To date, the application of computing to literary studies has been primarily in stylometry (or stylistics) and in the structural analysis of texts. In these areas, scholars have demonstrated the value of text analysis software to studies of authors and texts. But an important figure has been left out of the field of literary computing: the reader. Fortunately, the tools of stylometry and text analysis appear to be readily adaptable to studies of literary reception -- to the act of reading.  Such an application of these tools stands to lend much needed support to the empirical study of reading.  As Miall and Kuiken have noted,
Despite two millennia of theories..., we are still a long way from grasping what actually happens when a reader understands a text, or whether literary texts perform specific functions that set them apart from other texts. Moreover, the empirical study of these questions has only just begun.... (1994a: 338)
In order to be applied effectively to the empirical study of reading, the tools of computing will require a conceptual framework for their deployment. Any empirical study, after all, requires a theoretical model or framework from which questions and hypotheses can be conceptualized and investigated. Cognitive science can provide some elements for such a framework.  In this paper, I will consider how one tool of literary computing, TACT,  can be applied through a cognitive framework to illuminate one dimension of literary response: the reception of -- or "reader response" to -- the sound patterning of poetry. In considering this use of TACT, I hope to suggest how the computing tools of stylistics in general may be used to illuminate not only authorship but also reading.
The connection between reception theory and cognitive science is in principle a longstanding, although not frequently appreciated, one. Theorists of reception have always been interested in the mental processes of literary response. Since all mental processes are activity represented in neural substrates, and since the object of cognitive science's inquiry is the nature and organization of such cognitive processes and structures, it makes sense that cognitive science should be able to contribute something to an understanding of literary reception. But what can cognitive science tell us specifically about the cognitive activity that constitutes the response to the sound patterning of poetic language? What goes on in our brains when we respond to the sound patterning of a poem? 
At this point, one of the most promising cognitive frameworks seems to be one which allows the reception of poetic sound patterning to be conceived of in terms of different "modes" of response. The notion of different modes of reading seems to be the best way to account for the different responses or interpretations that can be attached to the same set of poetic sound patterns and, ultimately, to the same poem. Since we all have the same basic cognitive capacities, different acts of reception must result from different uses of those capacities -- different uses that result from different orientations to a poem before us, from different reception strategies. Since there are several theories that share concepts similar to the notion of reception strategies or modes of response, I will refer to them generically and call them "cognitive reception orientations" or CROs for short. 
In the CRO framework, different acts of reception are traceable to differences in cognition. The reception of any given poem is a multi-dimensional cognitive activity that involves a combination of different cognitive processes and functions. CROs are what determine the particular combinations. Each CRO, like a strategy or script, brings component cognitive systems into a specific configuration; each configuration results in certain kinds of cognitive activity and leads to a particular experience of a poem's poetic language.
In information processing terms, the sound patterning of a given poem can be seen as data that is to be processed in the brain by using various component processing mechanisms. How the data will be processed depends on how the array of component mechanisms is configured. CROs are what determine the different configurations. They can be thought of as maps for the different "paths" that reading or response can take through various cognitive "sites", or "response centres" in the brain.
What are the different response centres that can be involved in the response to poetic language? A comprehensive list would be unimaginably complex and is beyond the scope of this study but a minimal list would probably include these component cognitive systems: linguistic memory, music cognition, sound symbolism, and problem solving. The reception of poetic sound patterning, regardless of the CRO or type of reception involved, would be expected to involve each of these four basic cognitive systems to some extent. I'll briefly describe each before I consider how it might be possible to use TACT to study their configuration in two CROs: the aural CRO (hearing a poem in performance) and the visual CRO (reading a poem silently).
Memory might be the most fundamental of components in all CROs for response to poetic sound patterning. In order to perceive and respond to them, such sound patterns must be saliently available in memory for any processing or response to them to be significant. Since the seminal work of Sachs, it has become a common assumption in psycholinguistics that memory for the surface form of a text is poor. Since poetic sound patterning is encoded in the surface stratum of poetic texts -- rather than on their semantic or "gist" level --, the case for any account of reading in which sound patterning plays a significant role in response therefore depends on a refutation of this assumption.
Recent research on verbatim or surface-structure memory has emerged with strong evidence of cases where the retention of surface form is higher and longer-lived than has commonly been assumed. This research suggests that, in the reading of poems, memory for surface form is high and robust for two important reasons: the literary reading context and the mnemonic function of poetic form.
