Words weave through sentences, submerge from view, resurface, resubmerge, warp and weft of weave or whatever metaphorical model we can throw at it. A word indexes many situations; a situation indexes many words. Can we see words and narrative, i.e. semantics and episodics, as orthogonal to one another, at right angles? Say, “story” is the X axis, “wording” the Y. Cognitive psychologists for a long time have entertained the distinction of semantic from episodic memory, and sought to discern whether they are, as hypothesized, truly independent from one another. Steven Prince, at al. (2007) report in Psychological Science that “the neural correlates of EE [episodic encoding] and SR [semantic retrieval] are dissociable but interact in specific brain regions.” In an example of this kind of interaction, semantic associations enhance the retrieval of episodic memories (Menon, et al., 2002). One site of interaction between the semantic and the episodic happens at the very instant a word is semantically integrated into a sentence, the very instance, in effect, when it enters into the telling of an episode.

While Swinney’s and others’ interest is pitched at an understanding of how we “compose” an interpretation out of a sentence’s building blocks, from the words and the grammatical relations within it, it does not seem to me to be too great a stretch to map this fine-scale event to the wider action of narrative within which it happens. Those ephemeral moments interest me when irrelevant meanings are sloughed away and only meanings fit to the flow of narrative within which they are embedded are left. This is the moment I have called “reading between the words.”

For every word we read, our mind opens and closes on that word’s potentialities, like the gate of a movie projector opening and closing the light on frame after frame of a film that slips a pull-down of filmstrip though, click-click-click in those interstices between openings, a sentence operating on our minds like a filmstrip operating on our eyes and visual cortices. By persistence of sense just as to the eyes, the film works by persistence of vision. Open, closed, open, closed—but it all comes coherent in a single stream, illusory yes, but to the mind, real. In fluent reading, whole gobs of text go in before consciousness opens its aperture. Flick-flick-flick. But it seems continuous.

William Nestrick, citing Walter Benjamin’s essay “The Work of Art in the Age of Mechanical Reproduction,” states that “the film is the animation of the machine, a continuous life created by the persistence of vision in combination with a machine casting light through individual photographs flashed separately upon the screen.” Narrative, an older technology, is an animation of another machine. Our deeply naturalized habits of reading elide the composite, discontinuous nature of our construction of meaning. Like the animation of the machine, it invites comparison to the monster from Frankenstein—and hypertextual disruption, what Shelley Jackson encodes in Patchwork Girl.

In the prevalent “invisible style,” films build seamless scenes from successions of discrete shots much as the mind builds seamless sentence understanding out of successions of words. What is left out is simply not noticed. Maybe it is repressed or abjected—but to waking, normal cognition, something must be, because the entirety of the worlds we build in our minds arises from selectively attending to some things and not others. Repression in one sense is the inescapable flip side of being in the world with a mind. It is the flip side of the selective attention that gets knitted together by wetware magic into scenes and stories we assemble, recall, reform, tell, and retell. No word drawn from the batch of “non sequitur,” “relevant,” “irrelevant,” “emphasis,” etc. could bear on our way of writing and transacting communication if it were not in a context of narratives, stories, and episodes, with their characteristic features: topics, action, motivation…. And none of these concepts would have any utility to us if construction of stories in our minds were not inherently selective, if thought itself were not inherently so.

The process of selection and of assembly follows a pattern something like what Katherine Hayles describes in “Flickering Connectivities in Shelley Jackson’s Patchwork Girl” (2000). More than the signifier, though, it is signification that flickers, the association of signifier with signified, on a pattern of flickering consciousness, in which access to the results of the automatic, unconscious process of lexical access flickers in and out, in fluent reading seldom noticed. This kind of processing, highly automatized by skill, can be disrupted by divergence from conventional expectation. Since 1917,  when Russian “Formalist” Viktor Shklovsky described it, this disruption of the automatic has been prominent in aesthetics, theorized as “strange-making” (ostranenie to transliterate the Russian), or “defamiliarization.” The practice has its precedent in the work of William Wordsworth. Explained by Samuel Taylor Coleridge in his Biographia Literaria:

“Mr. Wordsworth, on the other hand, was to propose to himself as his object, to give the charm of novelty to things of every day, and to excite a feeling analogous to the supernatural, by awakening the mind’s attention to the lethargy of custom…”

In the frame of the present post, comparing the cognitive integration of text with the sensory integration of motion pictures, this would be akin to slowing the film down below the critical flicker frequency, so as to break the illusion. In the technoromantic frame, the mechanism of novelty’s action in the brain is the “orienting reflex,” first described by Ivan Petrovich Pavlov as the “shto eto takoi” (“what is it?”) reflex. The orienting reflex is the mechanism that breaks the automatic.

