Pages in this blog

Sunday, December 15, 2019

Some informal remarks on Jockers’ 3300 node graph: Part 3, Signs and mechanisms

I want to start at the point where I ended my previous post in this series, Some informal remarks on Jockers’ 3300 node graph: Part 2, structure and computational process. I was arguing that the issue of scale was misconceived. The fundamental issue is NOT a corpus of texts versus one or a handful of texts. That’s trivial. The issue has to do with the terms of analysis. So-called close reading exists within a conceptual and methodological discourse which is quite different from that of so-called distant reading.

The difference is one of conceptual ontology, as that term has come to be understood in computer science and the cognitive sciences. In that context the issue is not the ultimately real, a philosophical question, but the kind of concepts we are we using. I start with a simple example, salt versus sodium chloride and then build out from there to words versus signifiers and then on to texts and meaning.

Conceptual ontology

Consider for a moment, salt on the one hand and NaCl (sodium Chloride) on the other. Physically they are (almost) the same substance, but conceptually they are quite different. Salt is defined and understood in terms of its physical appearance and, above all else, its taste. We can taste salt even when we cannot see it, and so can animals. NaCl, however, is defined in terms of an atomic theory of matter that didn’t exist until early in the 19th century. Moreover, NaCl is a pure substance, consisting of nothing by sodium and chlorine atoms; salt on the other hand will always have some impurities. Thus, strictly speaking, salt and NaCl are not physically the same, close, but not exactly the same.

Salt and sodium chloride, then, are used in different intellectual contexts, each of which has its own vocabulary. Salt belongs with sugar, pepper, cinnamon, flour, and so forth all substances having to do with food, food preparation, and eating all related concepts. Sodium chloride is related to potassium chloride, sodium hydroxide, and so forth, electrolysis, ion-exchange, and so forth, for a long list. Thus we have two different conceptual worlds, each coupled to characteristic actions and processes, but both ultimately grounded in the same physical reality. Someone whose occupation has them working in the sodium chloride world has no trouble with table salt at mealtime. The transition from one world to the other is seamless, or nearly so (there may be a change of clothes involved).

The worlds of “distant reading” and “close reading” differ from one another in the same way. The physical texts and the symbols imprinted on them are the same in both worlds, but the concepts and methods of description and analysis are different. It’s that (kind of) different that Geoffrey Hartman had in mind almost a half century ago when he observed: “modern ‘rithmatics’—semiotics, linguistics, and technical structuralism—are not the solution. They widen, if anything, the rift between reading and writing” (The Fate of Reading, Chicago 1973, p. 272). He expresses himself using the standard trope of distance, but he’s certainly not talking about distance in any physical sense – no one is.

These days we might want to gloss distance in terms of explicit mediating steps. In close-reading you read a primary text and then does some thinking, perhaps some secondary reading and research, and then you write about that primary text. Just how you break that down is somewhat arbitrary, but it’s nothing like what happens in distant-reading, which starts when a collection of primary texts is digitized. The digitized texts are then cleaned up, tagged with metadata, organized in a database or databases, and then subject to analysis, which might be a straightforward matter of word counts, or the somewhat more strenuous process of topic modeling, or perhaps we’re going to make use of vector semantics, or any of a number of things. This more complicated process is likely to involve two, three, or more people at various stages along with way. At some point the analytic process will produce some set of visualizations and they, in turn, will be interpreted in terms appropriate to the primary texts (and their contexts).

As I said, that’s how we might gloss the close vs. distant distinction these days. The distance isn’t one of physical steps, but operational steps. However, it’s doubtful that Hartman had anything like this in mind. To be sure, stylometrics existed in 1973, and Stanley Fish was doing his best to skewer it, but I doubt that Hartman was thinking about stylometrics when he made that remark. However, he might well have been thinking about the kinds of tables and diagrams that Lévi-Strauss used, or that showed in linguistics article, those diagrammatic signs that interrupted the linguistic flow. They are indices of a different mode of thought, a different conceptual ontology.

Words and signifiers

We can begin to appreciate that difference by noting the difference between words, which are signs in Saussure’s sense, and signifiers, which are components of signs. Literary critics generally talk of words, and occasionally of signifiers. The concept of word is transparent enough in ordinary casual discourse. As ordinarily understood, the concept of words encompasses pronunciation, spelling, grammatical usage, and meaning, often several meanings. Words in this sense are the things listed in dictionaries. And for the most part, literary critics deal in words, even those who’ve read a bit of Saussure.

