A SUMMARY COMPARISON OF JUNCTION GRAMMAR
AND
TRANSFORMATIONAL GRAMMAR

Translation Sciences Institute
Brigham Young University
                   Provo, Utah                   

April, 1977[1]

Introduction

Since I am asked with increasing frequency to comment on the differences between Junction Grammar, as opposed to Transformational Grammar or Generative Semantics, it seems appropriate at this juncture to make a summary statement on this matter for the record. The fundamental differences will, I believe, be quite evident if we first examine the general theoretical considerations which motivate them. More specific points of similarity or difference can then be discussed within this broader context.

The Origins of Junction Grammar

The basic assumptions of junction theory were made during the period between 1965 and 1968. They were motivated by some rather strong intuitive notions of my own about language structure in conjunction with a general sense of dissatisfaction about the way meaning was handled by the theory of grammatical transformations. Permit me to elaborate by reviewing my train of thought during the period in question regarding developments in Transformational Grammar. 

Whereas Chomsky's original exposition of TG had asserted that not even the weakest semantic relation holds in general between kernel and transform (Chomsky, 1957, p. 101), subsequent work by Katz and Postal (Katz and Postal, 1964) led to the proposal that, since deep structure was assumed to support semantic interpretation, and was therefore the basis of meaning for sentences, transformations should be constrained so as not to change the semantic interpretation of sentences (the semantic-invariance constraint). This position came to be rather generally accepted, eventually espoused by Chomsky himself (Chomsky, 1965). It led to many changes in TG, including the abolition of double-base transformations - obviously, the embedding of a sentence by transform to one constituent rather than to another (a restrictive clause, for example), could affect the meaning of the constituent to which it was embedded.

This seemed reasonable enough, but the practice of appending markers in deep structure to trigger what seemed to be a potentially unlimited variety of obligatory structure-changing transformations (the markers in question were to be equated by the semantic component with the structures they yielded) struck me as a bad, if not simply ad hoc, solution. From my point of view, there was no reason at all to suppose that natural language uses P-rules plus T-rules rather than some as yet undiscovered mechanism for generating any and all required configurations in deep structure, plus language-specific lexical rules for realizing these as surface strings. Perhaps every transform-triggering marker was simply an indication that the formulation of the base rules was incorrect or incomplete, and not proof in any sense that humans use or need structural transformations. 

My suspicions about structural transformation were reinforced by specific instances in the evolution of TG where T-rules were eliminated by favor of P-rules. Here is the crucial point in this regard:  

Decisions to eliminate transformations seemed to be motivated by occasional insights into the generalities of constituent structure, rather than by the consistent application of some criteria provided by the theory of transformations for deciding whether a given configuration should be obtained via P-rules instead of T-rules. 

The continuing debate among transformationalists regarding how best to handle derivational configurations is a case in point: Chomsky has argued for more powerful P-rules, whereas others continue to advocate the use of transformations in order to obtain these forms. 

I was also troubled by the manner in which proposed T-rules were defended. The standard argument was structured in the following way: 

(1) Sentences ‘a’ and ‘b’ are divergent in their surface structure, but have identical semantic interpretations; therefore, 

(2) they derive via transformation from the same deep structure. 

This line of reasoning is, of course, consistent with the semantic-invariance constraint, but there was absolutely no formal device for showing that sentences ‘a’ and ‘b’ were, in fact, semantically identical, rather than, say, almost identical. For example, it was proposed that nonrestrictive relative clauses were derived from conjoined sentences, because they could be removed from their head without altering its meaning. This implies first, that structural rank (super-ordinate versus sub-ordinate) is not semantically relevant, and second that other nonrestrictive modifiers would somehow be derived from conjoined clauses also. I simply didn't believe that such notions were plausible. To my way of thinking, it was inconsistent to attribute semantic relevance to one aspect of constituent structure, but not to others. It seemed more likely that structural distinctions in general, i.e. the variety of relationships felt between surface constituents, were meaningful, some more subtly than others perhaps, but meaningful in some sense nevertheless. 