Hayles:

“In Patchwork Girl, one of the important metaphoric connections expressing this flickering connectivity is the play between sewing and writing. Within the narrative fiction of Frankenstein, the monster’s body is created when Frankenstein patches the body parts together; at the metafictional level, Mary Shelley creates this patching through her writing.”

Shelley Jackson’s work strategically disrupts narrative at several levels, from fine-grained lexical interactions to collisions between textual elements arising in the restructuring of narrative (as hypertext) and exploitation of a medium (the computer) that restructures the reading interface. Significant to this is Donna Haraway’s “Cyborg Manifesto,” which treats the reconfiguration of the human form itself. Hypertext operates, for now, on the evolved biological capacity for cognitive reconfiguration, hacking the machine via its present interface.

What lies at the asymptote of narrative restructuring by hypertext? Where does it lead? Again sketching: At the outer extreme of hypertext is the “chaotic novel” described in “The Garden of Forking Paths” by Jorge Luis Borges. Any (coherent) narrative we read picks out senses of the words within it, biases us to read words one way and not another. So, if a narrative like the Garden of Forking Paths were to be realized—in which we visit every possible line of a story that unfolds within it—must we then also visit every possible sense of every word within it? Not just as we encounter them, in the way Swinney shows us is so often unconscious, but as we integrate them into their contexts? What happens if word senses themselves mutate with context, with the events that happen around them in the stories they enter in? What does representation of this level of complexity represent?

——————————————————————————–

Caveat:

I have assumed in this series of blog posts that what is happening at the lexical level in sentences is a microscopic instance of a larger sweep of events in our cognitive construction of narrative meaning—that the sentence in some sense works like the scene, and so on up. This theoretical portrayal is adumbrated, sketchy: conjectural. Reflexively, though not necessarily destructively, it is subject to the same form of critique, of illusory continuity. Delve into the psycholinguistic findings, and extend them, and expect the results to richen or complicate the present picture. A side-effect of reliance on Swinney’s work may be acceptance of a modular model of word representation in the brain, something others have inferred from it, with fully independent representations of word meaning. This may contradict “interactionist” models of sentence interpretation, which in my view are essential to understanding how we handle figurative language. For one form of experimental challenge to Swinney et al., see “Early Integration of Context During Lexical Access of Homonym Meanings,” by Janet Lee Jones in Current Psychology 10, no. 3 (Fall 1991): 163-181. For a theoretical consideration of lexical interaction in sentences, see “Should Natural-Language Definitions be Insulated from, or Interactive with, One Another in Sentence Composition?” by L. Jonathan Cohen, in Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition 72, no. 2/3 (Dec. 1993): 177-197.

—————————————————————————————

Works Cited:

Borges, Jorge Luis. 1964. “The Garden of Forking Paths.” In Labyrinths: Selected Stories & Other Writings, 19-29. New York: New Directions Publishing Corporation.

Coleridge, Samuel Taylor. 1983. Biographia Literaria. Ed. James Engell and W. Jackson Bate. Princeton, New Jersey: Princeton University Press.

Haraway, Donna. 1991. “A Cyborg Manifesto: Science, Technology, and Socialist-Feminism in the Late Twentieth Century.” In Simians, Cyborgs and Women: The Reinvention of Nature, 149-181. New York: Routledge. (Available online at: http://www.stanford.edu/dept/HPS/Haraway/CyborgManifesto.html)

Hayles, N. Katherine. 2000. “Flickering Connectivities in Shelley Jackson’s Patchwork Girl: The Importance of Media-Specific Analysis.” Postmodern Culture 10, No. 2.

Jackson, Shelley. n.d. “Stitch Bitch: The Patchwork Girl.” MIT Communications Forum. http://web.mit.edu/comm-forum/papers/jackson.html

Jackson, Shelley. 1995. Patchwork girl, or, A modern monster by Mary/Shelley, & herself: a graveyard, a journal, a quilt, a story & broken accents. Watertown, Massachusetts: Eastgate Systems, Inc.

