Roman Ingarden [Ing73] developed Aristotle's concept that a literary work of art has at least 4 layers
Furthermore he thought the work composed of an infinite number of Unbestimmtheiststellen (indetermination spots). Individual concretizations (readings) only partly remove these places of indeterminacy, so that different concretizations remained possible. Ingarden's work was phenomenologically based, with few poetry examples. He assumed [Ing73, p.40] that "the literary work is conceived as a purely intentional product of the subjective, creative processes of its author." Since then other literary approaches have found favour and experimental psychologists have studied the reading process more deeply. Although they seldom use poetry texts, much of their work concentrates on elements that are common in poetry - semantic/syntactic ambiguity - and are especially relevant to how layers 3 and 4 affect the transition from layer 1 to layer 2. With this new evidence we might now be better placed to determine how a poem's meaning is resolved, what the resolution and scale of the derived meaning is, and how the layers interact.
Not all texts pose readers problems. The "more 'purely lyric' the poem is, the less - roughly speaking - is the effective determination of whatever is positively stated in the text" [Ing73, p.52]. When ambiguities occur they can be at the word level (homographs, like "wait for me at the bank"), the phrase (oxymorons can make a phrase irreduceable), anaphoric (where "it" might refer to more than one thing), etc. Whole sentences can be spiked with (perhaps retrospective) irony. Whenever the readers are presented with such choices, expectation (statistically, contextually or syntactically primed) guides them to a solution. Interpretations are preferred that match their knowledge of the world, or refer to something earlier in the text. But even these criteria often aren't enough to discount all the alternatives. Theories describing readers strategies fall into the following categories
The strategy chosen depends on the type of text (the chance of ambiguity) and also on the flexibility of the reader. Hybrid models have been developed suggesting that parallel processing is used until the load becomes too great, after which serial processing is used. See D. Michell's article [Ger94, p.375] or [Cla77, p. 82] for further details.
Once the little local troubles have been resolved there's a tendency to seek a single meaning from these multiple atoms, penetrating from layer 3 to layer 4 of Ingarden's schema. The presented objects are constructed from non-contiguous material, so that there's scope for the uncovering of new contradictions at this stage. The process is rarely cumulative. The whole can have properties denied the parts - a surreal poem can be made of non-surreal sentences; a story full of jokes can be sad. The emergent tone of the whole can in turn affect the interpretation of the parts. Much current psychological research investigates to what extent higher level semantic representations constrain the activity of lower processes. In literature the overall tone doesn't always dominate: a final line can reveal the irony of the whole; one disruptive pun can wreck a serious piece. I'll draw on some analogies from mathematics to discuss various approaches to integration.
Time and sequence figure in Ingarden's designs. He thought that works were both multi-layered and multiphase, having an ordered sequence of parts; a "peculiar quasi-temporal extension". He wrote that "The individual objects are not only enriched by new attributes, ... They are also constantly entering into new connections and relations with other objectivities of the portrayed world and are freed from those relationships which had previously been binding" [Ing73, p46]. This refers to changes in the population and interaction in the 4th layer. He seems not only to have a bottom-up approach to textual comprehension, but he also undervalues the problem of lower level processing. Disruption can happen at lower levels, disrupting a reader's linear progression through the text. Even when reading normal texts, about 10% of saccades (eye motions) are backwards [Ger94, p.58]. When bigger structures than words are involved, and readers have invested more time in parsing them, then reversals can be upsetting. Two sentenses with similar surface structure can have very different parse trees
Have the policemen whom you saw arrived?
Have the policemen whom you saw dismissed. [Mar80]or
Computers can't wreck a nice beach
Computers can't recognise speech
Realisation of this at level 2 requires structural regeneration and consequent level 1 backtracking. Mistaken bridging inferences (the reader mistaking the writer's implicit assumptions) can be more disruptive still. In general, the higher the layer in which the incongruity is discovered, the larger the backtracking required - in a whodunnit a new piece of evidence may cause us to re-evaluate a scene, tempting us to flip back through the pages.
When reading poetry (moreso than when reading prose), we are prepared to delay interpretation or make re-interpretations. Our eyes move more hesitantly too, making longer fixations, shorter jumps and more backward jumps. We may maintain a number of interpretations in mind simultaneously until a checkpoint (maybe a question or a linebreak) triggers a resolution. An analogy can be made here with quantum physics. Mathematically, the position of a particle is determined by a probability function - one cannot be certain of its position until a measurement takes place. It's a matter of interpretation whether the probability function describes the likelihood of a particle being in a particular place (say, 5% likely to be in a box), or whether the wave/particle is partly there (5% of it is in the box). When a measurement is made, the probability function is said to collapse, and we detect a particle. In a similar way, we can try to force meaning out of a text, but this process gives only one of what could have been many possible outcomes. Only after many reading/experiments can one expect statistcal significance.
An over-hasty collapse of the probabilities into a single meaning may be engineered by the writer. The lines
his work is good for nothingwould lose their suprise value were they a single line. I suspect that readers of modern literature are used to resisting premature resolution
Some pieces are not constructed hierarchically, or even linearly. Level 2 (meaning) is too crumbly to support any higher stuctures. The pyramid is flattened; wordplay and character development compete on a level playing field. Resolution, if it is felt necessary, is deferred until the end without the build up of pre-Modernist tension. "The meaning-relationship is completed only by the simultaneous perception in space of word-groups that have no comprehensible relation to each other when read consecutively in time...modern poetry asks its readers to suspend the process of individual reference temporarily until the entire pattern of internal references can be apprehended as a unity" [Fra45]. Expectation-based disambiguation is unavailable. Local sub-elements can't be solved first. The text is a single quantum system whose components, like the notes of a melody, cannot be usefully studied in isolation.
Though Ingarden's theory works well enough for technical documents, it doesn't well model situations like those above where interaction between layers is two-way or where (in pieces by ee cummings, for example) layers effectively overlap. Even at the level of letter identification, the layer model become clumsy. When readers see a round capital letter they look down to the bottom right corner because the higher processing layers know that the letter could be an O or a Q. This confusion of perceptual layers is a sign that the work lacks intrinsic layers. In descriptions of Ingarden's work there is some confusion over the role of the upper layers, partly because as the theory developed, definitions became more fuzzy. Ingarden viewed cognition as "composed of heterogeneous but closely connected processes" [Mcc85, p.11]. Psychological research has shown that these processes perform effectively only when they can accept feedback. He also considered the outside world (including texts, presumably) to be irrelevant to the work's structure [Mcc85, p.12], which is more true of lyric poetry than postmodern pieces. We are left with the impression that Ingarden's theories work best on a small subset of texts beyond which many 20th century works lie.
[Fra45] "Spatial Form in Modern Literature" J. Frank, Sewanee Review, 1945
[Ger94] "Handbook of Psycholinguistics", M.A. Gernsbacher (ed), Academic Press, 1994
[Ing73] "The Cognition of the Literary Work of Art", Roman Ingarden (trans R.A. Crowley and K.R. Olsen), Illinois, 1973
[Mar80] "A theory of syntactic recognition", Marcus, MIT Press, 1980.
[Mcc85] "Selected Papers in Aesthetics/Roman Ingarden", P.J. McCormick, The Catholic University of America Press, 1985.