Recent studies indicate that the affective value of literary texts (the way they "speak" to us personally, arousing emotions and memories) in general makes their surface form -- their exact wording or phrasing -- important to us.  Emotion and genre expectation are aspects of the context in which we read poetry that heighten our attention to and memory for sound patterning. Shapiro and Murphy, for example, argue that "listeners attend to the level of analysis of text that is most relevant, important, or salient, given their current goals" (1994: 87). They show that when the affective character -- humour, anger, sadness -- of a given discourse is important to a reader, more attention is given to the encoding of surface form in memory. In other words, we have good verbatim memory for texts to which we attach emotions (e.g. jokes, meaningful letters, favourite novels). Being such texts, poems can be assumed to contain important information in their surface structure that readers attend to for encoding in memory.
But affective considerations aside, the literary genre itself appears to trigger response or interpretative acts that pay close attention to surface form. Zwaan (1994) has shown that genre expectations influence the way texts are represented in memory. He shows that readers who are in literary reading mode tend to give a text's surface structure higher priority in memory. In his experiment, two groups of subjects were given the same text but each group was informed differently as to the genre of the text they were to read. Those subjects who were informed their text was literary made the surface structure the most important level of the text's representation in memory.
But the context of poetic reception is not the only factor in determining strong surface-form memory; poetic sound patterning itself has a strong mnemonic function. David Rubin (1995), for example, argues that the mnemonic function of poetic sound patterning plays a key role in the ability of oral traditions to transmit oral texts over long periods of time without significant change in their surface form.
To what extent is poetic sound patterning processed by the cognitive system for music perception and response? Cognitive science generally takes music to represent a non-speech or non-linguistic phenomenon and it typically assumes that the cognition of music is distinct from the cognition involved in the response to "verbal art". As Peretz and Morais argue, recent neurophysiological "data strongly suggest that at least three different systems of auditory recognition exist" (1993: 63). The three distinct systems are one for music, one for speech/language and one for the recognition of environmental sounds. But poets have long associated poetic language with music. Is the link between poetry and music, then, merely an analogical one?
The poets seem to have been on to something. There are, of course, no actual links between poetic language and the harmonic or melodic dimensions of music. But the perception and response to rhythmic patterning is one element of poetic sound patterning that appears likely to involve music cognition. This link is a possibility that has not yet been investigated empirically, but it becomes somewhat obvious when one tries to repeat lines of a poem -- say, a limerick -- while replacing all the syllables with an empty syllable (e.g. "da"). There is clearly some kind of non-speech rhythmic structure that the human ear perceives and responds to in such prosodic patterning.
Part of the response to the poetic patterning of sound also seems to involve perceiving it as expressive, mimetic or symbolic. Alliterative patterns, vowel and consonant alternations, line rhythms -- critics of poetry consistently assume that these and other elements of poetic sound patterning can "express" qualities of speaker voice (irony, fear); or somehow "capture" an action performed by the poem's persona (consonants that "dig" into vowels perform the digging action of the persona described in the poem); poetic sound patterns would also seem to be able to imitate movement (line rhythms contain the push and pull of the "sea") and environmental sounds (sibilants embody the wind's movement through branches). All of these "mimetic" or "expressive" functions are aspects of what is termed "sound symbolism". The nature of our perception and response to symbolic sounds and sound patterns is also a matter of strong debate: how do we perceive and respond to these patterns -- do we infer them or figure them out using acquired knowledge or do we have an inherent sense or innate "feel" for them? On the one hand, linguists and literary critics -- following Saussure -- have tended to argue that the connections between sound and sense are arbitrary, being defined by cultural factors. The perception of links between sound and meaning is an act of association that we have been conditioned or "trained" to perform. On the other hand is a growing body of empirical data that indicates that to a certain extent -- and possibly a large extent -- sounds have "senses" which are intrinsic to the way our brains perceive speech sounds.  Just as, in colour perception, we "see" light waves of different frequencies as "colours", so do we perceive certain sounds and sound patterns to have "sense" qualities (e.g. emotion, colour, brightness). 
It might seem anathema to consider problem solving to be a part of the response to poetic sound patterning, but this capacity of the brain has a stronger role in such response than one might think. Involving, for example, the capacity for expectation, inference, analysis, rule generation, and the application of knowledge structures, problem solving is an innate (and indispensable) capacity of the human brain. In the poetry-experiencer's perception and response to poetic form, problem solving is most visible in what might be called academic interpretation, in which, for example, one might attempt to distinguish historically between the functions of poetic form in period "x" vs. period "y" or to delineate the political encoding of certain forms.