Hence, instead of assuming that sentences whose meanings are similar, but whose surface constituent structures are different, derive by transform from the same deep structure, I assumed that such sentences actually derive from separate and distinct deep structures, which, if compared, would be seen to have a great deal in common, but would not be identical in every respect. This meant that not only similarities in semantic interpretation but also differences, regardless of their subtlety, would be a manifestation of deep structure contrasts, or processes of semantic interpretation associated therewith. The structure of semantic quantification would then become more than a binary decision form (semantically equivalent or not semantically equivalent). Rather, it would entail a study and comparison of both similarities and contrasts over a much more extensive set of relationships than had been utilized heretofore. In other words, it seemed to me that while TG was using only two absolute values of semantic comparison as criteria for discussing T-rules, what was actually required was a scale of values between those points for identifying a rather large set of constituent relations utilized by languages of the world but not as yet discovered or explicated by linguists, and hence not accounted for in TG or other models of language. Frankly, I was convinced that the author of TG was mistaken in setting semantics aside initially (Chomsky's original intent was to divorce the study of syntax from semantic considerations), and that consequently, when confronted with the necessity of explicating semantic phenomena, TG was simply not equipped to do so. 

Permit me to elaborate on this point briefly. You will recall that TG was initially motivated by arguments about the relative adequacy of different types of formal grammars for handling natural languages. It was argued that while a phrase structure grammar alone is inadequate to generate the vast variety of surface structures observed among the languages of the world, a phrase structure grammar plus a component of T-rules is adequate. These conclusions were used to advance a theory of syntax which did not appeal to semantics. The crucial issue raised by that whole argument, however, was, and still is, whether any grammar can be said to handle natural language, be adequate in any linguistically significant way, or even be of any substantive theoretical interest for linguistics, when restricted to syntax. Furthermore, for me, at least, the dichotomy drawn between syntax and semantics was simply counter-intuitive, like divorcing music from the study of harmony. From my point of view, syntax (structural relations) was an integral part of semantics.  

Although Chomsky has convinced many that his approach is viable, I think that the evolution of TG has demonstrated otherwise. A grammar designed to generate sentences without regard for their meaning, which is then adapted to accommodate semantic phenomena as an afterthought, experiences the same range of troubles that the "new wine in old bottles" analogy was intended to exemplify. My own experience with TG is just one example. It was bequeathed to me by my mentors with high recommendations. But once I made the decision to regard all structural relations as semantically relevant, the model was no longer a viable instrument with major modifications. In effect, either an alternative to the transformational generation of sentences was required or the notion that deep representations constitute the basis for semantic interpretation had to be scrapped, for while the TG model might succeed in generating the set of surface structures required, it could not in principle account for the semantics of those sentences. As then structured, it ignored the syntacto-semantics of constituent relations introduced by transformations. Chomsky has since repaired this "leak” by permitting semantic interpretation of surface structures as well as deep structures (Jackendoff, 1972) but even that falls short of the mark - The P-rule plus T-rule configuration in its revised form still fails to capture the range of structural relations required to account for the meaning of sentences. I will illustrate what I mean by this shortly. For the moment, however, let us return to the evolution of JG. 

For the reasons noted, I eventually discarded TG and proceeded to formulate the J-rule plus L-rule configuration described in the literature on JG. Since I considered the inter- and intra- constituent relations exhibited by lexical strings to be reliable reflection of meaning, and since I assumed that such relations were defined at the deep level, the notion that any rule of grammar should alter constituent structure during the derivation of sentences was rejected on theoretical grounds. I therefore imposed the condition on L-rules (these are the rules of JG which interpret deep representations and produce lexical strings corresponding to them for specific languages) that they might interpret and reflect deep constituent relations, yielding lexical strings with adequate information to recover deep representations, but not disturb or in any way alter them (Lytle, 1974). 

I defined the constituents upon which J-rules were to operate to be conceptual in nature, rather than lexical, so that one of the functions performed by L-rules would be to express the symbolic link which humans form between the words of a particular language and their stock of concepts. Actually, I also wished the component of J-rules, together with the stock of conceptual objects upon which they would operate, to form the basis for a conceptual grammar which would exist independently of any particular language (Lytle, 1969). 

This arrangement, I felt, was necessary to account for the intelligent behavior of mentally competent humans who, for one reason or another, speak no language, and animals, some of which could probably teach us a thing or two if they possessed a mechanism for externalizing information. Other language specific phenomena relegated to the L-rules were word order, lexical ellipsis, and affixation. It was also obvious, of course, that a component of phonological rules was necessary to mediate between lexical representations and their acoustic or graphic realization. Both L-rules and phonological rules were presumed to operate in the context of information available for reference at the deep level. An outline of the JG model thus conceived appears in Figure 1. 

