single word count as T-units with the missing word supplied. Other true fragments are discarded." As noted above, words represent content in STM. Discarding any of them therefore distorts the text as a representation of what is going on in the writer's head. And discarding "errors" is even worse since errors often reflect points of growth -- precisely what the researcher should be interested in.
     Mellon, however, was not interested in "improving" students' writing in the usual sense of the word. He uses the word "enhancing" in his title, and his primary concern is to evoke more syntactic complexity. Errors were sanitized or ignored. A subsample (85) of the students' writing was analyzed for overall quality, but "These were typewritten so that spelling and punctuation errors could be corrected . . ." (68) In summarizing this substudy, he states:
The writing of the experimental group was inferior to that of the subjects who had studied conventional grammar, but indistinguishable from that of subjects who had studied no grammar but had received extra instruction in composition -- curious results indeed. (69)
Put more simply and directly, the students who studied conventional grammar wrote better than both those who had studied transformational grammar with sentence-combining and those who had not studied grammar. Mellon's research, in other words, proves just the opposite of what the NCTE hierarchy claims that it proves.
      Mellon's study is worth reading, both for his results and for some of the questions that he raises. To my knowledge, his third rule is the first to raise the question of what to do with quoted discourse: "Independent clauses occurring as directly quoted discourses count as T-units. Speaker tags are discarded." Here again, however, he appears to be more concerned with what he is counting rather than why. Quoted discourse generated by the writer (as in writing a story) should be counted because the writer is generating the sentence structure. Quotations of other people's sentences (as in a research paper) should be discarded because the writer did not generate the sentence.
     Mellon directly addresses the question of whether or not grammar should remain in the curriculum. Although he claims that his study should not be used to address that question, his explanation is either too complex or too confused to be summarized here. The confusion surrounding this study is evident from the fact that it was used as an argument against the teaching of grammar, whereas Mellon thought that it might be viewed as an argument for teaching grammar. (75) It was also used as an argument for sentence-combining as a method of teaching writing, whereas Mellon claimed that "sentence-combining practice had nothing to do with the teaching of writing." (79)
     Perhaps the greatest contribution of Mellon's study are his observations on traditional grammar textbooks:
the experimenter finds without exception that all widely used seventh grade texts are limited to these puerile sentence types. Apparently they are employed on the assumption that students of this age cannot learn to speak about the grammatical structure of more complex language. Not only is this untrue, it causes these students to experience and perhaps emulate sentences far below their attained level of syntactic fluency. Despite their manifest undesirability, these activities were chosen as the control treatment simply because they are conventional, established, and well-nigh universal in current seventh grade grammar programs." (38-39)

it may very well  be the case that conventional grammar study fails to promote growth of syntactic fluency not because of the usage practice which it features, but rather because of the hundreds of simply structured and altogether childish sentences which it employs for parsing exercises. As noted, this is the case with conventional textbooks generally, not merely with the one used in this experiment. Nor are the sentences featured in grades eight or nine very much more mature, although they include a wider range of construction types. If true, this observation raises serious doubts as to the manner in which linguistic study is currently being introduced to junior high school students in the vast majority of American schools, since it serves to impoverish rather than enrich the language environments of these students. (63)

For my friends in linguistics, I'd like to suggest that they have the same problem. Students are subjected to countless phrase structure and transformational  rules for deriving the passive "The door was closed by John" from the active  "John closed the door." I have yet to see a presentation of a linguistic pedagogical grammar that deals with the analysis of passages from real texts, sentence after sentence, as does the KISS approach. This may explain students' dislike of linguistic grammars. (See the Elley study, discussed below, p. 67-70.)
     Although Mellon's objective had been to test the effectiveness of sentence-combining exercises and the study of transformational grammar,  he concluded that "Clearly, it was the sentence-combining practice associated with the grammar study, not the grammar study itself, that influenced the syntactic fluency growth rate." (74) O'Hare was to pick up on this and run wild with it.