Saussure distinguished between the signifier and the signified. The signifier is a physical entity, either sonic or visual. The signfied is a mental entity; it is the bearer of meaning. Signifiers are public; they can be transmitted between people. Meanings are, well, they’re not exactly private – Wittgenstein established that in a famous section in his Investigations – but as mental objects they exist in people’s heads and we do not have direct access to one another’s heads. Thus, as William Croft has argued in chapter 4 of Explaining Language Change [1], word meanings are negotiated in conversational interaction with one another.

Computational critics, in contrast, may talk of words, but what they are actually working with are signifiers, signifiers rendered in digital form. That is to say, that’s what’s in the databases at the heart of computational work, mere signifiers. The computer has no access to word meanings, to signifieds; it only knows the written signifier. Consider the following passage from the well-known 1949 memo on machine translation written by Warren Weaver, who headed the Natural Sciences Division of the Rockefeller Foundation:
First, let us think of a way in which the problem of multiple meaning can, in principle at least, be solved. If one examines the words in a book, one at a time as through an opaque mask with a hole in it one word wide, then it is obviously impossible to determine, one at a time, the meaning of the words. "Fast" may mean "rapid"; or it may mean "motionless"; and there is no way of telling which.

But if one lengthens the slit in the opaque mask, until one can see not only the central word in question, but also say N words on either side, then if N is large enough one can unambiguously decide the meaning of the central word. The formal truth of this statement becomes clear when one mentions that the middle word of a whole article or a whole book is unambiguous if one has read the whole article or book, providing of course that the article or book is sufficiently well written to communicate at all.

The practical question is, what minimum value of N will, at least in a tolerable fraction of cases, lead to the correct choice of meaning for the central word?
While that memo catalyzed early work on machine translation, the approach suggested in those paragraphs played no role in that work. The necessary computing power wasn’t available. That changed in the 1980s and especially 1990s.

The topic analysis technique that Jockers used follows from the insight expressed in those paragraphs, as does the work on vector semantics. The idea is simple. Words that appear together frequently must somehow have related meanings. Race, horse, saddle, jockey, and track, all mean different things, but they belong to the same discourse and so will appear in close proximity when that discourse is spoken, or written. And so it is with human, race, IQ, identity, American, black, and white, these too belong to a discourse and so will co-occur in texts using that discourse. Notice that race is common to those two very different discourses. Considered alone, without context, the meaning of race is ambiguous, we don’t know what it means. But if we see race in close proximity to horse, we’ll locate it in one discourse while if we see it in proximity to IQ we’ll locate it in that other discourse.

Signifiers in themselves have no meaning. But in actual usage signifiers are always bound to a meaning and that meaning connects them to related signifiers. Given a large enough body of texts, computers can determine which signifiers occur together. It is up to the investigator to make a judgment about why those signifiers are occurring together. The investigator will, of course, make that judgment on the basis of their knowledge of the language.

The conventional literary critic finds such judgments utterly trivial and has no way, no intellectual context, for understanding how remarkable it is that mere computation can discover such relationships. And since that critic is only interested in a handful of texts they have no reason to reconsider that judgment and consider the possibility that there is something to be learned in examining such patterns in collections of texts, such as the collection of 3346 Anglophone novels Jockers has been working with.

Text and meaning

But what is a text? For the linguist, or for the computational critic, and answer is simple: a text is a string of characters.

Things are not so simple for the literary critic. Yes, sure, a text is a physical object, a scroll or a codex, and it is a bunch of signifiers inscribed on such. That’s trivial and, for the most part, uninteresting except in very well defined contexts, such as physical preservation or, more interestingly, the preparation of an edition. And the preparation of critical editions has been very important in the development of digital humanities broadly conceived, but still, we can bracket that.