Jones, Janet Lee. 1991. “Early integration of context during lexical access of homonym meanings.” Current Psychology, 10 (3).

Menon, Vinod, et al. 2002. “Relating semantic and episodic memory systems.” Cognitive Brain Research, 13:261–265.

Onifer, William, and Swinney, David A. 1981. “Accessing lexical ambiguities during sentence comprehension: Effects of frequency of meaning and contextual bias.” Memory & Cognition, 9(3): 225-236.

Prince, Steven E. et al. 2007. “Distinguishing the Neural Correlates of Episodic Memory Encoding and Semantic Memory Retrieval.”  Psychological Science, 18 (2): 144-151.

Swinney, David A. 1979. “Lexical Access during Sentence Comprehension: (Re)Consideration of Context Effects.” Journal of Verbal Learning and Verbal Behavior, 18:645-659.

“Thinking is conducted by entities we don’t know, wouldn’t recognize on the street.” –Shelley Jackson, “Stitch Bitch: the Patchwork Girl”

In “Stitch Bitch,” an essay about her hypertext novel Patchwork Girl, Shelley Jackson treats us to a succession of metaphors for embodiment and the way we read: The body is a “statue,” a “hard kernel.” Hypertext is “the banished body”; it sets up “rendezvous between words never before seen in company”; when it diverges at a choice-point it dissolves as “a Cheshire aftercat.”

Hypertext brings into action properties that lie dormant in conventional linear narrative. Jackson enumerates these properties in the section entitled “Collage.”

“We don’t say what we mean to say. The sentence is not one, but a cluster of contrary tendencies…. But nobody can domesticate the sentence completely. Some questionable material always clings to its members. Diligent readers can glean filth from a squeaky-clean one. Sentences always say more than they mean, so writers always write more than they know, even the laziest of them.”

Hypertext through its strategy of design activates those dormant meanings:

“It was not difficult, for example, to pry quotes from their sources, and mate them with other quotes in the ’quilt‘ section of Patchwork Girl, where they take on a meaning that is not native to the originals.”

To understand what Jackson hypothesizes to be happening in hypertext, it helps to understand the workings of the narrative style it would subvert. Conventional linear narrative, which Jackson describes as “fated slalom,” is configured so as to shepherd readers away from divergent threads: “Plot chaperones understanding, cuts off errant interpretations.” How narratives repress the penumbral interpretations that could emerge from them, how they tame the many voices of inherent allusion and come to sound like one voice or one story, how they filter their rich harmonies down to singular melodies, is far from entirely understood. At the sentence level, this resolves to a question of how a sentence’s prevalent meaning is composed out of words that are in themselves inherently polyvalent and ambiguous. Arriving at sentence’s end, we usually have an unequivocal idea of what we have just read. But how does the brain make meaning, word by word, as it reads a sentence? When—at what moment in the reading—have we dispensed with alternative interpretations, including the senses of words that don’t fit?

In a series of experiments beginning in the late 1970’s, psycholinguist David A. Swinney developed an innovative way to pinpoint the moment of  word disambiguation in sentences. Swinney asked: As we are reading, do we access all of a word’s senses at once, and only then disambiguate them to fit the prior context, the flow of the sentence they occur in? Or, does the sentence we are reading, and our understanding of it, preselect what senses of a multivalent word’s meaning we perceive, so that we never even entertain irrelevant ones? At stake was an understanding of how verbal memory is organized, and whether word senses are accessed independently from sentence interpretation. Do we really have access to the “contrary tendencies” made possible by the breadth of allusion, and semantic potential, that its constituent words carry within them? We disambiguate words at some point; but when?

Swinney’s questions concern what psycholinguists call “lexical access.” To answer them, his group developed the ingenious “cross-modal priming task,” where one mode was auditory, the other visual, where listening to a soundtrack and reading from a screen could together be used to tease out the timing of disambiguation for words within sentences. In one experiment (1981), the auditory track plays a set of strategically constructed sentences, each with a carefully chosen word—let’s call it a target word–placed somewhere within it. That target word is in itself ambiguous, but the sentence is designed to support just one interpretation of it.   At the same time, via the visual track, other carefully chosen words are flashed on a screen.  The flashed words are, in fact, prompts that are semantically related to one or another meaning of the ambiguous target word from the auditory track. And all this takes place while the experimental subject engages in a psycholinguistic “lexical decision task” (LDT).

