INTRODUCTION

Style and Processing

GENERATIVE grammar, with its claim to represent the human mind and its activity, seemed a decade ago to promise a genuine insight into the function of language in literature. This hope was eloquently set forth in the concluding sentences of Richard Ohmann’s now famous programmatic essay, “Literature as Sentences”:

Although there has been much progress in generative grammar in the last decade, it has not developed into the handmaiden Ohmann hoped for. Writing six years later, Roger Fowler cited Ohmann’s words, but still found it necessary to ask, “What do transformations do to us?”² Fowler, however, had little to offer in answer to this question, and what he did suggest is vulnerable to the scathing strictures leveled against generative stylistics by Stanley Fish in his English Institute paper of 1973.³ Fish saw that generative stylisticians had never systematically specified how sentences are understood and had indeed made contradictory assumptions about the effect of deletion, for example, on the reader. Fish is correct both in his description of the facts and in his identification of the critical weakness. Generative stylisticians never have sketched a model of how sentences are read, and, lacking that model, have described connections between syntax and cognitive processing which strike others as vague or fanciful, a mixture of inspired hunch, intimation, and bald assertion in no way more explicit or insightful than traditionally based analyses.

The reason that generative stylisticians did not specify a model of reading is to be found, I think, in a naive understanding of generative grammar that was endemic in the 1960s—one eminent grammarian has confessed to it, and Chomsky warned against it in Aspects of the Theory of Syntax.⁴ That misunderstanding was to take the term generative psychologically—to assume that the model of surface structure, transformations, and deep structure was a direct replica of sentence production and perception. In reading a sentence, one parsed into surface structure, 'undid' the transformations to get a deep structure, and applied rules of semantic interpretation to the deep structure to get the semantic representation. Hence, one could assume that the transformations directly described some of the sifting done by the mind, some of the cognitive processes set in motion.

In the last decade, a body of evidence has accumulated which has raised doubts that the transformational model Ohmann assumed, or any of its descendents, is a direct rendering of the way we process sentences. Some sentences prove to be easier than their transformational derivation would predict, others harder. This research is summarized in Fodor, Bever, and Garrett’s The Psychology of Language and in Charles and Helen Cairns’s Psycholinguistics.⁵ Fodor et al. conclude that psycholinguists thereby discovered that they had something to do, which is, among other things, to specify the “computational systems” whereby hearers and readers interpret sentences. Psycholinguists have made some headway in this specification, and attempts to simulate these 'computational systems' with computers have also clarified certain points. It appears that the model that is beginning to emerge is what is needed to realize the objective of generative stylistics. The way a writer chooses to frame sentences and place their elements does affect the reader’s cognitive processes in predictable ways which analysis can explicate, but via the strategies of processing: a particular construction or preference of a writer is important insofar as it affects processing of the text. In this way, stylistics becomes concerned, as Fish argues it must, with the way texts and readers act on each other.

We will begin with the following assumptions and rudiments of a model of reading.

1. Reading has at least three levels, which we will call perception, comprehension, and interpretation. We will say we have perceived a sentence when we have specified its propositional structure. This includes identifying the propositions and the phrases that function in them, matching logical Subjects and Objects with the predicates that relate them, and associating modifiers with the elements they modify. These are relatively simple and uncontroversial notions, and we will not need any sophisticated logical formalisms to represent them. I thus assume no difference between the goal of perceiving sentences of literature and those of other discourses. Comprehension of a sentence involves the integration of its propositional content into one’s running tally of what is being described or argued in the passage. It is on this level that one identifies the individuals referred to by the noun phrases and pronouns; the time and place of the actions represented in the sentence; possible motives, purposes, instruments, and consequences of the actions; and so on.⁶ This integration may involve inferences establishing relations not explicitly stated in the sentences of the passage. Perry Thorndyke gives a simple example in the sequence,

where we infer the proposition 'there was a chandelier in the room' to link the second sentence to the first. Thorndyke states the general point nicely:

The difference between perception and comprehension can be illustrated by a paragraph devised by J. D. Bransford and M. K. Johnson where the propositional structure of each sentence is easy enough to make out, but the sentences do not give us enough information to construct a 'frame' in which they will all cohere:

Given any sort of clue to the frame, however, one finds that the paragraph falls easily into comprehensibility ("Washing Clothes").

