Difficult poetry processing: Reading times and the narrativity hypothesis

Abstract

This study presents an experiment that uses reading times as a measure of the processing effort demanded by ‘difficult’ poems, where difficulty is defined as a text-driven response phenomenon associated with resistance to reading fluency. Reading times have been used before to explore the processing of literature, but seldom with the aim of shedding light on difficulty. There is then scope to redress this research gap, also in light of Shklovsky’s claim that the technique of art is ‘to increase the difficulty and length of perception’. In the current experiment, a group of participants read six poems on-screen. The poems are by Mark Strand, Ezra Pound, Wallace Stevens, Geoffrey Hill, Susan Howe and Jeremy Prynne, and have been selected based on critics’ remarks on their difficulty or lack thereof. An extract from a novel by JG Ballard was also included to find out how its narrativity would compare, in processing terms, to the more elliptical narrativity of Strand’s and Pound’s poems. The time spent on each line was recorded by software E-Prime, commonly used in psycholinguistics. The results indicate that three of these texts – Ballard’s, Strand’s and Pound’s – were read at a much higher speed than non-narrative poems by Stevens, Hill, Prynne and Howe. The proposed explanation was that it is sufficient for readers to recognize traces of a narrative schema to read the text fluently, even if such text is low in coherence. By contrast, when prototypical narrative features cannot be mapped onto a text, the processing effort as measured by reading times remarkably increases. Overall, the results refine our understanding of the relationship between difficulty and the stylistic strategies associated with it.

Keywords

Difficulty empirical stylistics narrativity poetry processing reading times

1. The study of poetic difficulty and the need for empirical stylistics

Although the concept of literary difficulty ‘begins with commentary on Homer and Holy Scriptures’ (Adams, 1991: 23), the rise of critical studies on difficulty parallels the rise of Modernism. In his monograph The Difficulties of Modernism, Diepeveen offers the following definition:

Difficulty was the experience of having one’s desires for comprehension blocked, an experience provoked by a wide variety of works of art (“comprehension” is here defined broadly)’’ (Diepeveen, 2003: x).

In other words, difficulty is a response phenomenon and as such it involves the experiential dimension of reading. Yaron also appeals to readerly experience when studying the processing of difficult poems, arguing that ‘a poem is considered difficult if the representation constructed by the reader is defective’ (Yaron, 2008: 146). Despite the use of the negatively connoted adjective ‘defective’, Yaron does not intend to criticize or discredit the experience of difficulty, but instead she emphasizes the incomplete, incoherent or fragmentary nature of readers’ posited mental representations prompted by difficult poems. In a helpful though hardly intuitive distinction between obscurity and difficulty, White proposes that the former is a matter of ‘obscured vision’, whilst the latter ‘applies largely to the language of a text, its style or syntax’ (White, 1981: 17). This suggestion implies that stylistics is ideally equipped to investigate difficulty, and this has already been the case to an extent.

Indeed, scholarship on poetic difficulty falls within three main traditions: typological; stylistic; and reception-oriented. In reviewing each tradition, I argue that empirical stylistics can encompass all three, thus emerging as the most suitable framework for the study of poetic difficulty. The second section outlines a new stylistic–empirical model of poetic difficulty, focusing on one of its most interesting dimensions – narrativity. The third section reports the results of an empirical test showing that the presence and density of stylistic features associated with narrativity accounts well for the amount of time readers spend on poems. More precisely, the more easily a poem can be mapped onto a narrative schema, the quicker and presumably more effortless its online processing will be. This explanation rests on the empirical evidence that ‘narrative passages […] are easier to read than passages in the descriptive and expository genres’ (Graesser et al., 1980: 138), and as a consequence ‘the superstructure that people are most familiar with and handle most easily is the narrative schema’ (van Dijk and Kintsch, 1983: 252). Additional stylistic remarks will help explain processing differences among non-narrative poems.

Far from standing in self-contained camps, these three traditions interact with each other. Both stylistic and reception-oriented accounts of difficulty originated from the Russian Formalism of Shklovsky (1998 [1917]) and the Practical Criticism of Richards (1929) and Empson (1930), also a pioneer of typological approaches. Even so, each tradition appears guided by different premises: the poetic text as an observable object (stylistic approaches); the poetic text as an experienced event (reception-oriented approaches); and the poetic text, either at the object or at the event pole, as instantiating broader aesthetic or cultural categories (typological approaches).

1.1 Typological approaches to poetic difficulty

I begin with typological approaches since these lay out the foundations for the other two traditions. The first landmark is Empson’s Seven Types of Ambiguity (1930). Empson begins by providing a broad definition of ambiguity which brings it close to folk intuitions about difficulty, and then outlines his types by conjecturing a relationship between tropes and posited reader response. For instance, his first type of ambiguity features ‘comparison’ and ‘false antithesis’ (Empson, 1930: 28) since these tropes ‘leave it to the reader vaguely to invent something’ (Empson, 1930: 29). The second type occurs whenever ‘two or more meanings are resolved into one’ (Empson, 1930: 62) and are mainly exemplified through syntactically ambiguous passages. The conflict versus coexistence of meanings (the first and second types respectively) seems to underpin the taxonomy.

Three decades later, Press (1963) developed Empson’s typological enterprise by offering a wide-ranging perspective on poetic obscurity. He investigates both the textual side of poetry and its contextual embeddedness, as witnessed by the titles of each chapter: ‘vocabulary’; ‘syntax’; ‘reference’; ‘themes’; ‘images’; ‘common readers’; ‘indifferent writers’; ‘private countries’; and ‘public worlds’. Such a catalogue of topics enables Press to effectively categorize his arguments, but prevents him from showing how obscurity works holistically – that is, how each facet interacts with the others.

The typological tradition was revived in 1978 by Steiner’s influential essay ‘On Difficulty’. Steiner’s starting point is the common-sense question ‘what do we mean when we say: “this poem, or this passage in this poem is difficult?’’’ (Steiner, 1978: 18; emphasis is the author’s). To answer it, he develops a fourfold typology. Contingent difficulty stems from a mismatch between the encyclopaedic knowledge presupposed by a text and that available to readers, who therefore need to ‘look up’ a word, an expression or an allusion. Modal difficulty ‘lies in the eye of the beholder’ (Steiner, 1978: 33), having to do with the reader’s inability to cope with unfamiliar modes of expression. Tactical difficulty ‘has its source in the writer’s will or in the failure of adequacy between his intention and his performative means’ (Steiner, 1978: 33) and is mainly exemplified through syntactically ambiguous passages. Finally, ontological difficulty ‘confront[s] us with blank questions about the nature of human speech, about the status of significance’ (Steiner, 1978: 41). This type stands at the furthest remove from the text and it is not therefore exemplified.