It remained, then, for the J-rules to account for the variety of constituent relations not attributable to the operation of L-rules or phonological rules. Thus, while others were still apparently of the opinion expressed by Zellig Harris that: 

Some of the cruces in descriptive linguistics have been due to the search for a constituent analysis in sentence types where this does not exist because the sentences are transformationally derived from each other (Language, Vol, 33, p, 338). 

I had decided to make an attempt to improve the descriptive and explanatory power of the base component to the point where it would be possible to generate constituent structures at the deep level for many sentences which had previously been obtained only by derivation from trees having very different constituent relations. In short, I was investigating the possibility that failure to identify the constituent structure of a sentence does not necessarily mean that the sentence is a transform with a different set of relations than those defined by the base rules, but, rather, that such failure is occasioned by ignorance of the set of constituent structure relations which base rules might assign (Lytle, 1971, p. 86). And that is still the basic, fundamental, motivating assumption of JG. 

Figure 1. JG Model of Language

The Origins of Generative Semantics

During the formative period of JG, the generative semantic solution (GS) was advanced by others who observed these semantic troubles of standard TG. Like JG, GS was based on the assumption that deep representations must be semantic representations. GS is essentially different from JG, however, in claiming that there need not be any direct connection between surface constituent structure and underlying semantic representations. The reasons for the GS position are, if I understand correctly, twofold: First, the prime movers of GS, like those of standard TG, either did not notice, or disregarded as unimportant, certain basic correspondences between surface constituent structure and the semantic interpretation of sentences; and second, the authors of GS elected to borrow or build upon representational systems extant in logic and mathematics rather than construct a uniquely linguistic system of representation in essential harmony with relations reflected in surface strings. 

Consequently, GS representations, like those of symbolic logic, for example, purport to represent the meaning of sentences quite independently of what their surface constituent structure may be. This makes it necessary, of course, for GS to utilize structural transformations as a means for mediating between deep representations and surface strings. In fact, the abstractness and complexity of GS semantic representations requires even more extensive use of structural transformations than do other approaches to TG (Jackendoff, 1972, p, 9). 

 Thus, while both JG and GS were motivated by the desire to deal more effectively with semantic phenomena, each reached very different conclusions regarding the nature of deep representations. JG, unlike either standard TG or GS, makes the premise that if the base rules are correctly formulated, i.e. if they succeed in capturing the generalizations and specializations (the scale of semantic values mentioned above) upon which the similarities and differences of meaning we feel in sentences actually depend (these, I repeat, for the most part having been either overlooked or disregarded by other theories), then these rules, in conjunction with language specific lexical rules, will be adequate to account for both the meaning and surface form of any sentence without recourse to the transformation of structure. 

May I hasten to add, however, that one should not, from the foregoing statement, reach the conclusion that JG deep representations are claimed to express explicitly the entire meaning of sentences. In JG, the meaning of a sentence is defined to be a function of (see Figure 2):

(1) Syntax - junction relations and their associated patterns of categorization; 

(2) Reference - constituent referent attributes and other associations established by the information net; and 

(3) Context, including Logic – inferences computed by auxiliary information processing systems.

 Figure 2. Syntax is considered to be an integral part of semantics in Junction Grammar.

 To make a long story short, the meta-model of human intelligence of which J-rules are but a part is presumed to interact with the deep representations of JG (we refer to them as WFSSes, but you may think of them as junction markers if you like), to generate a great deal of information not explicitly represented by them. Notice that the hypothesis that deep representations stimulate more information than they contain is useful in explaining why the same sentence may "mean" different things to different people. It is interesting also to note that this hypothesis is somewhat akin to the notion of semantic interpretation proposed by certain transformationalists who oppose GS (they call themselves interpretive semanticists), but the formal mechanisms of JG for its implementation are very much different than those proposed by interpretive semantics (IS). (Jackendoff, 1972) Let us now turn to IS briefly. 