1971 (published by NCTE in 1973)
Frank O'Hare's Sentence Combining: Improving Student Writing without Formal Grammar Instruction

     Having read Mellon's study, O'Hare decided to test pure sentence-combining against "traditional" instruction in grammar. Interestingly, he rejected most of Mellon's revisions of the definition:

     "Mellon counted clauses of condition, concession, reason, and purpose as separate T-units because he believed that logical conjunctions behave much like coordinate conjunctions. In addition, he discarded clauses with repeating predicate phrases because he claimed they were elliptical and therefore vacuous. This experimenter remained unconvinced by Mellon's reasoning in either case and, therefore, retained Hunt and O'Donnell's simpler and more convincing methodology." (48)
What O'Hare failed to mention is that by returning to Hunt and O'Donnell's definition he was able to get more favorable statistical results. His sentence-combining exercises could be geared to push the natural blossoming of these clauses. Whereas Mellon would have counted "The man ran away because he was afraid of being caught" as two T-units averaging 5.5 words -- with no subordinate clauses, O'Hare gets to count it as one eleven-word T-unit with an adverbial clause!
     Although he treats fragments and garbles as his predecessors had (by discarding them or supplying a missing word), O'Hare does make some improvement in dealing with the problem of quoted discourse:
     "A very real difficulty arose when directly quoted discourse introduced by such an expression as 'He said . . .' was encountered. Mellon discarded the 'speaker tag.' This experimenter was unhappy with such a procedure because it soon became difficult to define the expression 'speaker tag.' There is no great loss when an expression like 'He said' is discarded. But what about the following example encountered in the analysis: 'Clutching the knife tightly in his bleeding hand, Joe painfully crawled towards the opening and said, 'I surrender'.' Exactly what is the speaker tag here? Technically speaking, it would include every word from 'Clutching' to 'said.' Surely this is not a two word T-unit with a sixteen-word speaker tag discarded!
     Hunt stated that 'there is some reason, then, to tabulate direct discourse along with noun clauses' (1965, p. 75). It would be easy to imagine directly quoted discourse consisting of a dozen sentences. Counting all of them as noun clauses would also be unsatisfactory. A compromise was reached by counting the first expression after 'He said' as a direct object, because it seemed to satisfy the minimally terminable requirement for a T-unit. For example, the following discourse --
Marsha said, 'I really like you, John. However, Clarence's father is a millionaire and I like the idea of Palm Beach.'
-- would have been segmented into three T-units -- between 'John' and 'However,' and between 'millionaire' and 'and.' The advantage of such a procedure was that it retained as much of the student's original writing as possible." (48)
This is probably the best way to handle quoted discourse generated by the student. It still does not address the problem of quoted material in research papers, but O'Hare was not facing that problem because of the nature of the writing samples he collected.
     Not surprisingly, O'Hare focussed on clauses and ignored other transformations. He counted the six measures of syntactic maturity which had already been shown to blossom starting in seventh grade: Words/T-unit, Clauses/T-unit, Words/Clause, Noun Clauses/100 T-units, Adverb Clauses/100 T-units and Adjective Clauses/100 T-units. As a result, he got impressive statistics, but they are essentially meaningless. 
     O'Hare ignored both Hunt's and O'Donnell's suggestions that a great deal of "maturity" takes place in the reduction of clauses. (See O'Donnell, p. 24.) Suppose, for example, that a student in the grammar group would have written "He hurt his foot while he was playing baseball." in the pre-tests, and in the post tests would have written "He hurt his foot playing baseball." Even though both Hunt and O'Donnell would have considered this as an increase in syntactic maturity, it would have shown up as a negative in O'Hare's statistics, both in words per clause and in clauses per T-unit. Without copies of what the students wrote, we have no way of knowing how often this happened. 