Overall, while typological studies bring together considerations on language, readers and context, their taxonomies are very broad and introspection-driven, and fail to account for the diversity of texts, readers and contexts.

1.2 Stylistic approaches

The richness and diversity of poetic texts as linguistic constructs which might determine difficulty is by contrast the purview of stylistic approaches. Nowottny, for instance, commenting on the shifting tenses in Dylan Thomas’s ‘There Was a Saviour’, argues that ‘it is difficult to make out at first what point in time we are supposed to be looking at or looking from’ (Nowottny, 1962: 188–189). Deictic shifting as a stylistic strategy to either involve or distance readers from the text-world of contemporary poems has also been the focus of a more recent study by Jeffries (2008). Importantly for the overall scope of the present paper, she proposes that there might be a relationship between difficulty and the diminished readerly involvement prompted by certain uses of deixis (e.g. ‘you’ with generalized reference vs lyric ‘I’). Taking a broader perspective and relying on the theory of foregrounding, Leech (2010: 116) emphasizes the relationship between linguistic deviation and enhanced interpretive effort: ‘dysfunctional features, e.g. distortions of linguistic norms, obscurities and ambiguities, challenge the reader to find new avenues of interpretation’ (Leech, 2010: 116).

Dillon’s 1978 monograph on the processing of literature is marked by a linguistic–typological commitment:

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 others; identification of main verbs is unusually problematic in James, and so on. (Dillon, 1978: xxx)

Dillon’s comparative work appears to inspire Chafe (1991), reporting how a group of readers unanimously deemed an extract from Henry James more difficult than one from Edith Wharton. Building on Chatman (1972), whose fundamental work on James’s later style focused on the stylistic basis of abstraction (e.g. nominalizations, nouns with indefinite reference, and mental verbs), Chafe traces James’s difficulty back to the following stylistic markers: decontextualized proper names; syntactic embedding; negation; passive constructions; and nominalizations.

In a pedagogical vein, Toolan (1993) shows how retracing patterns of lexical cohesion in a difficult poem by Geoffrey Hill can help readers overcome their initial disorientation. Toolan interprets unusual collocations by relying on paradigmatic observations:

the odd phrase ‘fatted marble’ surely calls to mind […] the only common collocate of ‘fatted’, in ‘fatted calf’; and the latter expression’s association with (propitiatory?) sacrifice has a complex relevance both to this stanza and the whole poem. (Toolan, 1993: 39)

While Toolan’s bottom-up approach extends from text to context Sell (1993) applies ideas from pragmatics in top-down fashion. He suggests that Eliot’s The Waste Land flouts each of the four maxims (quality, quantity, relation, and manner) of Grice’s cooperative principle (Grice, 1975). The effect of this strategy Sell concludes, is that readers need ‘to do a lot of the sense-making for themselves’ (Sell, 1993: 138).

Adamson analyses a wealth of poetic excerpts that showcase the linguistic innovations brought on by Modernism, which she relates to ‘the deliberate courting of difficulty in Modernist aesthetics’ (Adamson, 1999: 643). The lens through which many of these innovations are interpreted is that of an estranging mimesis of spoken language:

Olson’s topic-skipping may well be the most naturalistic in reproducing the associative leaps and self-interruptions we all practice when we talk to ourselves or to an intimate friend, but put in writing and addressed to a public audience, it strikes many readers as a perversely difficult form of communication. (Adamson, 1999: 597)

The apparent contradiction between poets’ interest in rendering speech and their sometimes forbiddingly difficult style is also explored by Milroy (1977), who focuses on Gerard Manley Hopkins’s dialect vocabulary, word-formation strategies, sprung rhythm and syntactic structures whose reliance on ellipsis is reminiscent of colloquial English.

In summary, while stylistic approaches have been invaluable in detailing the linguistic basis of difficulty, the attribution of difficulty itself typically rests on the response of one reader only – the critic. Independent empirical validation is therefore needed to find out whether these attributions can be generalized to account for other readers too. In other words, what tends to be overlooked is the crucial question of ‘how readers feel when confronted with particular stylistic devices’ (Hakemulder and van Peer, 2016: 190).

1.3 Reception-oriented approaches

The first part of Hakemulder and van Peer’s statement (‘how readers feel’) is the chief concern of reception-oriented scholars. Their stance with regard to difficulty is well exemplified by Fish:

Of course, the difficulty is itself a fact – of response; and it suggests, to me at least, that what makes problematical sense as a statement makes perfect sense as a strategy, as an action made upon a reader rather than as a container from which a reader extracts a message. (Fish, 1980: 23)

While for some stylisticians (e.g. Toolan) linguistics democratically enables us to approach challenging texts, Fish believes that approaching difficulty as a meaning-retrieval task is misguided because the difficulty exists to be experienced, rather than tamed; similarly to Fish (1980), Diepeveen contends that difficulty ‘arises from the extreme responses it elicits, provoking people to anxiety, laughter, or anger’ (Diepeveen, 2003: 45). Sutherland reiterates the point along similar lines: ‘difficult art baffles, intimidates, stretches and upsets’ (Sutherland, 2010: 767).

Reception theory at its most radical can be seen in the essay collection The Idea of Difficulty in Literature (Purves, 1991). Here all the authors emphasize the non-textual – or only secondarily textual – nature of difficulty. A couple of examples will suffice: Purves considers difficulty ‘a social construct’ (Purves, 1991: 2), whilst for Hynds ‘“difficulty” is not a feature of particular texts, but the result of the similarity or disparity between dimensions of the text and the socially embedded and motivated interpretive processes of particular readers’ (Hynds, 1991: 117).

This receptionist standpoint is openly endorsed by Yaron’s (2002, 2003) empirical work. By comparing responses to an ‘accessible’ poem by Mark Strand and a ‘difficult’ one by E. E. Cummings, she found that participants faced with Strand’s poem recalled complete textual sequences (clauses and sentences), whereas those faced with Cummings’s recalled shorter sequences (words and phrases). She also found that the length of the sequences inversely reflected the accuracy of their recall: readers tended to re-paraphrase passages of the Strand poem, but left Cummings’s recalled units unaltered (Yaron, 2002: 150–151). Yaron’s finding is insightful in terms of processing, but this observed reading behaviour would have been better explained through a stylistic scrutiny of both poems from which Yaron refrains.

Overall, reception-oriented studies have little to say about how stylistic features impact on readers with regard to perceptions of difficulty. This lacuna is surprising, given the abundant evidence provided by stylistic approaches. That difficulty has a strong linguistic basis is hard to dispute; that difficulty is not wholly immanent but it needs an experiencing reader to be felt and assessed is also hard to dispute. This is where empirical stylistics comes in.