But problem solving -- or at least several major aspects of it -- is an integral part of all complex human activities and the affective response to poetic sound patterning is no exception. It would be difficult to imagine a case in which a reader's prior encounters with poetic sound patterning -- their experience in the classroom or otherwise -- never came into play in their response to a poem at hand. Problem solving is a process by which prior prior experiences (and knowledge inferred from them) are brought to bear in new experiences. The fact that problem solving can be unconscious or automatic only makes it more difficult to rule out of even the most casual and "unproblematical" reading contexts. Consider those instances, even in "recreational" reading, where we encounter minor incongruences in poetic form -- at these points, don't we engage in a little low-level, on-the-spot problem solving? Explicit problem solving (involving relatively intense analysis) would certainly be impossible to rule out in poems written in philosophical or intellectual modes, and in poems that contain ironic or otherwise self-conscious uses of conventional poetic form. But cognitive science has also recently begun to decouple, with studies on the role of emotion in problem solving, problem solving from cold artificiality. As such studies become more substantial, the connections between problem solving and emotion -- and the connections between problem solving and the experience of poetry -- will become clearer. Indeed, problem solving is the primary capacity involved in all forms of play (including parodies, chess, even scholarly close reading), all activities with which emotions can be strongly connected.
A CRO model of poetic response tells us that the patterns of sound in poetic language will be perceived and responded to differently when read with different CROs. In order to test for differential responses to a given poem, patterns in the poem must first be identified. TACT is well suited to such task. I conducted a trial experiment in which TACT was used to identify a very low-level (consciously imperceptible) aspect of poetic sound patterning: phonemic collocation. Differential reception of phonemic collocation was tested for in the visual  vs. the aural CRO.
Two subject groups were presented with the same poem, Rita Dove's "Teach Us to Number Our Days" (1993). The eight subjects -- four in each group -- were graduate or postgraduate members of the University of Toronto English or Comparative Literature Departments. All had relatively high levels of experience with poetry in both creative and academic contexts. None had encountered the poem before. The visual CRO group was presented with the poem on paper. The aural CRO group listened to the poem recorded on tape. For comparison, the two groups of subjects were required to perform an identical set of response tasks. I attempted to curtail aural reception in the visual group by instructing them not to mouth (subvocalize) or recite the words while reading.
After two passes through the poem, both groups answered a series of questions which tested their memory for phonemic collocate associations. Phonemic collocates are phonemes that tend to occur near to each other. TACT was used to analyze a phonemic transcription of the poem and obtain the collocation data. TACT was able to indicate, as it were, the strengths of association between different phonemes in the poem. For any given phoneme, TACT could indicate which other phonemes were most likely to be found near it and which other phonemes were most likely not to be found near it.  I used TACT to select three phonemes from the poem that had strong collocates -- whether the collocation was one of strong "attraction" or "repulsion". As can be seen in Table 1 below, for each selected phoneme, two strong/weak pairs of collocates were chosen (one pair consisted of vowel collocates, the other, of consonants).
|Selected Phoneme||Strong Collocate||Z-Score||Weak Collocate||Z-Score|
|/iy/ (as in eat)||/ey/ (wait)||+ 2.013||/u/ (nut)||- 1.089|
|/iy/ (as in eat)||/ch/ (church)||+ 5.319||/r/ (run)||- 0.441|
|/a/ (as in odd)||/o/ (home)||+ 1.712||/iy/ (eat)||- 1.321|
|/a/ (as in odd)||/r/ (run)||+ 2.395||/t/ (top)||- 0.218|
|/ae/ (as in at)||/iy/ (eat)||+ 1.340||/i/ (sip)||- 1.392|
|/ae/ (as in at)||/n/ (nine)||+ 3.212||/b/ (ban)||- 0.222|
After their two passes through the poem, subjects were presented with each of the selected phonemes in separate questions. For each selected phoneme they were asked to consider one strong/weak pair of collocate phonemes (vowels) and to decide which collocate phoneme they felt was more strongly associated with the selected phoneme. In a subsequent question, subjects considered the same selected phoneme but chose between the other strong/weak pair of collocate phonemes (consonants). This procedure was repeated with the remaining two selected phonemes. For each question, the instructions reminded the subjects that their collocation decisions were to be based on what they felt to be the association between phonemes in the poem rather than between phonemes in general. For each answer, subjects rated their confidence in their decision. The following is a sample question:
Consider the phoneme /ee/ as in eat or jeep. Of the two phonemes below, /ee/ is more strongly associated with
/ey/ as in hate or wait
/u/ as in nut
Rate your confidence in your answer: 1 2 3 4 5 (1 = no confidence; 5 = practically certain)
The results are summarized in Table 2 below.