     Without copies of what the students wrote, O'Hare's study is useless. His results were predictable without his study. He already knew that subordinate clauses begin to blossom in seventh grade, and that the blossoming of subordinate clauses produces a jump in the number of words per T-unit. He already know Mellon's conclusions. One must, therefore, question his selection of an experimental population: "The seventh grade was selected as the level on which to conduct this experiment simply because Mellon chose seventh graders." (37) O'Hare knew that he was going to get positive results, and he was almost certainly aware of the basic behaviorist theory of conditioning -- if one has students spend hours combining sentences into longer ones, longer ones will transfer to their writing -- until the conditioning wears off. One could, therefore, say that O'Hare used -- and abused -- Hunt's T-unit. The abuse is obvious in the title of the study. Whereas Mellon had used the word "enhancing," O'Hare used "improving." 
     Mellon, in his substudy, had found that the traditional grammar group did better in "overall quality." O'Hare conducted a much more limited study of overall quality, a study that is highly suspect. He concluded that "The experimental group wrote compositions that were judged to be significantly better, at the .001 level, in overall quality than those written by the control group." Because of this conclusion, we need to examine what he did in some detail.
     As in Mellon's study, errors in spelling and punctuation were corrected before the samples were evaluated for "quality." Then, thirty matched pairs of writing samples were assembled:

"it was decided that the system of forced choices between matched pairs of compositions would be utilized. Members of the control group were listed and numbered in ascending order of IQ for both boys and girls. A similar list was compiled for the experimental group. A subject was randomly chosen from the control group and a subject of the same sex and approximately equal IQ (within three of four IQ points) was chosen from the experimental group to make up a matched pair." (49)
Although this sounds, at first, very objective and "scientific," note that it does not state that the members of the experimental group were randomly chosen. There were only 83 students in the experiment, 41 in the experimental group, and 42 in the control group. (36) This raises serious questions about the decision to use thirty matched pairs -- why not 20? Why not 40? Did the decision to use 30 enable the researcher to randomly select a subject from the control group and then choose the best of two or three essays from the experimental group to create a pair?
     The evaluation of the (already corrected) matched pairs raises further questions. The evaluation was done by eight experienced English teachers. O'Hare notes that:
"Of course, these evaluators had no knowledge of the nature of the present experiment. They were simply told to make a single judgment on the overall quality of the compositions in each pair, basing their decision on ideas, organization, style, vocabulary, and sentence structure." (50)
Noting that "rater training helps rater reliability," O'Hare explains that 
"during an initial practice period the evaluators were given two matched pairs of compositions, one pair exemplifying very good seventh grade writing and the other, the contrary. On the blackboard, from left to right, were written ideas, organization, style, vocabulary, and sentence structure. After each item was discussed in turn, the evaluators were asked to choose the composition they preferred, basing their judgments equally on all five factors. They were to indicate their preference by making a large check at the top of the preferred composition. Then there ensued a discussion of the relative merits of each of the paired compositions to establish some sort of general agreement concerning the five criterion factors. These five factors were left on the blackboard, and the evaluators were encouraged to glance there occasionally to ensure that they were taking all five into consideration in their judgments." (51)
It would be very interesting to know if O'Hare was involved in the discussions of the group, particularly those about "style" and "sentence structure." Was O'Hare aware of the fairly well established psychological theory that the last item in a list (or in a discussion) is the best remembered? (How was style distinguished from sentence structure? Why was sentence structure last on the list, and last to be discussed?) Finally, and most importantly, why weren't the two essays in each pair rated separately for each of the five items? Was O'Hare trying to suggest that exercises in pure sentence-combining improve students' ability to generate and organize ideas using a better vocabulary? Or was he using these categories to hide something? As it stands, this part of O'Hare's study is meaningless.
     The study, however, was not ineffective. Many in the  English profession are notoriously bad at math. Nor are they particularly adept at scientific reading. The study, its title and conclusions, were thus more talked about than read. And they were hailed as "proving" that instruction in grammar is useless. O'Hare abused Hunt's concept of the T-unit.
     O'Hare himself did not believe the conclusions drawn from this study that supposedly concludes that students' writing can be improved "without formal grammar instruction." In 1986 he published the 454-page The Modern Writer's Handbook (Macmillan), the first half of which is entirely devoted to very traditional slice-and-dice grammar instruction.