2. A stylistic–empirical model of poetic difficulty and the narrativity hypothesis

A stylistic–empirical model should detail as systematically as possible the linguistic cues and readerly outcomes of the (literary) phenomenon to be investigated. Ideally, it should enable and test predictions about the prototypical response prompted by different stylistic configurations.

The model I set out here distinguishes between offline and online difficulty. Offline difficulty is a post-reading feeling of incomplete or unsatisfactory understanding, hence overlapping with Diepeveen’s (2003) definition. It is related to the ‘elusiveness’ of poetic meaning (Lamarque, 2009), calling either for hermeneutic labour or, alternatively, for an experiential, less point-driven reading mode (e.g. Fish, 1980). Online difficulty, on the other hand, is the real-time effort experienced when reading a poem, and is therefore associated with textual resistance. It rests on the psycholinguistic assumption that ‘the amount of time spent reading the region […] provides an indication of processing difficulty’ (Sanford and Emmott, 2012: 50). Compared to offline difficulty, online difficulty involves more basic, text-driven cognitive operations such as decoding word meanings and parsing structures. That is why a stylistic–empirical approach is conducive to assessing online difficulty. Of course, the offline–online distinction is only a first step towards a more nuanced understanding of difficulty, and much more work needs yet to be done.

My review of stylistic approaches revealed that several linguistic features are deemed responsible for difficulty. In a previous paper (Castiglione, 2013) I examined a number of remarks by critics and stylisticians in order to develop an extensive linguistic checklist associated with difficult poems. In order to place the current study within its original context, an updated and refined overview of such features, called linguistic aspects of difficulty (LAD henceforth) is provided below. Lack of narrativity, the aspect I will be focusing on in this paper, is number 20 in the list (see Table 1).

Table 1.

Linguistic aspects of difficulty.

Stratification and structure	Linguistic aspect of difficulty	Processing operations challenged/elicited
Graphology (words to whole text)	1. Non-standard orthography 1a. Eye-dialect 1b. Creative misspellings	Word recognition
	2. Typographic foregrounding 2a. Unusual layout2b. Capitalization, italics2c. Special fonts, font size2d. Insertion of non-linguistic signs2e. Blank spaces	Inferencing (interpretation of foregrounding)
Morphology (morphemes)	3. Morphological deviation 3a. Word-compounding, neologisms3b. Ungrammatical word derivations	Word recognition, decoding, integrating
Phonology (syllables)	4. Adjacent stresses (spondaic rhythm)	Subvocalization (phonological mediation)
Lexicogrammar/ Semantics (word to clause complex)	5. Low-frequency, technical words (full semantic representation often out-of-reach)6. Abstract nouns (richer semantic representation than concrete nouns)7. Proper names 7a. First names, decontextualized (generic, void referents)7b. Full names of existing individuals (homophoric, encyclopaedic referents, e.g. John Adams)	Decoding
	8. Lexical ambiguity 8a. Two grammatical classes8b. Two distinct referents	Parsing + Decoding
	9. Word combinations (semantic clashes) 9a. Restricted collocations (inappropriateness)9b. Collocation contradictions (paradox)9c. Violation of selectional restriction (incongruity)	Decoding + Integrating
	10. Non-standard word order 10a. Marked order (fronting, inversions)10b. Ungrammatical string (e.g. ‘boy the run who’)	Parsing + Integrating
	11. Use of negation	Integrating
	12. Novel metaphors and figurative language 12a. Semantic distance between source and target domain12b. Inversion of domain mappings (concrete to abstract)12c. Other tropes with non-literal meaning (i.e. irony, sarcasm, and allegory)	Integrating + Inferencing (interpretation of foregrounding)
	13. Lack of main constituents 13a. Omission of essential constituents (e.g. ‘Gertrude Stein’s ‘Never which when where to be sent’ [who, what?])13b. Ellipsis of inferable constituents (e.g. Clark Coolidge’s ‘where [there is] no one’)	Parsing + Inferencing (interpretation of foregrounding)
	14. Noun phrase juxtaposition (nominal style) 14a. List (contingent relatedness of nouns)14b. Apposition (semantic relatedness of nouns)	Integrating + Inferencing
	15. Complex syntax 15a. Trailing and anticipatory constituents15b. Multiple embedding15c. Subordination15d. Parentheticals15e. Unclear clause boundaries15f. Garden-path constructions (i.e. sentences with two possible parsing routes)	Parsing + Integrating
	16. Lack of coherence 16a. Title either missing or misleading16b. Lack or misuse of connectives16c. Exophoric (decontextualized) referents16d. Lack of recognizable semantic domains/schemata16e. Weakly implicated/unrelated clauses16f. Shift of tenses, pronouns and viewpoints16g. Clash of registers	Integrating
	17. Line/syntax interference (information distribution disrupted by enjambments)
	18. Intertextuality/cultural allusions	Recognition of allusion + Inferencing
	19. Lack of subjectivity/presence of objectivist stance (lack of immersion)19a. Denotative language19b. Passive constructions19c. Lack of attitudinal/emotive/evaluative lexis19d. Lack of deixis or estranging use of deixis19e. Lack of progressive forms of verbs19f. Poetic persona concealed19g. No human presence or depersonification19h. Prominence of inanimate objects(minor empathetic recognizability)	Mind-modelling/attribution of intention
	20. Lack of narrativity 20a. Structural distance from prototypical narratives20b. Few or no linguistic markers of narrativity (past/present tenses, proper names, personal pronouns, adverbials of time and space)	Prototypical literary schema blocked

Linguistic aspects of difficulty cut across all linguistic levels, from orthography to semantic/pragmatics. Following key terms in systemic-functional grammar (Halliday, 2014), they are ordered according to stratification (linguistic levels) and structure – from syllables and morphemes up to clause complex and the whole poetic text. The right-hand column of Table 1 specifies the cognitive processes likely to be challenged by each LAD: word recognition (i.e. interpreting a string of letters as an existing word); decoding (assigning meanings to words); parsing (assigning thematic roles and grammatical class to words, interpreting syntactic structures); integrating (building a global mental representation for the text); and inferencing (making sense of aesthetically foregrounded patterns through interpretive acts). For a more technical definition of such operations, see Kintsch (1998) and Harley (2008). LADs seek to pinpoint the linguistic basis of a specific type of readerly perception – that of difficulty, both offline and online. Of course, since LADs are qualitative and context-sensitive, the checklist provides a preliminary analytical orientation rather than a formula to score the difficulty of poems. As mentioned earlier, in this paper I focus on lack of narrativity, as this is the LAD that emerged most forcefully from the examination of the reading time measurements discussed in Section 3.