The experiment uses a series of sentences and target words, but as an example, let’s say the ambiguous target word is “scale,” and that two among its available meanings are: “a weighing device” and  “a protective plating on a fish or reptile.” We can’t be sure which meaning is relevant until we hear it used in a sentence: “The postal clerk put the package on a scale to see if it had enough postage” vs. “The dinner guests enjoyed the specially prepared river bass, although one did get a scale caught in his throat.” You (the experimental subject) are asked to listen to one of those sentences read aloud as a soundtrack over headphones, but at the same time you have a task to perform: Watch a monitor screen and when a string of letters appears, just as quickly as you can, press a button telling whether it is a word or not . This is the “lexical decision task.” Ideally, when you see “Glmople” or “~!@#$%^&” you press the “no, not a word” button and when you see “breakfast” you press  “yes, that is a word.”

For a long time psycholinguists have known that performance on the LDT can be facilitated, speeded up, if you have read semantically related words just beforehand. Swinney’s addition of the auditory track to the visually presented LDT enabled his uniquely time-sensitive measure of lexical access. Thus, using the ambiguous target word (e.g. “scale”) Swinney could flash related words on the screen and measure the time it took for people listening to that audio track to press the button for “yes” or “no,” “word” or “not a word.” He could probe with those visually flashed words to see which senses of “scale” were active in the listeners’ minds. All he had to do was to choose probe words to flash on the screen that were related more to one sense than the other, at any given moment. A probe word related to “scale” as “a device to measure weight” could simply be “weight,” then; and a word related to “a protective plating on a fish or reptile” could simply be “fish.” If there is priming—if hearing “scale” helps us to answer the LDT more quickly—then we can measure it.

Swinney found—contrary to earlier experiments by other researchers who had probed for lexical access only after the sentence was over—that at the moment the ambiguous word (“scale” again) was played on the soundtrack, multiple senses of that word were accessed.

That’s at the moment we read a word. How long, though, do all senses of a word remain available to us, after we read it? Swinney knew from others’ experiments that alternative meanings were unavailable by the time a sentence had been read. How long during sentence reading were alternative word senses available? Swinney’s second experiment was designed to provide an answer. Probing with the same sets of visually flashed words a second-and-a-half later, a second-and-a-half after the onset of the ambiguous word in the soundtrack, Swinney found the priming effect was gone. Only the probe words related to relevant senses of “scale” gave rise to a faster LDT while irrelevant senses that had been primed in the previous experiment were gone–knocked right out of the sentence interpretation. As Onifer and Swinney (1981) conclude, “In the absence of any strongly biasing context, it appears as though all meanings of a lexical ambiguity are accessed, at least momentarily. Such access is not available to conscious introspection, and the listener eventually becomes aware of only one of the meanings accessed for the ambiguity.”

Swinney has illuminated the cognitive processing that goes into reading within a small unit of narrative, the sentence. On this fine scale, his group’s results bear out Shelley Jackson’s descriptions of a conventionally linear “slalom” narrative, a narrative that does not invite attention to its inherent ambiguities. Each word arises in turn with its full multiplicity of meaning, only to be delivered to consciousness in a tightly narrowed sense that fits the sentence’s unitary whole. There’s a verbal sleight of hand in an unequivocal narrative that sluices the course of consciousness: It’s a magic trick. To disperse attention is to disrupt what happens between the words, before reading resolves into an unequivocal interpretation.

None of this is to say that we never access multiplicities of meaning within the “slalom” linear narrative, or that we cannot. It is simply a typical case, and subject to habits of reading as much as it is to the formulae of writing. What Swinney’s work shows is a particular instance where, as Jackson declares, “We don’t say what we mean to say. The sentence is not one, but a cluster of contrary tendencies.” Lexically, it inescapably is. Hypertext works to subvert and call attention to a multiplicity of meaning inherent in any text. Its point of intervention is that interstitial moment when we are at the business of unconsciously sloughing away meanings that fail to fit a larger frame of narrative, where we are reading, literally, between the words.

(Part 2 is here.)

One thing, or many? Is a monster a fully integral creature, or literally or etymologically, several? Stitches mark the external evidence of internal division in Frankenstein’s monster, suturing together a gathering of disparate parts made animate. As the parts are ill-fit together, the monster is ill-fit to the world, internal relations forming, foreordaining outer.