The Origins of Interpretive Semantics

The theory referred to as interpretive semantics (IS) is actually a reaction to GS, and entails a number of proposals for preserving the original position of Chomsky regarding syntax and semantics, which is, of course, that syntax can be dealt with independently of semantics. Hence, IS rejects the notion that the base rules of the grammar generate semantic representations directly. Rather, it is argued, the base rules, i.e. the P-rules, generate syntactic representations. Semantic representations are then obtained from these by processes of semantic interpretation. There are some new twists, of course. For example, surface structures as well as deep structures are now subject to semantic interpretation, and other descriptive paraphernalia have been added to accommodate "functional" and "modal" phenomena. But not withstanding these addenda, not to mention an enlarged system of P-rules, structural transformations are still required to obtain surface constituent structures. Hence, from the JG point of view, IS represents another attempt to patch up a model which is fundamentally defective. It is not surprising to me, therefore, that the adherents of GS and IS are currently stalemated amid mountains of “empirical evidence” which is used alternatively to argue both for and against both GS and IS. 

It was alleged earlier that attempts to adapt standard TG to deal with semantic phenomena fall short of the mark because those adaptations "still fail to capture the range of constituent relations required to account for the meaning of sentences." The substance and motivation of this statement are rooted in certain crucial differences between the JG conception of what syntax (constituent relations) consists of and the TG conception of what syntax consists of. In order to clarify this point, it will be necessary to consider the format of J-rules and the implications of that format for semantics. A review of how that format was arrived at will also be instructive.

What Junctions Are

The decision to construct a grammar with no structure changing rules, as previously noted, established the need for a more powerful component of structure-generating base rules. Although my intuition suggested to me that constituent structures were a reflection of linguistic operations which join constituents in ways much more specific than the simple concatenation provided for by rewrite rules, it was not at first obvious what those ways were, or how to make them explicit. The first significant step towards constructing the type of rules required came when I realized that noun-phrase relative clauses, noun complements, and other modifiers were all manifestations of a linguistic operation whose output literally permeates human speech. I refer, of course to what we now term subjunction. The sequence of events immediately preceding and following the birth of subjunction in JG were also important in establishing the methodological orientation of the theory. 

I had been contemplating for some time what struck me as serious theoretical problems associated with the phrase structure rules NP > N S and NP > N S of TG (see Figure 3). 

RELATIVE CLAUSE RULE: NP --> NP S 

COMPLEMENT RULE: NP --> N S 

 (1) The embarrassing fact that John was able to learn frustrated his adversaries.

Figure 3. The TG treatment of noun complements and relative clauses.

 These rules, proposed to account for restrictive relative clauses and noun complements, respectively, both embedded an entire sentence to a nominal antecedent, encompassing both constituents with brackets labeled NP. It seemed clear that the two structures in question were in fact related, i.e. in some sense similar, but not in the way suggested by the P-rules proposed to generate them. Specifically, in each case there seemed to be an overlapping of constituents in the main clause with those in the subordinate clause: In the relative clause, an NP of the main clause coincided referentially with an NP of the dependent clause; in the complement, a noun, or noun phrase potentially, of the main clause was equated referentially with the entire dependent clause, Thus, sentence (1) is ambiguous over the relative clause and the complement readings: 

(1) The embarrassing fact that John was able to learn frustrated his adversaries.

 (Notice that you can replace that with which for the relative clause reading but not for the complement reading.)

There was nothing in the P-rule formulation to make this overlapping of constituents explicit, however. Hence, a mechanism for checking the coreference of NP's was required so that the relative clause transformation could apply to sentences embedded by NP > NP S and produce the appropriate relative pronoun. It was not clear whether this mechanism was supposed to establish a coreference relation between the head N and the complement S in NP > N S, however, since no T-rule depended upon such a check. Moreover, there was no

clear justification for using N rather than NP to the right of the arrow in the complement rule, since the head of a complement can have articles and modifiers too. 

Still more serious, to my way of thinking, was what appeared to be implied by the relative clause rule. Namely, the entire clause was bracketed with a nominal category (NP), suggesting that it, like the complement, was functioning in its entirety as a nominal constituent. I concluded that the structural symmetry of these two rules resulted in a false generalization (the illusion that both clauses were nominalized) while failing to make explicit the generality which actually existed and was semantically crucial (the referential overlap between constituents in the main and dependent clauses). I therefore proceeded to devise structural representations which reflected the overlapping of constituents without violating what I sensed were the correct categorizations. This was first done in the manner shown in Figure 4 and later as shown in Figure 5. 

 Figure 4. Initial JG diagrams for noun phrase relative clauses (A) and noun complements (B).

Figure 5. Later JG diagrams for relatives (A) and complements (B).