Compared to most fictional writing, the narrativity of poetry is often present in weakened forms. Pound’s poetry, for example, has been described as a ‘fractured narrative’ (Nadel, 2007: 61) or as displacing ‘linear narrative of cause and effect’ (Moen, 2010: 296). Similarly, McHale defines John Ashbery’s ‘The skaters’ as ‘fragmented narrative’, while Susan Howe’s The Europe of Trusts he considers as ‘antinarrative’ (McHale, 2004: 6). How to stylistically rank the narrativity of poems is an issue I tackle in Section 3.2.

3. Narrativity in poetry and reading times: an empirical test

3.1 Aim

Reading time measurements in the empirical study of literature have served a variety of purposes: from investigating readerly expectations (Zwaan, 1993 [1991]) to detecting response differences across genres (Hanauer, 1998) and within genres (Carminati et al., 2006); from testing the effects of stylistic features (Mahlberg et al., 2014; Sanford and Emmott, 2012) to exploring matters of affect and foregrounding (Miall and Kuiken, 1994). This last study is partially comparable to the current one as Miall and Kuiken found that ‘slightly different components of foregrounding were predictive of reading times’ (Miall and Kuiken, 1994: 403). However, only once throughout their entire paper do they comment on the relation between processing times and stylistic differences (Miall and Kuiken, 1994: 403).

The current test aims at integrating the above picture by looking at reading times as manifestations of online reading difficulty. At this point an objection may be raised concerning the adequateness of reading times to capture reading difficulty rather than other response phenomena. As Wallot and colleagues aptly put it: ‘a certain level of reading speed can be reflective of both a time investment (careful reading, rereading, etc.) and severely compromised reading activity (problems with decoding, prolonged but unresolved uncertainty, etc.)’ (Wallot et al., 2014: 1751). Although in principle there seems to be no way to discriminate between these two functions, an informal reading survey I had previously conducted suggests that reading times in my sample of poems are indicative of ‘compromised reading’ (online difficulty) rather than of ‘time investment’ due to other causes (e.g. aesthetic appreciation). In that survey, another sample of readers was asked whether or not they enjoyed the poems. I found a strong tendency for poems assessed as difficult to be enjoyed the least, thus pointing to the latter of the two functions mentioned by Wallot and colleagues.

Having built up a case for reading times as response indicators of online difficulty, I am interested in discovering whether there are appreciable differences in the online difficulty of narrative, semi-narrative and non-narrative poems. In doing so, I wish to address Yaron’s main criticism of reading time studies, namely, their neglect of full texts (Yaron, 2002: 134). Where applicable, I will briefly assess the influence of other LADs on the degree of online difficulty displayed by stylistically heterogeneous poems.

3.2 Method

Materials

The materials comprise six poems and one narrative extract. Table 2 offers an overview of key bibliographical details and line length. Each text will be referred to by its author’s initials (see Identification column) for ease of reference.

Table 2.

The seven literary texts tested.

Identification	Author	Title	Collection/work	Year	Length
MS	Mark Strand	‘The Late Hour’	The Late Hour	1978	21
EP	Ezra Pound	Extract from Canto LXXXI	Pisan Cantos	1948	16
WS	Wallace Stevens	‘What we see is what we think’	The Auroras of Autumn	1950	18
JP	Jeremy Prynne	Lobster-orange, shag in parvo….	Pearls that were	1999	10
GH	Geoffrey Hill	33	Speech! Speech!	2000	12
SH	Susan Howe	A small swatch bluish green…	Bed Hangings	2001	8
JGB	J.G. Ballard	Opening of The Drowned World	The Drowned World	1962	18

The criteria considered for the selection were ‘difficulty’ and ‘narrativity’. Regarding difficulty, the first requirement is that one of the poems functions as an accessible baseline against which to assess the difficulty of the others. I have chosen Mark Strand’s ‘The Late Hour’ (henceforth: MS), whose accessibility has already been demonstrated by Yaron (2002, 2003). As the extract below illustrates, in this poem a heterodiegetic narrator evokes the end of a relationship by exploring the male lover’s psychological state of obsession and desperation. He is the main focalizer of the events:

A man walks towards town,

a slack breeze smelling of earth

3 and the raw green of trees blows at his back.

He drags the weight of his passion as if nothing were over,

as if the woman, now curled in bed beside her lover,

6 still cared for him.

The authors of the remaining poems have been regarded as difficult by literary critics. At this stage, Diepeveen’s broad definition of difficulty as impeded comprehension (see Section 1) was used. Pound is deemed difficult by Steiner (1978: 22–27), Tuma (1998: 47), Mellors (2005: 33) and Nadel (2007: 63). An extract from the famous Canto LXXXI, ll. 42–57 (Pound, 1954) (henceforth: EP) was chosen for the study since it displays ‘elusive connections’ (Perloff, 1985: 9) alongside cultural references and a congeries of personae. In EP, named statesmen and literati are involved in a seemingly dialogic exchange that, on closer inspection, turns out to be a series of juxtaposed utterances violating the Gricean maxim of relation:

47 ‘You will find’ said old André Spire,

‘that every man on that board (Crédit Agricole)

has a brother-in-law.’

50 ‘You the one, I the few’

said John Adams

speaking of fears in the abstract

53 to his volatile friend Mr. Jefferson,

(to break the pentameter, that was the first heave)

or as Jo Bard says: ‘They never speak to each other,

56 if it is baker and concierge visibly

it is La Rochefoucauld and de Maintenon audibly.’

Stevens ‘is hard to understand’ according to Serio (2007: 1), and Steiner illustrates his ‘tactical’ difficulty category through one of his poems (Steiner, 1978: 38). Quoting the two final lines of ‘Anecdote of the Jar’ (‘It did not give of bird or bush, / Like nothing else in Tennessee’), he claims that ‘no reading of the text has come up with a coherent parsing or equivalent transposition into normal syntax’ (Steiner, 1978: 39). The Stevens poem I have selected (henceforth: WS) comes from the late collection The Auroras of Autumn, in which, according to Leggett (2007: 62), the widely perceived difficulty of Stevens is manifested more forcefully than elsewhere. Opened by a defamiliarizing metaphor and a coinage, WS unfolds a meditation on the distinction between imagination and reality. Its stylistic fulcrum is an extended noun phrase series that can be interpreted as either a list or an apposition (exemplified by lines 7–9 in the extract below):

At twelve, the disintegration of afternoon

Began, the return to phantomerei, if not

3 To phantoms. Till then, it had been the other way:

One imagined the violet trees but the trees stood green,

At twelve, as green as ever they would be.