1979
Elley, Barham, Lamb, and Wyllie: The Role of Grammar in a Secondary School Curriculum

[For a little background on this study, see the entry in the bibliography.]

     Because this study is sometimes cited to show that instruction in grammar is useless, the first thing that needs to be pointed out is that none of the researchers had a background in grammar or linguistics. They were administrators and teachers who were interested in the question of the effectiveness of formal instruction in grammar, and, because of chance and connections, they received a grant to do an extensive research project. Even a brief reading of their book, however, shows that their study went far beyond the question of grammar into questions of the teaching of literature, the effects of individual teachers on instruction, etc.  They deal with T-units in Chapter Eight, and their presentation is somewhat muddled:

"The segmentation of sentences into T-units followed the principles outlined by Hunt, with minor modifications based on Mellon. Thus:
(1) Each independent clause, plus its subordinate clauses and embeddings counted as one T-unit.

(2) The coordinating conjunctions 'and', 'but', 'yet', and 'so' (when it meant 'and so') were regarded as markers which separate adjacent T-units (except in cases where they separated two subordinate clauses).

(3) A clause was defined as an expression which contained a subject (or coordinated subjects) and a finite verb (or coordinated finite verbs).

(4) Sentence fragments which resulted from the omission of a single word were counted as T-units, with the missing word supplied.

(5) Interjections, parentheses, unintelligible words, vocatives and speaker tags in conversation were discarded.

Examples
(1) The policeman tracked down the thief. (1 T-unit)

(2) The policeman and the dog hunted and tracked down the thief. (1 T-unit)

(3) The policeman hunted through the thick bushes / and tracked down the thief. (2 T-units)

(4) The policeman, with his dog, hunted all night for the thief who had stolen and abandoned the new car / but did not catch him. (2 T-units) (74-75)

No further examples are given, but examples three and four are sufficient to raise enormous questions about the results of the study. They violate the researchers own rule three! Not only that, they also eliminated from consideration what Hunt found to be a major aspect of early syntactic development -- the compounding of finite verbs. Did these researchers apply their rule two such that whenever they encountered an "and" (except between subordinate clauses), they counted a new T-unit? Would the phrase "on land and sea" have resulted in a count of two T-units? (See Rule 2.) Clearly there is conflict between their second and third rules. Without access to copies of the students' writing, the results of this study are totally meaningless.
       Although it goes beyond the question of the T-unit, these researchers also produced a basically meaningless count of the "number of transformations" in ten "T-units" (already meaningless, as noted above) in selected students' writing. They counted the following twelve "transformations" -- 1. Prepositional phrases, 2. Subordinate clauses, 3. Adjectives preceding their nouns, 4. Participles, 5. Gerunds, 6. Possessives, 7. Absolutes, 8. Passives, 9. Appositives, 10. Comparatives, 11. Adverbs, 12. Deletions. (76) As is typical of this project, the researchers were assiduous in their methodology -- "As in the previous analysis, the anonymity of the pupils was retained, and essays from each group mixed together, to safeguard against possible bias." (76) 
     But these researchers appear to be extremely vacuous in substance. They appear to have no awareness of relevant theory, including, for example, O'Donnell's concept of "formulas." According to O'Donnell, a phrase such as "cat in a hat" is probably learned as a vocabulary string -- a formula. No transformations would be involved in producing it. In addition, the twelve categories overlap. Among others, participles, absolutes, and appositives involve deletions. What, then, were these researchers counting when they counted "Deletions" as a separate category? Beyond that, they present the twelve categories as apparently equal, apparently unaware of Loban and Hunt's argument that subordinate clauses blossom in grades seven to nine, or Hunt's argument that participles and appositives are "Late Blooming."
     They do present an interesting little table of "Mean Number of Specific Transformations per 10 T-Units" and comment that "The only difference of any magnitude in the 36 possible contrasts occurred in the case of participles, where t tests showed that the TG group used significantly fewer than the RW and LLE groups (t-2.30, p?0.05 and t-2.62, p?0.01). There is no support in these figures for the hypothesis that a special study of any kind of transformation increases the propensity to use them." (77-78)  The mean number of participles per ten T-units for the TG group as 1.33; for the RW group, 2.10; for the LLE group 2.40. Interestingly, the mean number of appositives per ten T-units for the TG group was 0.91; for the RW group, 0.62; and for the LLE group 0.57. But apparently this difference was not statistically significant."