|"Correct" Responses||Confidence Avg.||"Correct" Responses||Confidence Avg.|
|/iy/ with vowel collocates||4 / 4||4.25||3 / 4||2.0|
|/iy/ with consonant collocates||2 / 4||3.5||1 / 4||2.25|
|/a/ with vowel collocates||3 / 4||3.0||1 / 4||3.0|
|/a/ with consonant collocates||2 / 4||2.25||0 / 4||1.75|
|/ae/ with vowel collocates||2 / 4||2.5||2 / 4||2.0|
|/ae/ with consonant collocates||3 / 4||2.0||2 / 4||1.5|
|Visual:||67% "correct" responses (16/24)||Aural:||38% "correct" responses (9/24)|
|2.9 Avg. Confidence||2.1 Avg. Confidence|
While there is a great deal of variation within the scores and ratings for each of the test phonemes, the visual group was consistently more accurate than the aural group. The averages confirm this pattern. This would seem to suggest that memory for phonemic collocates is better in visual rather than aural reception. Stronger generalizations are plausible but will not be conclusive until further testing is completed. Does differential memory for phonemic collocates indicate a differential configuration of linguistic memory in a visual CRO? Why would this be so? Questions such as these will have to wait, for two main reasons. First, the phonemic collocate test is an as yet not investigated methodology.  More importantly, though, the subject sample of my experiment is too small to be confident that the differences between subject groups would replicate. Nonetheless, the findings of this trial are quite suggestive and warrant further and more substantive testing using or extending this methodology and theoretical framework. Similar tests and measures could be developed to study the involvement of other cognitive systems (sound symbolism, music cognition, problem solving) that are active in the response to sound patterning in poetic language. This kind of study would help us to understand better the differences between the visual and aural reception of poetic language.
The limited use  of TACT that I have reported on here is meant merely to signal the much larger potential the resources of literary computing have to aid the empirical study of literary reception. All of the computing tools of stylistics have the potential to serve a double function: to illuminate not only authorship but also reception. The methodological principles by which TACT was used in this study can be extended to make similar use of practically any of the computational tools of stylistics. The uses for such software in the empirical study of literary reception divide into two categories. On the one hand, they can be used, as in the experiment reported here, to identify systematically structures or elements of texts for which different reader responses can be empirically studied. On the other hand, they can also be used to analyze structures or elements of texts that readers identify as significant. Differences in patterns of identification could be examined between readers or between different reading contexts or practices. In other words, we can use the computational tools of stylistics to generate patterns we'd like to test on readers, and we can also use them to help us analyze the patterns which readers themselves mark out as important or striking.
Whether we are studying the reception of poetic language  or plot structure, the point remains that literary computing shouldn't be limited to accounts or analyses of regularities or linguistic structures in texts. Those accounts and analyses can be used to study different acts of reception. If a program such as TACT can give us a map, as it were, of where all the features or structures in a text occur, why not use that map to examine the routes that different acts of reading (or CROs) will trace over it? Texts, after all, are not just sites of authorship but also sites of multiple and different readings.
 This paper is based on "Sound Sensing: Cognitive Science and the Response to Poetic Form", which was read at the Consortium for Computing in the Humanities (COCH/COSH) sessions of the 1997 Canadian Congress of Learned Societies. I wish to thank Ian Lancashire for his immeasurable guidance. I am also grateful to Carol Krumhansl and an anonymous reviewer whose comments have aided in revisions.
 Reading here is meant -- in the broad semiotic sense -- to refer to the interpretative acts and/or responses that constitute a person's experience of any literary "text", whether the "text" is a poem, novel, film or play.
 This is something which literary theorists unfortunately have been largely uninterested in discussing or pursuing. I say "unfortunate" because any theory of reading is based, after all, on some kind of data, whether the collection of this data is implicit or declared, whether it consists of internally monitoring one's own act of reading or whether it consists of intuiting patterns from one's observations of others' reading practices. When pressed, most reception theorists seem to claim that their theories are heuristic in value (see Iser 1989: 49) -- basically, "I'll generate the ideas; let someone else test them" -- or that reading (like the concept of the author) is a conceptual device used to theorize reading (see Mailloux 1982: 192) and is therefore not important or impossible to examine as an "actual" or "real" process of the brain that can be scientifically studied.