6 The sky was blue beyond the vaultiest phrase.

Twelve meant as much as: the end of normal time,

Straight up, an élan without harrowing,

9 The imprescriptible zenith, free of harangue,

As for Prynne, his work has been claimed to enact a readerly experience typical ‘of any difficult literary work’ (Mellors, 2005: 167). In keeping with Mellors, Lopez (2006: 135) writes that ‘much has been made of the difficulty of Prynne’s poems’. After reading Prynne’s voluminous anthology Poems (2005), I have chosen the collagist ‘Lobster orange, shag in parvo’ (henceforth: JP). I was attracted by the sudden shifts in grammatical subjects, the presence of syntactic and lexical ambiguities and the surrealist frame of each micro-event:

Lobster-orange, shag in parvo. Peaceful/

pushful kid wants it better, wants sex not fish upfront

3 as well in touch. Spring peaks red-inked, blissful dogged

doggerel at joint screaming with rind orange blind-gut

Hill’s difficulty is such a deeply engrained critical assumption that the Norton Anthology of English Literature (2000: 2717) uses the word when prefacing him. Baker, a reviewer of Hill’s Speech! Speech! (2001) mentions ‘the spectre of Difficulty’ (Baker, 2002: 34), so my chosen text, stanza 33 (GH henceforth), comes from this collection. GH is characterized by register mixing and a critical re-employment of everyday language, whereby the heteroglossia of contemporary society embeds the poetic persona’s unstable self:

4 Something mùst give, make common cause,

In frank exchange with defamation.

So talk telegraphese, say: FORTITUDE

7 NEVER MY FORTE. BLOOD-IN-URINE SAMPLES

RUIN EURO-CULTURE.

Finally, Howe is the subject of a demanding close reading by Quartermain, who notes, among other things, how in her poems ‘semantic grounds shift’ (Quartermain, 1992: 184). The poem I have chosen (henceforth: SH) is unlike the remaining five in that its blurring of constituency boundaries poses exceptional parsing challenges:

A small swatch bluish-green

Woollen slight grain in the

3 Weft watered and figured

Right fustian should hold

Altogether warp and woof

The only prose extract (JGB) warrants a different sort of justification. It functions as a control text for narrativity, since it displays many linguistic features typical of narrative fictional texts: third person pronouns; proper nouns; past tense verbs; and adverbials of time and place (Biber and Conrad, 2009: 150). In addition, it fits well the definition of narrative provided by Toolan:

A narrative is a perceived sequence of non-randomly connected events, typically involving, as the experiencing agonist, humans or quasi-humans, or other sentient beings, from whose experience we humans can ‘learn’. (Toolan 2001 [1988]: 8)

Importantly, JGB has been disguised as a poem by adding lineation:

Soon it would be too hot. Looking out

from the hotel balcony shortly after eight o’ clock,

3 Kerans watched the sun rise behind the dense groves

of giant gymnosperms crowding over the roofs

of the abandoned department stores for a hundred yards

6 away on the east side of the lagoon.

This decision was guided by Zwaan’s (1993 [1991]) finding that genre awareness influences processing time. Since I am interested in literary language rather than in genre expectations, my purpose was to predispose participants to mistake JGB for a standalone narrative poem. Alongside the added lineation, such induced genre misclassification should be aided by Ballard’s literary style, which draws on devices typically associated with poetry: similes (like a colossal fire-ball); lexical foregrounding (the botanical name gymnosperms); alliteration (giant gymnosperms); and an elaborate use of attributive adjectives (dense, abandoned, massive, relentless…).

Although it contains unmistakably poetic features (e.g. the genitive metaphors scars of light and wounds of night), MS also fits Toolan’s definition well while displaying key linguistic features of narrativity (third person pronouns, adverbials of time and space). From the perspective of a ‘comparative textology’ (Carter, 2014 [1989]: 181), then, JGB and MS are ideally positioned next to each other as a poetic narrative and a narrative poem, respectively. By contrast, EP can be classified as semi-narrative: on the one hand, it displays several of the aforementioned features (proper names, adverbials of space, reporting verbs and direct speech); on the other, it fails to meet Toolan’s criterion of ‘non-randomly connected events’: the transitions between the speech events in EP look erratic, and to make them cohere one must venture on quite far-fetched interpretive inferences.

The remaining four poems are hardly narrative: WS only features past tense verbs and one adverbial of time (at twelve) and lacks a central ‘experiencing agonist’. What is more, its relational verbs and extended noun phrase series (ll. 7–17) suspend all sense of temporal progression. JP hardly exhibits any narrative features either: although there are a proper name and some third person subjects, their presence is backgrounded; there is no spatiotemporal anchoring and, structurally, Toolan’s definition does not apply: there are no experiencing agonists, the events appear randomly connected and the point of the narration is utterly unclear. GH is not narrative either, as it displays the stylistic features of interior monologue (first person pronouns, present tense, foregrounded deixis: see Wales, 2011: 232) whilst at the same time undermining one of the key functions of this writing technique. Indeed, we are denied full access to the poetic persona’s consciousness because some of his utterances are ironically borrowed from media discourse, thus simultaneously affecting LAD 19f (‘poetic persona concealed’) in Table 1. Finally, SH is even more uncompromising in its antinarrative stance: the poem is constructed as an epistemically modalized declarative clause followed by two semantically unrelated interrogatives, even though its phrasal and clausal boundaries are blurred and need to be inferred. In addition to this, neither narrative features nor experiencing agonist can be identified.

In summary, JGB and MS are fully narrative, EP is semi-narrative, WS, GH and JP are barely narrative, and SH seems the least narrative of all. If lack of narrativity is a strong LAD, then the average reading speed is expected to decrease alongside the narrativity of a poem. The empirical data presented and discussed in Section 3.3 appear to support this hypothesis.

Participants

Twelve 1st year undergraduates from the University of Nottingham (UK) volunteered for the test as part of the Investigating English Language module. All were native speakers of English (all female; mean age 18.2 years; standard deviation = 0.4 years). Their print exposure and reading habits have been gauged through the personal questionnaire reproduced in Figure 1.

Figure 1.

The personal questionnaire.

From Q3 on the questionnaire it emerged that none of the participants knew any of the texts used in the study, so they all were read for the first time during the test. The list in Q4 is an Author Recognition test, providing ‘a measure for “print exposure”’ (van Peer et al., 2012: 221). It features poets, other intellectuals and made-up names to check whether the participants had heard of some of the authors tested beforehand. On average, participants correctly identified four out of the twelve poets in the list. Except for Pound, the authors used in the test were little known if not completely unknown: Stevens and Strand were identified twice in total, Hill only once, Howe and Prynne never. As for leisure reading time, 16% of participants read less than 30 minutes per day and none more than three hours. The majority of responses are split between 30 minutes and one hour (50%) and from one to two hours (50%).

A comparison between poetry collections and novels (Q6 and Q7) reveals the much greater popularity of the latter genre. As many as 42% of the participants had read more than ten novels per year, against 0% for poetry; conversely, one-third of the participants (33%) had read only one to three poetry collections per year; higher still was the proportion of those who had not read any poetry collections at all (58%). In contrast, all the participants had read at least one to three novels per year. When it comes to leisure reading, therefore, even the well-read participants in our study by and large choose novels over poetry collections. These data echo the already quoted claim by Graesser and colleagues regarding readers’ deep familiarity with narrative schemata. Such familiarity is of course likely to have significant bearings on the outcome of the test.