What I am interested in here is the text as an object of interpretation. THAT text is an enigma [2]. The following passage is from the introduction Rita Copeland and Frances Ferguson prepared for five essays from the 2012 English Institute devoted to the text:
Yet with the conceptual breadth that has come to characterize notions of text and textuality, literary criticism has found itself at a confluence of disciplines, including linguistics, anthropology, history, politics, and law. Thus, for example, notions of cultural text and social text have placed literary study in productive dialogue with fields in the social sciences. Moreover, text has come to stand for different and often contradictory things: linguistic data for philology; the unfolding “real time” of interaction for sociolinguistics; the problems of copy-text and markup in editorial theory; the objectified written work (“verbal icon”) for New Criticism; in some versions of poststructuralism the horizons of language that overcome the closure of the work; in theater studies the other of performance, ambiguously artifact and event. “Text” has been the subject of venerable traditions of scholarship centered on the establishment and critique of scriptural authority as well as the classical heritage. In the modern world it figures anew in the regulation of intellectual property. Has text become, or was it always, an ideal, immaterial object, a conceptual site for the investigation of knowledge, ownership and propriety, or authority? If so, what then is, or ever was, a “material” text? What institutions, linguistic procedures, commentary forms, and interpretive protocols stabilize text as an object of study?
What? “Linguistic data” sounds like it might be mere signifiers, and perhaps “copy-text” as well. But the rest of them, those various sites “for the investigation of knowledge, ownership and propriety, or authority”, those texts clearly consist of words, in the ordinary sense, complete with their multiple and often ambiguous and contradictory meanings.

That is the text that is conceptualized in vague spatial metaphors. Do we talk of meaning as being IN the text or as being somehow outside the text, in the CONTEXT? I think it would be a mistake to interpret that as a vague assertion about mechanism, though I am sorely tempted to do so. Rather, it’s a statement about critical methodology. If you think of meaning as IN the text, then you do not invoke anything but the text itself in the process of interpreting it. That’s what was originally meant by close reading. But if you mean that meaning is located in the context, well then you must bring other evidence to bear in your close reading. Depending on your particular methodology you may call on historical materials of various kinds – newspapers and other contemporary periodicals, correspondence, legal documents, and so forth – or you may want to mount a psychological argument in terms of this or that theory, diaries, and what have you. And of course the text itself.

And then we have the idea of “hidden” meaning. That’s clearly a spatial metaphor. But a metaphor for what? Not the text itself; it’s not as though anyone is imagining a secret text bound into the spine or sandwiched into the front or back boards. The idea seems to be something like a large and elaborate old house with hidden rooms and passages. Expressed in that way it seems just a bit foolish, but then no one ever expresses it in that way, do they? Again, we’re dealing with a covert methodological injunction. Words have many meanings and are linked to other meanings through various figures, such as the master figures of metaphor and metonymy – need I invoke, for example, Roman Jakobson’s work on the subject [4]? And so, as a methodological precept, one must explore those possibilities.

The fact of the matter is that, by contemporary standards, those of the so-called cognitive revolution and after, literary criticism lacks an account of language mechanism, whether in the reader, the author, or the critic. It is non-mechanistic or pre-mechanistic and certainly non-computational – an issue I’ll take up in the next and last post in this series. For now I want to end on the assertion that meaning, as the object of critical investigation, is inherently and irreducibly subjective. That doesn’t mean that critics cannot agree on the meaning of texts, for subjectivity necessarily implies intersubjectivity. It means only that meaning resides in subjects, in readers, writers, and critics and is not subject to objective determination. It is thus like color, which is also subjective in that sense. Sensations of color may be closely related to wavelengths of light, but cannot simply be reduced to them. Color arises within subjects, though psychologists have made a great deal of progress in figuring out how color perception functions in humans and animals.

The objective study of meaning, that is, of semantic mechanisms, has not yet been so successful. It was the search for semantic mechanisms that lead me to David Hays and computational linguistics, which I discussed in the previous post in this series, Some informal remarks on Jockers’ 3300 node graph: Part 2, structure and computational process.

References

[1] Croft, William (2000). Explaining Language Change: An Evolutionary Approach. Longman.

[2] I’ve written a number of posts at New Savanna about the concept of the text. They’re at this link: http://new-savanna.blogspot.com/search/label/text.

[3] Rita Copeland and Frances Ferguson, “Introduction”, ELH, Volume 81, Number 2, Summer 2014, p. 417.

[4] For example, Roman Jakobson and Morris Halle, Two Aspects of Language and Two Types of Aphasic Disturbances in Fundamentals of Language, The Hague & Paris: Mouton, 1956.

[5] I’ve got a good many posts on color, https://new-savanna.blogspot.com/search/label/color. They are of various kinds. Some are simply sets photographs where I was interested in colors and color contrasts. But others are about color perception in the mind and in photography. If you are interested in this topic you might what to search the web on “color perception”.

No comments:

Post a Comment