 The applications of cognitive science to literary studies are not limited to reader theories. The increasing frequency with which cognitive approaches are appearing in literary scholarship indicates that, as F. Elizabeth Hart argues, "the time has come for an exploration of the specific ways cognitive science can augment literary theory" (1995: 28).
 TACT stands for Text Analysis Computing Tools and is a software suite that was developed at the University of Toronto. The manual was published in 1996 by the MLA. See Lancashire 1996.
 My conception of cognition -- cognitive activity -- here is a broader one which includes emotions and other bodily phenomena (kinaesthetic response) that cognitivists have tended not to address. The cognitive psychology of emotion is an expanding field. Present and future studies of the influence of emotion in other dimensions of cognition, such as memory and cognitive organization (see Isen, Shalker, Clark & Karp 1978, Isen 1990), will likely contribute to the increasing capacity of cognitive science to provide insights into aesthetic experience. In any case, many interesting discussions on the value of cognitive science to aesthetics are sure to follow. For a recent example specific to literary study, see Herbert Simon's five-part "'Bridging the Gap': Where Cognitive Science Meets Literary Criticism" and the thirty-three published responses to it.
 Roman Jakobson's notion of a "set to the message" (or Einstellung; 1960) might be the earliest. Iser's notion of reader repertoires shares a kind of analogical similarity (1989). Script theories in discourse analysis are generally similar and more recently, Rolf Zwaan's "cognitive control systems" (1993) and Christopher Collins' "cognitive modes" (1991) are theorized explicitly in terms of cognition. Also consider Reuven Tsur's "poetic mode of speech perception" (1992).
 In a study of reading times, Miall and Kuiken found that poetic features -- "foregrounding" -- increase reading times, indicating that readers attend more closely to a literary text's surface structure in which these features occur (1994b).
 See Hinton, Nichols & Ohala 1994.
 Ivan Fonagy researched universal phonological symbolism in these categories during the 1970s. Reuven Tsur discusses these in relation to cognitive science (1992).
 Visual or "silent" reading is unlikely to be purely visual -- utterly devoid of aural processing. The visual processing of alphabetic script is likely to involve access to aural processing centres (see Parkin 1996: Chapters 7, 8). Cognitive neuropsychological evidence does suggest, however, that alphabetic script can be "read" for meaning without access to aural processing. The evidence comes from testing of individuals (stroke victims with brain lesions or the congenitally deaf) in whose brains the link between graphic and aural processing systems is blocked. But no evidence exists to show that visual language processing in "normal" individuals can exclude -- by choice, as it were -- access to aural processing. This means that even the most visual poetries (e.g. concrete poetry) are likely to involve some form of "hearing" simply because all language processing includes some access to aural processing centres. Further arguments for the aurality of "silent reading" can be made using research on subvocalization.
 The measure TACT uses to determine collocate strengths is a Z-score. A Z-score is a standard measure of the unlikelihood that a collocation is purely random. The higher above or below zero the score, the less random the collocation relationship.
 The phonemic collocate task may privilege visual reception. Why? One possible explanation is that visuality in poetry may be like literacy in general (Olson 1994): it encourages metapoetic conceptualization. This makes it easier to perform what is essentially an abstract, problem solving task: accessing an abstract representation of the poem in memory and searching it for relationships in its structure. A "phoneme" is itself an abstraction of language (as is an alphabet). But the privileging of visual reception in the phonemic collocate task may also be explained by considering how the oral presentation of poem tends to discount the listener's own sense of "voice", thus implicating less effectively in the listener the affective memories or metaphors upon which phonetic response and memory may be based.
 TACT, for example, can do much more than analyze phonemic collocates. My analyses of the phonemic transcription of the poem using TACT yielded other significant patterns which were not incorporated into the test questions. Beyond its abilities to search out formal sound patterns such as alliteration and rhyme, TACT, for instance, was able to locate unusual phoneme concentrations in specific lines. TACT also includes a program that finds full and partial anagrams -- two words being anagrams of each other if they contain the same set of letters in different combinations. Orthographic anagrams, as with sight rhymes, are aspects of poetic form that are likely to be purely visual and thus insignificant in the aural CRO. More interesting, though, are phonemic anagrams -- such as "eat" and "tea". To what extent might phonemic anagrams at the syllabic level be significant in the response to poetic sound patterning? Would they be more significant in the aural or the visual CRO? TACT could generate thorough lists of all these and other patterns for testing.
 TACT is only one of several promising technologies in the study of the reception of poetic language. The continuing development of automated metrical scansion (e.g. Robey 1993, Beaudouin 1996) and phonological transcription software (see Beaudouin 1996) will greatly extend such study.