Material in the contextual frame can constrain perception of new sentences. For example, the words in the following sentence can be grouped into phrases two different ways,

He put the block in the box on the shelf.

but if we had a context involving a block in a box we would not hesitate to group 'the block in the box' as a phrase in this sentence, and if the context has established a box on a shelf, then we would perceive 'the box on the shelf' as a phrase. This example suggests two major points: first, we may locate noun phrases not by looking for strings of words that could be grouped as noun phrases (with functions and reference to be decided later) but by looking for words that refer to things to which we can expect reference to be made. Second, we may not, therefore, construct all the possible parsings of the words before us if the first one we try is a plausible continuation of the contextual frame. We will examine many instances of 'garden path' phenomena in the following chapters which suggest that at least some of the time we choose one analysis rather than keep track of them all. This is an obscure and controversial point. All agree that many ambiguous strings are not noticed to be ambiguous—the question is whether the alternatives are momentarily constructed at some level beneath awareness. Such evidence as there is is conflicting and inconclusive, and it is also not clear what difference it makes. If these alternatives are never integrated into the networks of meaning one is constructing, they would seem to vanish as quickly as they arose (if they arose) and to be of negligible conceptual import.

Further, in the case of word recognition, we may be able to specify pretty closely the sense of a word before getting to it, and recognizing it then consists of determining that it has the right orthographic shape to bear the sense we have projected for it. In other cases, seeing its written shape will enable us to decide which of the possible concepts is in fact being invoked. This means that the sort of process envisioned by Empson where one retrieves or activates all of the senses of a written form which then reverberate in the subconscious while one sense is preferred by the conscious mind is an unusual or atypical one in general reading rather than the unavoidable basis of even the most practical kinds of reading (as Empson believed it to be). The view that skilled readers read word by word and from phonetic/graphic shape toward possible meaning is now generally doubted.⁹

Comprehension, then, synthesizes material into a world with actors, places, forces, and so on. In doing this it may be guided by conventions special to the type of discourse being processed. For example, there is a convention of narrative that descriptive material following a verb of perception is to be taken as specifying what was perceived. Hence the italicized he in the following sequence is not even momentarily ambiguous:

A more complex example would be the conventions of free indirect style in narrative, where one figure is referred to as he (or she) but must in some other respects be treated as an I if the passage is to cohere into a congruent scene.¹⁰ The question of when one should stop searching for a character’s motives is also settled in part by the type of work one is reading; inferences about physical causality may not be applicable in works of fantasy, and so on.

Interpretation, finally, is the most abstract level where we relate the sense of what is going on to the author’s constructive intention—why he is saying what he says, or what he is getting at in terms of the themes and meaning of the work. In this sense of the term, all discourses require interpretation, and they must be interpreted by means of conventions to various degrees special to the type of discourse. Literary interpretation, then, is not something different from 'usual' interpretation but simply interpretation of literature according to the conventions appropriate to literary discourse.

Interpretation governs comprehension and perception in that we tend to see what we have inferred the writer wants us to see. Samuel Fillenbaum conducted a study in which subjects were asked to paraphrase sentences which included such 'perverse' items as

In their paraphrases, over sixty percent of the subjects changed the sentences into the more likely sentences,

More than half were unaware of having changed the sentence, and many of those who were aware said that they knew what the original sentences were trying to say and so 'corrected' them.¹¹ Our capacity to take certain utterances as ironic and to respond to indirect speech acts, interpreting, for example,

not as a request for information as to ability but as a request for action, depends on this ability to infer the Speaker’s probable intentions.¹² William Empson’s sixth and seventh types of ambiguity furnish many other examples where the most obvious interpretation of lines taken out of context is inconsistent with the interpretation of the poem they occur in and so are 'adjusted' to conform. When we read the last line of Ben Jonson’s “Drink To Me Only With Thine Eyes,”

we do not read “If I could drink Jove’s nectar, I wouldn't give it up for yours,” but rather “I would not exchange yours for it.” Again, this line may not send one scurrying off to the OED to find out whether there is an archaic sense of change that will warrant the reading that has to be right on the grounds of tone and intention of the lyric (there is). One may simply not notice and read “change thine (for it).” Similarly the odd possibilities Empson pointed out in the famous stanza from Lovelace’s “To Althea”

namely, that the pronoun that, being singular, should be taken as referring to prison/cage, and hence we might get “minds innocent and quiet can take prisons and cages for hermitages (i.e., sentimentalize them), but they cannot do the same thing with (real) stone walls and iron bars.” In another context, or century, these lines might well bear this force. It is so wildly inconsistent with the tone of the rest of the poem, however, that we may read that as referring instead to a general “imprisonment.” There is a kind of dismissive aloofness conveyed by that (as opposed to this or a more accurate plural [those, them]) that does harmonize with the tone of the poem. Again, in context we may never notice, or indeed construct, the reading Empson says is there “on the face of it.”¹³