 As I further contemplated the relative clause diagram of Figure 4 (A), it occurred to me that the main clause and the dependent clauses could be construed as intersecting structures. In the case of the relative clause, the intersection occurred on categorially homogeneous nodes (on NP's), whereas in the complement structure, the intersection occurred on heterogeneous nodes (NP/S) so that the entire subordinate clause intersected with an NP of the main clause. It then struck me that perhaps any subordinate clause could be analyzed in terms of its intersection with the main clause. Turning to English, and other languages familiar to me, I observed almost immediately that in addition to relative clauses which intersected the main clause on NP's, there were others which intersected on adjectives, and adverbs, as exemplified by sentences (2) and (3):

 (2) Dogs such as Duke is require little food.

 Clause main: Adj dogs require little food.

 Clause dependent: Duke is Adj

 (3) As ye sow, so shall ye reap. (You will reap like you sow.)

 Clause main: Ye shall reap Adv

 Clause dependent: Ye sow Adv

 I likewise noticed adverbial and adjectival complements, in addition to noun complements, as exemplified by sentences (4) and (5): 

(4) He walks so that his toes point outward. (manner)

(5) The set such that it has no elements is empty· (attributive)

 By this time I had reformulated the original rewrite rule to be a junction rule:

 NP / NP S "NP subjoin NP of S"

 In order to account for clauses intersecting on adjectives and adverbs I added other appropriate rules: 

(Relative) AdjP / AdjP S "AdjP subjoin AdjP of S"

(Relative) AdvP / AdvP S "AdvP subjoin AdvP of S"

 

(Complement) AdjP / S "AdjP subjoin S"

(Complement) AdvP / S "AdvP subjoin S"

 At about this point, I made an important change in my approach to the data. Up till then, I had simply added to my list of subjunction rules as I happened to encounter structures which corresponded to them. Now, however, I began to anticipate what could plausibly be expected to happen in the way of intersecting structures. In other words, I generalized the subjunction rule to be a formula (X / Y S), and began to deduce specific possibilities in terms of the constituent categories extant in my system of diagramming at that time. Moreover, rather than rely on available data to motivate or confirm new rules, I began to use rules predicted by the subjunction formula to suggest data not otherwise available. 

Why this change in my approach to data was significant becomes evident when considered in the context of rules which correspond to certain rarely used but perfectly legitimate junctions. Consider, for example, sentences (6) and (7): 

(6) I fristicized my suspenders, like you did your garters.

(7) The book is on the table, which is not the same relation you observe between the picture and the wall.

 Imagine how long one might search a text before finding an instance of either of these relative clause types. Each would require a special context. Yet, any grammar of English would be incomplete without the subjunction rules for (6) and (7). The fact that some rules are used less than others in no way invalidates them. In a number of cases I deduced what at first struck me as the most ridiculous rules for subjunctions, only to find that sentences formed to fit them were perfectly acceptable. It soon became apparent to me that any grammar which is based on rules arrived at inductively, i.e., simply set down in a list as available data suggests them, will be incomplete, at best, and most probably not motivated by any significant generalizations. 

 Not long after, of course, came the realization that it is also possible to overdo a good thing. Some of the ridiculous looking rules I deduced turned out to be as ridiculous as they looked, i.e., no language that I or my colleagues are familiar with confirmed the validity of the predicted junctions. This turned out to be a healthy period of frustration, however, since it forced the formulation of linguistic constraints on both the form and content of junction rules. The identification of these constraints in turn led to new insights into the nature of constituent structure relations and their semantic correlates. 

Our research eventually produced what has turned out to be a very stable notion of what a junction, i.e., a linguistic operation which joins and inter-relates constituents, entails. Every J-rule consists of the following:

 (1) Categorized constituent operands;

 (2) An operation symbol which specifies the relation to be defined between the operands; and

 (3) A designation of the category of the new constituent formed by the junction.

 This definition of junction, plus a system of constraints designed to keep predicted junctions within the range of generalizations (similarities) and specializations (differences) evidenced by natural languages, does in fact provide a scale of semantic contrasts of the type I had originally envisioned. In fact, it has been possible to define increasingly more refined semantic distinctions, corresponding to higher orders of specificity in operand and operation symbols. A host of subtle differences of meaning can now be explicated in terms of junction alone. All of the proposed adaptations of TG still fall far short of the mark because none of them to date shows any signs of capturing even the most obvious constituent structure generalizations: To my knowledge, complements and relatives are still handled with the same old rules (the notion of structural intersection has escaped transformationalists entirely); such fundamental distinctions as subjunction, conjunction, and adjunction, are absent in their representations, not to mention the lack of any device to account for categorization patterns (see literature on JG for a discussion of these topics). 