Design and procedure

The test was carried out using E-prime software. The test type is a self-paced reading task: participants control the reading speed by pressing the space bar button on the keyboard to move from one line to the next one. In the meantime, the program records in milliseconds the time elapsed between each pressing of the button. The technical name of the test (‘window-moving accumulating test task’) implies that previously read lines remain on the screen after pressing the button. This choice suits difficult poems, which intuitively encourage readers to return to passages they found problematic in order to retrace coherence. The text presentation order was randomized to eschew order effects like fatigue or improved performance through exposure to the task (van Peer et al., 2012: 130). Each poem appeared once only, and readers could not return to it once they proceeded to the next one. This ensured that reading times were recorded for first readings only, as the focus here is on the impact of poetic difficulty rather than in its overcoming. Sessions took place on a one-to-one basis in the psycholinguistics laboratory at the University of Nottingham. Each participant was asked to sit in front of a computer screen and, when ready, to open the file of the experiment. At this point, they read the instructions on the screen, reported verbatim in Figure 2.

Figure 2.

On-screen instructions for the reading task.

Bearing in mind the influence of reading modes on reading times (Zwaan, 1993 [1991]), the penultimate point encouraged a leisure mode of reading. Although it may sound odd that an instruction should ask readers to behave naturally, the omission of such instruction would have resulted in non-comparable response data (e.g. some participants trying to analyse a poem and others skimming through it). There is no problem-free approach in empirical studies, especially when it comes to complex literary texts rather than constructed examples. The general rule is therefore to do ‘the next best thing’ (van Peer et al., 2012: 135). The task was completed in five to ten minutes. Before leaving, the participant received the personal questionnaire reproduced in Figure 1. Once this was completed too, it was handed back and the participant could leave the room.

4. Analysis and discussion

The analysis of reading times was done per item (i.e. poems) rather than per participant, as the interest was in the average response elicited by each text rather than in the variability of readerly skills. The reading speed per line was obtained by dividing the reading time in milliseconds by the number of letters of that line, counted by the software. This practice is standard in psycholinguistics for non-manipulated data (see Carminati et al., 2006; Mahlberg et al., 2014). Of course, the need for an unambiguous unit of measure (i.e. characters) prevents other aspects (e.g. punctuation, and syntactic structure) from being incorporated into the calculation of reading speed at this stage. The average reading speed for all the lines of a poem thus calculated is equal to the average reading speed per poem. Another preliminary step was that of removing outliers, that is, abnormally low or high values (i.e. exceeding the average by 2.5 standard deviations) stemming from momentary distractions and likely to skew the results. This operation resulted in the removal of 2.87% of the original reading speed data. Table 3 provides the average reading speed for each text (also visualized in Figure 3) alongside the standard deviation.

Table 3.

Average speed and standard deviation (SD) across lines per text.

Text	Average speed	SD across lines	Degree of narrativity
JGB	59.01 milliseconds (ms)/character (char)	8.77 ms/char	Narrative
EP	62.76 ms/char	11.79 ms/char	Semi-narrative
MS	63.51 ms/char	8.38 ms/char	Narrative
GH	72.45 ms/char	10.90 ms/char	Non-narrative
WS	73.39 ms/char	10.47 ms/char	Non-narrative
JP	79.34 ms/char	12.48 ms/char	Non-narrative
SH	90.41 ms/char	13.05 ms/char	Non-narrative

Figure 3.

Visualization of average reading speed per text.

To begin with, it is possible to trace a processing divide between JGB, EP and MS on the one hand, and WS, GH, JP and SH on the other. JGB, EP and MS are read at similar speeds which are, interestingly, very close to the 61.85 milliseconds (ms)/character (char) figure reported for a passage from Dickens in Mahlberg et al. (2014: 12). Three out of these four texts display a marked narrativity; by contrast one (EP), as argued before, is linguistically – though not functionally – narrative.

Not only does EP not fit Toolan’s definition of narrative; lacking inter-sentential cohesion, it cannot certainly be deemed an easy-to-read poetic extract. Indeed, EP was deemed hard to understand in an earlier survey probing post-reading comprehension (offline difficulty). So, at least in this instance, ease of processing does not necessarily run in parallel with ease of understanding. My proposal is that these two dimensions of difficulty are to an extent independent. EP has the readability of a traditional narrative, and its low online difficulty is enhanced by intra-sentential cohesiveness and a scantiness of local LADs. What seems to matter for the activation of a narrative schema is, quite counterintuitively, the extent to which the style of a literary text resembles a narrative locally (i.e. at clause level), rather than the overall fulfilment of the structure of a prototypical narrative.

On a first reading, then, I suggest that readers focus on the syntagmatic order of linguistic constituents but do not apply top-down their genre knowledge. The familiarity with narrative fiction attested to by participants (see 3.2.2) may have made their narrative schema more strongly activated than their poetry schema. This helps explain the similarity in reading speeds for texts that are so heterogeneous stylistically, as my survey has shown. 60 ms/char seems to mark the upper processing limit for texts displaying narrative features and little or none of the most disruptive among the LADs (e.g. violation of selection restrictions, asyntactic string − 9c and 10b in Table 1). My initial hypothesis that narrativity eases online processing by speeding reading up is therefore empirically supported. At the same time, the hypothesis has been made more precise by defining narrativity in linguistic rather than in functional terms.

GH, WS, and JP were read at a remarkably lower speed (from 70 ms to nearly 80 ms/char). As argued earlier, these poems are barely narrative, which implies that lack of narrativity is a powerful LAD in terms of online difficulty. In addition, they feature several local LADs (see Castiglione (2015) for an extensive analysis of these poems). This means that, in contrast to JGB, EP and MS, they pose processing hurdles at decoding (e.g. low-frequency, technical words or even neologisms) and integration level (e.g. novel metaphors, intra-clausal lack of coherence).