Just as interpretation can shape comprehension and perception, so it may become the guiding consideration when a text is grammatically irregular, obscure, or elliptical: the processes of perception and comprehension do not get enough data, or enough consistent data, to select a reading, so one imposes a propositional structure or contextual frame which supports one’s sense of what the passage should be saying. We will examine passages, especially in Chapters III, IV, and V, where reasoning from probable artistic intention is the most efficient way to set the text in order.

This three-way distinction (perception/comprehension/interpretation) is a useful one which we will employ throughout. Since everyone uses these terms differently, it is perhaps well to repeat the crucial basis of the distinction as we draw it, which is the goal of each level of processing (propositional structure/contextual frame/constructive intent). The three levels are not processes as such; rather, various processes may be carried out to achieve the goals of each, and the three levels may interact as described in achieving them. In what follows, we will be investigating the ways we may try to achieve the goals of propositional structure and contextual frame. How we arrive at constructive intent exceeds the bounds of this study, though E. D. Hirsch, Jr., argues that the process of arriving at the author’s intention is of the same nature and a continuation of these Tower level' processes of forming and testing hypotheses about the meaning of the text. His remarks succinctly characterize this process: “For that which we are understanding is itself an hypothesis constructed by ourselves, a schema, or genre, or type which provokes expectations that are confirmed by our linguistic experience, or when they are not confirmed, cause us to adjust our hypothesis or schema” (Aims, pp. 33-34). We proceed now to specify more narrowly our initial assumptions about perception and comprehension.

2. Readers have learned strategies for constructing propositional structure directly from serial order of words; they may not (reconstruct a surface structure (a labeled bracketing into phrases) in all details (though I assume a grouping into basic phrases), nor pass through a deep structure representation by 'undoing' transformations. For example, if one is to process the following sentence correctly,

a person must be identified as the antecedent of who, but this identification need not pass through an intermediate stage in which the entire relative clause is placed next to a person in some sort of mental representation.

3. These perceptual strategies can be semantically based (e.g., if one hypothesizes that a certain noun phrase is the Subject of a verb because it is the most semantically congruent noun phrase around) or syntactically based (e.g., if one hypothesizes a noun phrase as Subject because it is in a position often occupied by a Subject). One may use one sort of strategy or another, or use one as a check while relying on the other, and shift strategies as they prove to be more or less efficient and reliable. Psycholinguists have found evidence that some passive sentences are perceived somewhat more slowly than their active counterparts, but this is not the case if semantic clues identify the logical Subject and Object. For example,

is perceived as quickly as its active counterpart, though

is not, presumably because the latter order and words would also constitute a possible active sentence (The horse followed the cow.). In other words, when either noun phrase could be the logical Subject, one must attend to whether the markers of passive voice (be, past participle, by) are present. Similarly, the first sentence of the next pair is easier than the second (though still very hard),

presumably because semantic clues help to identify the lion as the Biter, the girl as the Asker, and the question as the thing which is complex. It is still hard to get the structure of the first sentence into focus, but the semantic clues do assist us in getting our bearings. Semantic clues are not always available, but when they are, they facilitate perception.¹⁵ Several different proposals have been made about the exact way semantic clues might function in perception, including John Kimball’s suggestion that semantic compatibility is used as a check in initial parsing to help in choosing between alternative analyses (as it is in Terry Winograd’s parsing program),¹⁶ and the proposals of Langendoen, Kalish-Landon, and Dore and of Lakoff and Thompson that one might use semantic information to make initial hypotheses as to function.¹⁷ Lakoff and Thompson, for example, propose that one might identify the following italicized noun phrases as logical Indirect Objects by a partially semantic strategy:

The strategy could be roughly phrased: “If you have a noun phrase which is not the logical Subject and which is animate, it is likely to be the logical Indirect Object (if the verb takes one).” This strategy will of course give some erroneous analyses, as it would to

'Erroneous results' can be good, however, if they correspond to actual misunderstandings we experience—in that case, they explain how we misunderstood.