     Overall, this study addresses some interesting and important questions about the methods and procedures of conducting a research project designed to compare different methods of instruction. But the contradictions within their rules for defining T-units, and between the rules and the examples, make that aspect of this study meaningless.
 

The KISS Definition of a "T-unit"

     If it accomplished nothing else, the preceding discussion should have demonstrated the problems in defining the "T-unit" and the necessity of providing copies of the students' writing.  The research that appears on this site (which does include transcripts) is based on Hunt and O'Donnell's original definition of the "T-unit." That definition, however, is also the best definition of a "main clause," and thus I have used "main clause" in the research to "Keep It Simple." {Note that the final "S" in "KISS" does not imply that students are stupid; rather, it suggests that educators are because we give students numerous, often contradictory and complex definitions of terms and then wonder why the students can't understand and/or apply them.)
     Some adjustments have to be made, however, either because Hunt and O'Donnell did not address the questions, or because they lacked a psycholinguistic theory. These adjustments are:

Word Count: Words are counted as they are by a spelling checker. This means that hyphenated words and contractions count as one word. Word count raises some interesting questions, but O'Donnell's idea of counting "snowball" as two words seems unfounded. In KISS research, I am trying to explore how the brain chunks words (which represent concepts) in working memory. I doubt that anyone who mentions a snowball thinks of (and then chunks) the two separate concepts, "snow" and "ball." Hyphenated words raise an interesting question, but they are relatively rare, and, because transcripts are included, anyone who wants to can study the effect of counting them as two words instead of one. Contractions are counted as one word because, I would suggest, not all  transformations are equally important at all stages of language development.  In essence, by the time a person uses a contraction, I am suggesting that the underlying transformations have become almost automatic. (For more on this, see below.) Again, contractions are relatively rare, and anyone who wants to is welcome to explore the effect that counting them as two words would have. Transcripts are provided.

Garbles: As noted in the discussion of Hunt's work, garbles represent concepts in the writer's STM. They are therefore counted as words.

Fragments, Comma-splices, and Run-ons: Most of the researchers discarded fragments, but as with garbles, fragments represent concepts in STM. An orthographic fragment counts as a "T-unit." The assumption here is that an orthographic fragment is the result of an overloading of the writer's STM. The overloading causes the writer to put down an ending punctuation mark and begin a new "sentence." Counting fragments as "T-units is therefore consistent with the KISS psycholinguistic model. A student's average words per main clause (T-unit) is thus a reflection of the average number of words that the writer can process in STM before dumping to LTM. Discarding fragments would distort this number, probably upward. Attaching fragments to the appropriate main clause (as Mellon did) would also inflate this number.
     None of the researchers discussed above specifically looked at comma-splices and run ons, but the KISS research does. A problem arises here in that most main clause breaks have a double sign -- a closing punctuation mark followed by a capital latter. If either of these is present, and if the first word in the next main clause does not always begin with a capital (such as "I"), no error was counted. (The problem of interpreting and transcribing handwritten texts remains, but that is a different issue. Ideally students' handwritten texts should also be transcribed by someone who is competent but unfamiliar with the nature of the research project.)

Direct and Indirect Discourse: As noted above, O'Hare's decision about quoted discourse is probably the best. The first clause after something such as "He said ...." is counted as a noun clause, direct object. Subsequent direct object clauses (as in "He said he would go, but he would be back.") are counted as main clauses unless they are preceded by "that" ("He said he would go, but that he would be back."). The same rule applies to things such as "He thought ...."
     A problem that none of the researchers have addressed (because of the age groups that they are primarily working with?) is that of quotations and paraphrases in "research" papers. At this point in time, the KISS approach to this problem is to skip any main clause that includes such material.
 