Samuel Taylor Coleridge uses the chimera to illustrate the disunity he perceives in a work of literature that seems patched together. The chimera exemplifies what he calls “mechanical form” in literature, the lesser of two kinds of artistic production, the more capably integrated and naturally constructed counterpoise to which he calls “organic.” Frankenstein’s monster breaches this distinction of Coleridge’s, dissolving the boundary between an exalted formative principle and a deprecated one, between the interwoven textures of life and assembly in the style of gears and gadgets.

Chimera depicted on plate

How stable, though, is the category of the “organic?” If something organically formed is somehow uncontrived and with better fit among its various parts than the chimera, we should be able to see how its parts make sense together.

Look at this bird:

Bald eagle

And this:

Pumpkin the presidentially pardoned turkey

Or this:

Hornbill bird

Or the mandrill:

Mandrill

How could you say, if you had never seen a hornbill, or a mandrill, that these creatures’ parts fit together beyond their apparent points of attachment? The head of the bald eagle looks to me entirely out of place in its coloration and just as ill-fit to the body as the head of a chimera to its. A turkey’s head, a mandrill’s face? In their superficial anatomy, they visually clash with their bodies.

By contrast, the digital image artist Martin of Humandescent.com unites visually disparate animals’ parts, using morphing techniques to blend and metamorphose images of different creatures into graceful conjunction with one another. One of my favorites from this site is Kittguin, formed from a blend of a penguin’s body and a black-and-white kitten’s head, the color patterns of head and body assimilated with one another. Fetch! seems to have brown fur, rather than feathers, continuing the dog’s facial texture across its body, and Crog, the crow-dog, has black feathers that blend texturally with the salt-and-pepper jowls of a hound.

Kittguin from Humandescent.com

Fetch! from Humandescent.com

Crog from Humandescent.com

The techniques Martin applies in these pictures answer a question that Coleridge’s use of the chimera only implies: What is a chimera’s graphic opposite? Applied to a living creature, what does the formal transformation from chimerical to organic look like when visualized? In electronically morphed images like these, the Coleridgean chimera rehabilitated boils down to a kind of mixing or exchange of surfaces and textures, or of finer-scale parts among larger-scale ones. Taking these images as examples, a distinction of organisms from badly integrated synthetic creatures seems at least in part to be about texture and scale.

By performing these cosmetic operations on grafted-together images, though, does Humandescent.com defuse the affective potency of composite creatures, taken as monsters? Can we really call Kittguin a monster, as a grafted-together creature? Isn’t it just cute? How does Dowlog (below) look to you? Monster? Not a monster? What about Robin Red Thug? To my eyes, it is not texture per se that makes these images a little disturbing, but the presence of a head that I do not expect in its bodily context. Is it evidence of a composite status what gives Dowlog its affective impact? Or the inclusion of the almost otherworldly ferocity of an owl’s visage?

Dowlog from Humandescent.com

Robin Red Thug from Humandescent.com

If we see a monster like Frankenstein’s as an assemblage of mismatched parts, a Coleridgean chimera brought to life, and then apply pressure to this model, we find weaknesses in it: actual organisms in the world are not always morphologically unified in the way Coleridge imputes, while visibly unified organisms, unchimerical in appearance, may still appear monstrous to us. The chimera, if taken as a model for monstrosity, has properties that turn out to be insufficient to account for monstrosity itself. But are they necessary? The abstracted, formalized idea of the chimerical does not take into account the acculturation that has made some animal forms familiar to us but others bizarre. It does not take into account the deeply embodied nature of our minds and our capacity for revulsion and fear, something we can feel so readily for loathsome creatures, unified in appearance or not. Textural unity after all does not efface the strangeness, the mismatch with expectation, that a novel creature presents. Some morphed creatures still resist unification.

It makes a kind of formal sense that a monster unassimilable to the world around it can be made of parts unassimilated to one another. Beyond this, Coleridge’s rough equation of chimerical form with mechanical has another advantage, when we turn it back on the question of what is monstrous: it gives us the kernel of an explanation of why we would see something entirely inorganic, a robot, also as a monster.