The notion of a semantic clue must be taken in the broad sense that includes pragmatic knowledge or assumptions and inferences drawn from the context. We did smuggle a piece of pragmatic information into the example with the lion, the girl, and the question: we assumed that the girl was incompatible as the Subject of bit, at least in the presence of a lion (i.e., given a choice between girl and lion as the Biter, take the lion), but this is not a true incompatibility, just an improbability. Further, in a given context, certain individuals are likelier to do certain things than others: in the case at hand, the probabilities change greatly if the lion has been shot full of tranquilizers. Thomas Bever speculates that, so enriched, semantically based strategies are very powerful:

We will see examples of contextual information clarifying formally obscure lines in the next chapter (especially examples 28, 29). One problem with exclusive reliance on semantic clues, however, is that the information they depend on is not always available, either because candidates are all roughly equal in eligibility (examples 30, 31, 41 of the next chapter) or because one has to look too far ahead or behind to check for compatibility (passim). Notice by the way that with this enriched definition of semantics we set up an interaction between comprehension and perception: entities can be eligible for a certain function in propositional structure because of their place, powers, or other attributes in the 'contextual frame' or 'world' we construct in comprehension. This means that the reference of a noun phrase may be determined before its function is assigned; hence, in some cases at least, perception is not prior to determination of reference.

4. We will examine many passages where clues of parallelism, repetition, alliteration, rhythm, and rhyme assist perception of propositional structure. Conceivably they might do this by being the bases of perceptual strategies, but it is more likely that we construct other sorts of groupings besides syntactic ones, and when these groupings align with syntactic groups they may facilitate perception of propositional structure. Rhetoric books bear witness to the force of such non-grammatical indicators of patterning by warning against using parallelism and repetition unless the sense is truly parallel, the phrases logically alike.

5. (Implicit in 3 and 4): One evaluates and adjusts the strategies one uses so that the particular structures favored by a writer can be most efficiently processed. If, for example, one is reading an author who inverts word order frequently, one might begin to give more weight to semantically based strategies, though if he inverts just in certain ways one might simply make certain adjustments in syntactically based strategies.¹⁸

6. (Follows from foregoing): Processing strategies are probabilistic in nature: they are based on expectations and likelihoods; they are not categorical; they can cope with common but ungrammatical constructions while being unable to cope with grammatical but uncommon ones; in all of these respects they belong to the domain of performance—they are indeed a quintessence or abstract of previous performances—rather than competence. Nils Erik Enkvist has quite properly emphasized that insofar as a text is stylized it departs from a norm, and it presents the reader with special problems insofar as it deviates from a norm of expectation:

As the word generate suggests, however, Enkvist is also somewhat confused about the status of generative grammar as a model of perception and production (see also his p. 43), and we must qualify his argument in one respect: a statistically uncommon construction is not necessarily a source of difficulty—it is so only if it cannot be readily processed by the strategies the reader is using. Further, passages can be difficult without involving any statistically rare features. Of course, the combination of common features that produced the difficulty might prove rare in some cases, but that is tendentious in the extreme. Fish’s conclusion holds here: no profile of an author’s syntactic proclivities, even if quantified along the lines Enkvist suggests, can directly predict how his style will shape our apprehension of his meaning.

7. Perceptual strategies are designed to accommodate limits of 'primary' or so-called Short Term Memory. Sentences are read linearly, with processing done as one reads, whenever possible before one gets to the end of the sentence. When a tentative analysis is assigned to an item, it can be removed from Short Term Memory, so there is a high value on strategies that assign an analysis quickly—with little Took ahead', for example. In general, we experience difficulties in perceiving a sentence when a piece of information needed by the strategy we are using is not in its expected place but is either delayed or in a different place.

8. The fundamental condition that a perceived propositional structure must satisfy is that it be based on a grammatical analysis of the sentence. To illustrate with an absurdly simple example,

with the revolver cannot be perceived as modifying the guard even though it is semantically congruent as a modifier of guard. Similarly, whatever we take the referent of the pronoun in the next example to be, it cannot be the referent of the boys (by grammatical rule):

We must always choose propositional structures and reference assignments that are based on a grammatical parsing of the sentence over ones that are not. I will leave open for the time being the question of how one deals, or should deal, with cases where two propositional structures are grammatically possible (Empson’s “double syntax”) and cases where no grammatical analysis is possible, but we will touch on these matters at the end of Chapter Two and face them squarely in Chapter Five. A final point here is that perception is self-correcting: a hypothesized structure that leaves a piece of the sentence unanalyzed, or requires a piece that is not there, will automatically be questioned or rejected.