 One final comment regarding the notion of structural transformation. I think it has never been satisfactorily explained from a theoretical point of view how one justifies the assumption that the transformational manipulation of structure without regard for meaning (either consciously or subconsciously) has any relevance for natural language phenomena other than, perhaps, idioms, whose structure is characteristically at odds with their semantic interpretation. Moreover, unless the realization of deep structures as lexical strings is totally unlike any other natural coding system in my experience, that process is more readily presumed to operate within constraints which make the path between levels of representation direct and straightforward rather than tortuous and obscure. Those who have attempted to devise analysis or synthesis algorithms using TG will appreciate the practical if not the theoretical implications of his point. 

 And that is the final broad distinction I wish to draw between JG and current versions of TG. It is, of course, a corollary of JG's basic premise - that base rules generate all constituent structure directly, while lexical rules interpret and reflect that structure, but do not alter or add to it in any way. Hence, except for idiomatic expressions or instances where morphological devices have become ambiguous or inert, JG makes the tacit assumption that surface strings are a reliable and direct reflection of the deep representations underlying them. 

This concludes my discussion of broad differences between JG and TG. There are, of course broad similarities to be noted also, but a consideration of these is beyond the scope of this paper. For the present, given the general areas of divergence discussed thus far, an enumeration of some of the more salient specifics of JG is offered in Appendix A.

APPENDIX A

Salient Aspects of Junction Grammar:

 1. Meaning is considered to consist of:

 A. Structure (junction relations and categorial contrasts)

 B. Reference (the denotational value of linguistic tokens)

 C. Context (connotation, inference, presupposition, etc.)

 2. Trees are generated either from the bottom up or from the top down.

 3. Relations between nodes are shown as junction operations. Three basic junction types are:

 A. Subjunction - accounts for modifiers of all kinds,

 B. Conjunction - accounts for nodes coordinated by and, or, but.

 C. Adjunction - accounts for relations between subjects and predicates; also between predicators and objects.

4. Patterns of categorial predominance are expressed.

5. Junction rules are categorically schematic as well as referentially schematic, i.e. every J-rule is derived from a junction formula.

6. Lexical rules interpret rather than transform. (There are no rules in JG which bring about a structural change.) 

7. A. Lexical ordering rules specify order of interpretation (trees do not express any explicit linear order for nodes)

 B. Lexical hiatus rules suppress the application of lexical matching rules and never delete nodes from trees. 

C. Lexical matching rules do not insert lexemes into the tree, but into a lexical string which exists independently and apart from the tree. 

D. Lexical agreement rules do not copy feature information from one node to another, but cause affixes to appear in the lexical string if certain feature information appears in the tree.

 8. Features are used in JG to abbreviate tree structures or to specify information furnished by the information net or the logic. 

9. Junction rules can be used to diagram fragments of discourse as well as to diagram sentences. 

10. Each junction relation has a semantic correlate. Even subtle junction contrasts result in semantic contrasts. 

11. Junction relations are directly associated with phonological phenomena.

12. Junction rules are considered to be a universal pool of possibilities from which all languages draw. 

13. Lexical rules are language specific, although certain formal properties of lexical rules are shared by all languages.

WORKS CITED

 Chomsky, Noam. Syntactic Structures.The Hague: Norton & Co., 1957.

 _____________. Aspects of the Theory of Syntax.Cambridge, Mass.: MIT Press, 1965.

 Jakendoff, Ray S. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press, 1972.

 Katz, Jerald J. and Postal, Paul M.In Integrated Theory of Linguistic Descriptions.Cambridge, Mass.: MIT Press, 1967.

 Lytle, Eldon G. A Grammar of Subordinate Structures in English.Hungary: Mouton & Co., 1974.

 ______________. Structural Derivation in Russian.Ph.D. Dissertation, University of Illinois (Champaign-Urbana), 1972.

 ______________. "A New Look at Grammar." Unpublished paper presented at BYU College of Humanities Seminar, 1969.

[1] Paper originally presented at the Languages and Linguistics Symposium, Brigham Young University, 1975

[Home] [Origins] [Article Archive] [Foundations] [Formalizations] [Analyses] [Pedadogy] [Forum] [Guest Book] [BYU]

Copyright© 2004 Linguistic Technologies, Inc.