SH elicits by far the slowest reading speed: 90.41 ms/char, around 30% more than JGB (59.01 ms/char). The narrativity of SH is arguably null, since it does not stage an experiencing agonist (only an implicit speaker) and tells no events. In addition, its garden path problems account for such a slow reading even more compellingly. Syntactically, the poem opens with a string of nouns and adjectives that refuses to be neatly parsed:

A small swatch bluish-green

Woollen slight grain in the

3 Weft watered and figured

Right fustian

In the first line, it is possible to identify a noun phrase in which the colour adjective bluish-green simultaneously fills a predicative syntactic role (i.e. it refers to swatch, assuming the copula ‘be’ is omitted) and an attributive one (i.e. it premodifies grain in the second line). The fact that bluish-green can be linked to two Head nouns creates a garden path effect. This structural ambiguity is replicated in l. 3. Here the participial adjectives watered and figured can simultaneously refer to weft, in which case they fill a predicative role, and to fustian, in which case they fill an attributive role. In addition, the absence of verbs means that three noun phrases (the Heads of which are swatch, grain and fustian) and one prepositional phrase (the Head of which is weft) are mutually embedded, with an ensuing feeling of incrementally increasing complexity. As a result of this stylistic configuration, readers may have been forced back to previous lines once misled by the garden paths. In turn, this explains the extremely slow reading speed elicited by SH. Incidentally, the fact that SH takes longer to read than the no less deviant JP suggests that parsing problems caused by blurred constituency are more central to online difficulty than integration problems caused by violations of selection restriction.

5. Conclusions

The central idea behind this research has been to work backward from effect to structure, from reading speed to linguistic properties. Reading time data exhibited unexpected processing similarities between texts JGB, EP and MS, whose only commonality is the presence of local stylistic features associated with narrative text-types. Theoretically, this finding supports the inclusion of the ‘lack of narrativity’ LAD in my stylistic–empirical model of poetic difficulty. Of course, tests like this dealing with original materials leave the door open for interpretive leeway, and the explanations that can be offered have to be tentative.

Despite this encouraging outcome, design limitations are worth stressing too. First, lines were presented without the left blank space which graphically marks them off as lines. This choice gives primacy to their linguistic properties, but downplays the effect of enjambments documented by Jagt et al. (2014). Another drawback is that the E-Prime does not incorporate the graphic accents used in GH nor does it allow representing the long blank space in the last line of JP, so this aspect of graphological foregrounding could not be captured. Modifications to the software are needed if difficult poetry is to be tested more rigorously in the future.

These limitations derive from the lack of a directly comparable study and reveal the challenges associated with testing as elusive and complex a notion as poetic difficulty. In the future, research could compare reading times for the first and second reading of poems varying in the kind, amount and distribution of LADs, in order to gain a fuller picture of how such poems are processed over time. Furthermore, it will be useful to replicate the study with a sample of readers with a strong poetic schema (e.g. poetry lovers, practising poets, and critics) to see whether the narrative hypothesis can be generalized across different reading populations. It has already been demonstrated (Peskin, 1998) that expert readers of poetry have quicker and more systematic access to certain interpretive strategies (e.g. allusions to other literary works, and contextualization) compared to novice readers, and tend to treat poems not as ‘a finite problem but an open-ended task’ (Peskin, 1998: 243). However, since online difficulty affects early stages of processing, it may be expected that reading differences due to expertise will be minimal. Finally, another promising path for future research involves the use of eye-tracking equipment (see Staub and Rayner, 2007), which by recording eye-movements in real time is able to pinpoint the linguistic sources that impair reading fluency. The researcher would also have the opportunity of visualizing and comparing the preferred reading routes for poems varying in their narrativity or along other relevant textual dimensions.

Footnotes

Acknowledgements

This paper was originally a part of my PhD thesis, and therefore it has benefited from the expert advice and encouragement of my supervisor Violeta Sotirova and my examiners Peter Stockwell and Andrew Caink. Many thanks also to Patricia Canning and to Sara Whiteley for their support and constructive feedback on the first draft of the paper, and to the anonymous reviewer for referring me to additional relevant sources and for helping me clarify some terminological issues. I am, of course, responsible for whatever weakness the paper may still contain.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author biography

Davide Castiglione is currently a lecturer in Language and Literature at Vilnius University. His specialism is the notion of difficulty in poetry, which he addresses by combining the stylistics of poetry and empirical approaches to literature. As a creative writer, he is also the author of two poetry collections.

References

Adams

(1991) The difficulty of difficulty. In Purves

(ed.) The Idea of Difficulty in Literature. New York, NY: State University of New York, pp.23–50.

Adamson

(1999) The literary language. In: Romaine

(ed.) The Cambridge History of the English Language, 4, 1776–The Present Day. Cambridge, UK: Cambridge University Press, pp.589–692.

Baker

(2002) Review of “Speech! Speech!” by Geoffrey Hill. Poetry Nottingham International 56(3): 34.

Ballard

(1962) The Drowned World. London, UK: Berkley Books.

Biber

Conrad

(2009) Register, Genre, and Style. Cambridge, UK; New York, NY: Cambridge University Press.

Carminati

Stabler

Roberts

. (2006) Readers’ responses to sub-genre and rhyme scheme in poetry. Poetics 34(3): 204–218.

Carter

(2014 [1989]) What is stylistics and why can we teach it in different ways? In: Simpson

(ed.) Stylistics: A Resource Book for Students. London, UK: Routledge, pp.175–182.

Castiglione

(2013) The semantics of difficult poems: Deriving a checklist of linguistic phenomena. Journal of Literary Semantics 42(1): 115–140.

Castiglione

(2015) Difficulty in Anglo-American Poetry: a stylistic and empirical perspective. PhD Thesis. The University of Nottingham, UK.

10.

Chafe

(1991) Sources of difficulty in the processing of written language. In: Purves

(ed.) The Idea of Difficulty in Literature. New York, NY: State University of New York, pp.7–22.

11.

Chatman

(1972) The Later Style of Henry James. Oxford, UK: Basil Blackwell.

12.

Diepeveen

(2003) The Difficulties of Modernism. New York, NY; London, UK: Routledge.

13.

Dillon

(1978) Language Processing and the Reading of Literature. Bloomington, IN; London, UK: Indiana University Press.

14.

Empson

(1930) Seven Types of Ambiguity. London, UK: Chatto and Windus.

15.

Fish

(1980) Is there a Text in this Class? The Authority of Interpretive Communities. Cambridge, MA; London, UK: Harvard University Press.

16.

Graesser

Hoffman

Clark

(1980) Structural components of reading times. Journal of Verbal Learning and Verbal Behavior 19(2): 135–151.

17.

Grice

(1975) Logic and conversation. In: Cole

Morgan

(eds) Syntax and Semantics 3: Speech Acts. New York, NY: Academic Press, pp.41–58.

18.

Hakemulder

van Peer

(2016) Empirical stylistics. In: Sotirova

(ed.) The Bloomsbury Companion to Stylistics. London, UK; New York, NY: Bloomsbury, pp.189–207.

19.

Halliday

MAK

(2014) An Introduction to Functional Grammar. 4th edition, revised by Matthiessen CMIM. London, UK: Routledge.

20.

Hanauer

(1998) The genre-specific hypothesis of reading: Reading poetry and encyclopedic items. Poetics 26(2): 63–80.