 

Other Measures of Syntactic Maturity 
and the Need for Copies of the Students' Writing

     As Hunt noted, using his definition of the T-unit, "Any two grammarians should be able to agree on an analysis, whether their denominations be transformational or structural or traditional." Although he is right, and although the T-unit is the best basic yardstick of syntactic maturity, Hunt himself, plus all the other researchers looked for, and thought they found, additional measures (subordinate clauses, number of transformations, etc.). As noted above, however, there are substantial differences in the way that researchers have defined these measures. In addition, none of the research projects discussed above were based on a theory that could explain the nature of natural syntactic development. In effect, they were simply counting either constructions or transformations, as if all constructions or transformations of a given type are equal.
     Loban attempted to use a weighted scale, but even his scale overlooks the fact that some constructions (or transformations) may be more or less difficult in different stages of development. In some often cited research, for example, Carol Chomsky has pointed out that young children cannot distinguish between "eager to see" and "easy to see," but a year later -- without any specific instruction -- they have all mastered this difference. In terms of the KISS psycholinguistic model, this means that, confronted by one of these phrases, or in an attempt to generate one of them, the child has to devote a lot of the limited space in STM to it. As a result, there is less space left free in STM to handle the rest of a sentence. Chomsky's example also raises the question of the effect of vocabulary on T-unit length. Then too, numerous researchers have pointed out the effect of mode (narrative, expository, etc.) on T-unit length.
     Then there is the question of "formulas," first pointed out, to my knowledge, by O'Donnell (p.92), although neither he nor any of the other researchers deals with the question adequately. A "formula" is simply a grammatical string such as in "We have a cat named Sam." When a fourth grader writes this, do we say that she is using a participle (which Hunt considers to be a very advanced construction), or do we say that she has internalized the formula "_____ named _____." Clearly young children hear "_____ named _____" relatively often. Thus, if I understand him correctly, O'Donnell was suggesting that children learn formulas as vocabulary strings with slots for substitutions, as in "When daddy get(s) home." These are not true participles, subordinate clauses, etc., because the children have not mastered the underlying transformations which produce true participles, etc. Not all participles, in other words, are of equal value. And, in order to explore which are, and which are not, formulas, we need to see what the students' wrote (or said).

     Once again, Hunt's original definition of the "T-unit"  was such that  "Any two grammarians should be able to agree on an analysis, whether their denominations be transformational or structural or traditional," but we have seen how every researcher discussed above redefined it in some way that would affect the statistics. And once they go beyond the T-unit, things become a hundred times more complicated. The web was not available, of course, at the time that the studies discussed here were done. And reproducing transcripts of all the students' writing in the research reports themselves would have been economically impossible. But now we have the web. And any research project that wants to be considered seriously should provide (available on a web site) both copies and transcripts of all the students' work. Any project that doesn't will look as if it has something to hide.

***

See also the discussion of fragments in the Essay on Fourth Graders' Writing.


Notes

1. Mellon's study contains a number of unexplained redefinitions. Among other things, he counted the "Number of Relative Clauses: The number of unreduced relative clauses. Although traditionally labeled adverbial, clauses of time, place, and manner were counted as relative clauses whose head nouns had been deleted." No examples are given, and I simply do not understand this. How can "When we went to the store" in "When we went to the store, we bought candy," be considered as a relative clause whose head noun has been deleted?

2. I have been unable to find any research that demonstrates long-term effectiveness of sentence-combining exercises, and there are theoretical reasons for suggesting that some such exercises can be harmful, but that is a subject for a different essay. [Basically, sentence-combining exercises can be harmful when they violate the sequence of natural syntactic development by, for example, attempting to get fourth graders to embed appositives in their sentences.]