Magnus, Robot Fighter with giant robot

I’m coming to the poem in the era of “ubiquitous computing,” where we are tracked in our transit through myriad electronic media, watched over in an everywhere-dense, ever more finely granular manner by people and systems we have not asked to do this for us. We do not have to imagine being watched over by machines; we are. In the past, the tools of mass surveillance have not had a happy history of use. In imaginative literature, they are a staple of dystopia.

“But,” Jen conjectures, “maybe… Brautigan’s speaker is a genuine optimist?” With this in mind, I reread “All Watched Over by Machines of Loving Grace,” seeking evidence for either case. What I discovered surprised me, though pleasantly: I could find no explicit support for a foreboding or dark reading, and a preponderance of support for the contrary. Throughout, exclamatory parentheticals strain at the bonds of its present, three times urging: “right now, please!” or similar. Throughout, the emphasis is on harmony, mutuality (“mutually programming harmony”), peace (“where deer stroll peacefully”), liberation (“free of our labors”), Edenic return (“joined back to nature”), primitive domesticity (“returned to our mammal brothers and sisters”), affectively positive all the way from opening “I” to closing period. Everything in the poem rings with unalloyed enthusiasm. The poem seems to offer no intrinsic home to foreboding.

Brautigan tells us directly, too, what he feels so exuberant about.

In his envisioning, electronic machinery, an artifact of human craft, is naturalized, reintegrated with the nature it has emerged from and, imbued with agency, assumes a reciprocal relationship with living creatures. The computers, “watching over” all in title and the last two lines, are themselves now in some sense alive; and figuratively or not, they are imagined to someday “live together” in harmony with mammals (lines 4-6).

Harmony in this poem lives in the mutuality of agency between mammals and computers, in the “cybernetic meadow” (stanza 1, line 3), “cybernetic forest” (stanza 2, line 3), and “cybernetic ecology” (stanza 3, line 3). Reciprocity is essential to the original idea of the cybernetic. As developed by Norbert Wiener and those who came after him, cybernetics is a multidisciplinary study of regulatory systems (<http://en.wikipedia.org/wiki/Cybernetics>)—closely related to the mathematical theory of control, and control theory is fundamentally a theory of feedback loops, of actions and reactions, a give and take between causes and effects, inputs and outputs. Brautigan has imagined what physicists call a dynamic equilibrium, an equilibrium of flows: “Pure [liquid] water / touching clear sky.” Across this boundary water circulates, liquid to vapor, vapor to liquid, endlessly. Schematically, this reciprocity is a loop.

Figure 1: Cover of the 1954 paperback edition of Norbert Wiener's _The Human Use of Human Beings_, with reciprocating arrows.

Figure 2: U.S. Geological Survey illustration of the water cycle, in which water circulates through Earth's ecosystem, a cycle of action inviting cybernetic analysis. http://en.wikipedia.org/wiki/Water_cycle

We have come far from the cultural moment into which Brautigan brought his poem. In our time, a kind of anxiety has interposed itself into the harmony Brautigan sees beckoning. This anxiety is sufficiently strong to reverse the poem’s gestalt, to flip it over to an ironic reading contrary to the one the poem most directly supports. While an ironic intent cannot be ruled out, nothing seems to be objectively present in the poem to disrupt its imagined harmony.

Naturalizing the artificial, Brautigan’s poem enacts a classically romantic breakdown of categories—the fusion of ideas that Samuel Taylor Coleridge so persistently espoused.  What it operates on, though, are the formative tropes of romanticism itself. It doubles back on romanticism’s pastoral commonplaces, uniting nature with the most highly evolved of human artifice, computer machinery. The poem is paradigmatically techno-romantic. It is a techno-pastoral.

When we acquire a skill, it becomes a “second nature” built upon the original, as if we were born knowing how to do it. It starts out artificial and becomes natural, as we cease to have to think about performing it. In the same way technology, once adapted to the world around us, becomes as if it had always been there. Thinking about how the artificial is assimilated into the natural, and the unaccustomed becomes accustomed, I became curious about the career of the phrase “second nature.” I ran a quick Google Ngram Viewer query:

Figure 3: Google Ngram viewer for "second nature," from 1740 to 1900, with a 3-year smoothing applied. http://books.google.com/ngrams/

Curiously, the bigram “second nature” suddenly ticks upward in frequency around 1790, taking off rapidly later in the decade and on through the early years of the next century. That is, the phrase “second nature” came to prominence (in Google’s dataset) over the same time course that the first generation of English romantic poets came to maturity.