9. The notion of choice of an analysis must be treated with some delicacy. There are times when we may avoid making a choice, holding the point open as we read ahead in hope of further information that will aid in making the choice. Some choices are typically on-the-spot decisions (e.g., whether have or be is a main or auxiliary verb), primarily because the information needed to make the decision is usually found in close proximity. These decisions may be harder to delay than others which often require suspensions of decision (e.g., pronominal reference), and hence a writer whose style complicates an on-the-spot decision may give us an amount of difficulty disproportionate to the importance of the choice. Some critics, moreover, have suggested that if a writer makes a choice difficult for us and creates situations where either choice fits into the interpretation of the work, perhaps we should not choose between alternatives but 'take both'. This suggestion involves fairly radical revisions in our model of reading which we will consider in Chapter Five.

Obviously these initial assumptions are very loose, allowing for variation within one’s own practice, among the practices of different readers, and over particular texts being read. How, in practical terms, can readers discover what strategies they are employing? The key here is the experience of difficulty or confusion—in extreme cases, of incomprehension or miscomprehension. At such moments, the strategies jam, and we can become aware of the origin of the difficulty by noting what in the text is unusual. Normally we are not conscious of choosing or using a particular strategy, but when the results of the processes fail to tally as a well-formed structure, we shift to a more consecutively organized problem-solving routine, constructing and weighing alternative readings.

Usually this shift is accompanied by a sense of greater effort or concentration being expended. Paradoxically, we can learn about how we read simple texts by analyzing the sources of our troubles with difficult ones. Most of the sentences to be discussed in these chapters have given me difficulty. In some cases, the difficulty was only momentary; in other cases, I have repeatedly misread the example; in yet others, I am still uncertain what the best reading is. This is plainly a subjective and introspective way of proceeding, but, as I see it, the only practical one. I have tried to ensure that the sentences cited are typical of their authors' styles by drawing most of the examples from randomly chosen stretches of text, but I have also included sentences from outside those samples which other writers have cited as typical. I appeal to the reader to determine whether he finds the passages difficult, and in the ways described, bearing in mind that they are to various degrees 'pre-analyzed' by being cited in a certain context and by italic printing of the problematic element.

In each of the first four chapters, a particular task, or group of related tasks, will be examined: the first chapter will deal with identification of phrases and their functions; the second chapter, with identification of clause boundaries; the third and fourth chapters, with problems of reference, coreference, and attachment. Chapter Five will, as noted, consider whether we should modify the way we carry out these tasks when we read literature, but Chapter Six returns to ways that texts can present difficulties in the task of integrating material into its context.

Most of the examples are drawn from five writers with deserved reputations for difficulty: Edmund Spenser, John Milton, Henry James, Wallace Stevens, and William Faulkner. Each of these is difficult in some areas but not in others. We can in fact establish a taxonomy of difficulty: a principal source of difficulty in Stevens has to do with appositives; relations of Subject and Object are more problematic in Milton than in the others; identification of main verbs is unusually problematic in James, and so on. Further, we can describe which structural predilections of particular authors create difficulties of perception and comprehension.

The following pages are to be considered an exploratory and tentative account of some processes involved in reading literary texts. It is far from complete and does not consider some of the more complex constructions critics have called attention to such as the 'diabolical' logical syntax in Milton discussed by Stanley Fish or the long strings of adjectives, some of them contradictory, that Faulkner is fond of.²⁰ I would hope that the model sketched here could be extended to describe the processing of these and indeed numerous other constructions.

Asserting that the model outlined here will realize the objectives of generative stylistics leaves a major question having to do with the notion of an assistant unanswered. Does criticism need such a model? How has it fared without one? Many critics, relying only on tact, mother wit, and Empson’s metaphors, have described the experience of reading various authors acutely and insightfully. Studies which focus on the reader’s experience of the writers and works that concern us include those by Paul Alpers on Spenser, Stanley Fish on Milton, and Helen Vendler on Stevens (and we will refer to Stephen Booth’s work on Shakespeare’s sonnets). These critics are among those who have anticipated psychologists and linguists in redefining the focus of their endeavor in terms of realtime experience of texts. With this focus on response has come an interest in the general principles which shape perception and comprehension. Stanley Fish, for one, sees the need for a model of processing which could specify responses by setting forth principles that “restrict (make predictable and normative) the range of response.”²¹ It is precisely these principles (some of them at least) that we undertake to provide and to give thereby some answers to the questions: what sorts of principles guide, or should guide, the reader and critic in reading literary texts?