21.

Harley

(2008) The Psychology of Language. 3rd edition. London, UK: Psychology Press.

22.

Hill

(2001) Speech! Speech! London, UK: Penguin Books.

23.

Hynds

(1991) Questions of difficulty in literary reading. In: Purves

(ed.) The Idea of Difficulty in Literature. New York, NY: State University of New York, pp.117–140.

24.

Howe

(2001) Bed Hangings. With pictures by Susan Bee. New York, NY: Granary Press.

25.

Jagt

Hoeks

JCJ

Dorleijn

. (2014) Look before you leap: How enjambment affects the processing of poetry. Scientific Study of Literature 4(1): 3–24.

26.

Jeffries

(2008) The role of style in reader-involvement: Deictic shifting in contemporary poems. Journal of Literary Semantics 37(1): 69–85.

27.

Kintsch

(1998) Comprehension: A Paradigm for Cognition. Cambridge, UK: Cambridge University Press.

28.

Lamarque

(2009) The elusiveness of poetic meaning. Ratio (New Series) 27(4): 398–420.

29.

Leech

(2010) Language in Literature: Style and Foregrounding. Abingdon, UK: Pearson Longman.

30.

Leggett

(2007) Stevens’ late poetry. In: Serio

(ed.) The Cambridge Companion to Wallace Stevens. Cambridge, UK: Cambridge University Press, pp.62–75.

31.

Lopez

(2006) Meaning Performance: Essays on Poetry. Cambridge, UK: Salt.

32.

McHale

(2004) The Obligation Toward the Difficult Whole: Postmodernist Long Poems. Tuscaloosa, AL: University of Alabama Press.

33.

Mahlberg

Conklin

Bisson

(2014) Reading Dickens’s characters: Employing psycholinguistic methods to investigate the psychological reality of patterns in text. Language and Literature 23(4): 369–388.

34.

Mellors

(2005) Late Modernist Poetics: from Pound to Prynne. Manchester, UK: Manchester University Press.

35.

Miall

Kuiken

(1994) Foregrounding, defamiliarization, and affect: Response to literary stories. Poetics 22(5): 389–407.

36.

Milroy

(1977) The Language of Gerald Manley Hopkins. London, UK: Deutsch.

37.

Moen

(2010) ‘ Chi È Questa? – Who is she?’: Transformation, displacement, and narrative refraction as structural procedures in The Cantos of Ezra Pound. Textual Practice 24(2): 287–312.

38.

Nadel

(2007) The Cambridge Introduction to Ezra Pound. Cambridge, UK: Cambridge University Press.

39.

Nowottny

(1962) The Language Poets Use. London, UK: Athlone Press.

40.

Perloff

(1985) The Dance of the Intellect: Studies in the Poetry of the Pound Tradition. Cambridge, UK: Cambridge University Press.

41.

Peskin

(1998) Constructing meaning when reading poetry: An expert-novice study. Cognition and Instruction 16(3): 235–263.

42.

Pound

(1954) The Cantos of Ezra Pound. London, UK: Faber and Faber.

43.

Press

(1963) The Chequer’d Shade: Reflections on Obscurity in Poetry. London, UK: Oxford University Press.

44.

Purves

(ed.) (1991) The Idea of Difficulty in Literature. New York, NY: State University of New York.

45.

Prynne

(2005) Poems. Tarset, UK: Bloodaxe Books.

46.

Quartermain

(1992) Disjunctive Poetics: From Gertrude Stein and Louis Zukofsky to Susan Howe. Cambridge, UK: Cambridge University Press.

47.

Richards

(1929) Practical Criticism. London, UK: Kegan Paul, Trench, Trubner.

48.

Sanford

Emmott

(2012) Mind, Brain and Narrative. Cambridge, UK: Cambridge University Press.

49.

Sell

(1993) The difficult style of “The Waste Land”: A literary–pragmatic perspective on Modernist poetry. In: Verdonk

(ed.) Stylistic Criticism of Twentieth-Century Poetry: From Text to Context. Florence, KY: Routledge, pp.135–158.

50.

Serio

(2007) Introduction. In Serio

(ed.) The Cambridge Companion to Wallace Stevens. Cambridge, UK: Cambridge University Press, pp.1–7.

51.

Shklovsky

(1998 [1917]) Art as technique. In: Rivkin

Ryan

(eds) Literary Theory: An Anthology. Oxford, UK: Blackwell Publishing, pp.15–21.

52.

Staub

Rayner

(2007) Eye-movement and online comprehension processes. In: Gaskell

(ed.) The Oxford Handbook of Psycholinguistics. Oxford, UK: Oxford University Press, pp.327–42.

53.

Steiner

(1978) On difficulty. In: Steiner G On Difficulty and Other Essays. Oxford, UK: Oxford University Press, pp.18–47.

54.

Stevens

(1950) The Auroras of Autumn. New York, NY: Knopf.

55.

Strand

(1978) The Late Hour. New York, NY: Atheneum.

56.

Sutherland

(2010) Wrong poetry. Textual Practice 24(4): 765–782.

57.

Toolan

(1993) Approaching Hill’s “Of Commerce and Society” through lexis. In: Verdonk

(ed.) Stylistic Criticism of Twentieth-Century Poetry: From Text to Context. Florence, KY, USA: Routledge, pp.32–45.

58.

Toolan

(2001 [1988]) Narrative: A Critical Linguistic Introduction. London, UK: Routledge.

59.

Tuma

(1998) Fishing by Obstinate Isles: Modern and Postmodern British Poetry and American Readers. Evanston, IL: Northwestern University Press.

60.

van Dijk

Kintsch

(1983) Strategies of Discourse Comprehension. New York, NY; London, UK: Academic Press.

61.

van Peer

Hakemulder

Zyngier

(2012) Scientific Methods for the Humanities. Amsterdam, The Netherlands: John Benjamins.

62.

Wales

(2011) A Dictionary of Stylistics. 3rd edition. Harlow, UK: Longman.

63.

Wallot

O’Brien

Haussmann

. (2014) The role of reading time complexity and reading speed in text comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition 40(6): 1745–1765.

64.

White

(1981) The Uses of Obscurity: The Fiction of Early Modernism. London, UK: Routledge and Kegan Paul.

65.

Yaron

(2002) Processing of obscure poetic texts: Mechanisms of selection. Journal of Literary Semantics 31(2): 133–170.

66.

Yaron

(2003) Mechanisms of combination in the processing of obscure poems. Journal of Literary Semantics 32(2): 151–166.

67.

Yaron

(2008) What is a “difficult” poem? Towards a definition. Journal of Literary Semantics 37(2): 129–150.

68.

Zwaan

(1993 [1991]) Aspects of Literary Comprehension. Amsterdam, The Netherlands: John Benjamins.