Disclaimer: The post below refers the implementation of SFL in NT studies, which may or may not necessarily reflect Halliday’s intentions. My goal is to address the methodological and theoretical flaws of the approach to SFL popularized by Stanley Porter. There is reason to believe that it represents a departure from Hallidayan linguistics (10/2/09).

This is my fourth post regarding some fundamental shortcomings of Systemic Functional Linguistics (SFL) as as applied to studies of Koine Greek, particularly those involving pragmatic choice. My focus today is on the area of syntax proper—word-order analysis—more officially referred to as information structure by linguists.

Two monograph-length studies have been conducted analyzing portions of the Greek NT using SFL. These include Gustavo Martín-Asensio’s 2000 volume Transitivity-based foregrounding in the Acts of the Apostles: a functional-grammatical approach to the Lukan perspective, and Ivan Shing Chung Kwong’s The word order of the Gospel of Luke: its foreground messages, published in 2005. Both use SFL as a methodology, but I will focus on the latter.

SFL, as applied in Kwong’s analysis studies the statistical distribution of given clause orders within the book of Luke. He groups them by kind of clause (primary, secondary, embedded), then ranks them using a classic SFL cline where the markedness or prominence of a particular order is inversely proportional to its frequency. In other words, the more uncommon a particular clause ordering is, the more it is said to stand out in the discourse, according to Kwong.

As in my other post on clines, this sounds very reasonable and compelling, until one considers the presuppositions that underlie the methodology. Discerning and considering presuppositions, both your own and those of others, is critical to good research. All too often presuppositions are overlooked or just ignored. Let’s take a look at some presuppositions associated with SFL’s statistically-based approach to word-order prominence.

In my review of Kwong, I used his identification of Complement-Complement-Predicator (CCP) and Subject-Complement-Predicator (SCP) as highly marked (i.e. as standing out in the discourse w.r.t prominence) as a test case (Runge, 4; see also Kwong, 106-107). He claims that since these orders are used infrequently by the writer, they are theorized to stand out in the discourse and be recognized as prominent by the reader. The passage to which he applies this claim is Luke 7:44-46.

SFL’s approach to prominence-marking in syntax theorizes that there is a correlation between markedness/prominence and frequency. Generally speaking I am in agreement with this, as long as there is appropriate interpretation of the data. In contrast to any other study I have seen on information structure, Kwong claims that infrequent clause orderings make the whole clause stand out stand out in the pericope or discourse. Prototypically, focus is placed on the individual elements within the clause, particularly the one(s) that makes the order so unique. To summarize, Kwong uses SFL to claim that the writer used a highly unusual word order to make the content of that clause stand out in its context. In order for this claim to be valid, a number of presuppositions must be true.

Authorial intent:
In order to assign meaning to the clause ordering, one would need to infer some kind of intentionality on the writer’s part. This need not be completely conscious, since most of the time we just phrase things in a way that “fits” the context. It seems reasonable to suppose the ancients did the same. If the writer was not intending to associate some meaning with the order, it would be absurd to make any such claim about prominence. Choice implies meaning. Removing intentionality makes this approach nothing more than a computational experiment, with no potential contribution to exegesis. Though Kwong never makes the claim explicit, he regularly infers intentionality based on the significance he assigns the rare orders.

Processing by the reader:
So if we assume authorial intent, the next logical question is whether humans actually process language in this way. When we hear an oddly-ordered clause, do we assign significance to the whole clause, or to the placement of a single element that makes it odd? SFL has opted for the former. If I were to say, “To the store must go I,” are you going to assign prominence to the whole clause within the discourse, or to “to the store” or “I,” depending on the intonation? Just for fun, let’s assume it is the former.

Multiple readings:
Assuming that humans really are wired to pick out oddly-ordered clauses in a discourse, there are other complications to this method. Kwong’s highly-marked example is from Luke 7, and the book consists of 24 chapters. The SFL statistical methodology postulates that the marked orders are determined by the overall distribution of the order within the overall text, in this case, Luke. How exactly is the reader expected to pick up on the use of a marked form early in a book, before they have processed the whole thing? Is it reasonable to expect that enough data has been crunched to perceive that the order in Luke 7 is marked?

What if the writer wants to mark something as prominent early in the book, like in a prologue? How is he to accomplish this? If one is going to make such a claim, then consideration needs to be given to the processing constraints faced by the reader. One solution to the problem of using marked forms early in the book is to assume that the book is to be read multiple times. This seems reasonable; after all it’s Scripture, right?

Assuming I am correct about multiple reading, does this mean that SFL’s statistical approach is only valid for sacred texts, not the kind of study of the Wall Street Journal or other such corpora that most computational linguists utilize?

Kwong omits any discussion of such matters. He moves directly from the statistical data to claims of prominence and markedness, without consideration of the processing constraints that it presupposes. But wait, there is more!

Ratio of the pattern to the overall discourse:
Another issue deals with the granularity of the ratio. How infrequent a pattern can a human really discern? How many times must something occur before a pattern can be discerned? How embedded or subordinated can a pattern be and still be picked out? Here is an excerpt from my review:

If the reader is expected to discern the rare collocations, how granular a discernment can be reasonably expected? Some marked orders cited for Luke 8:27b are ratios of 27 times versus 147 times, 78 times versus 306 times, and 7 times versus 27 times (109–10). Out of more than 19,000 words and over 2,500 primary clauses in Luke’s Gospel, how fine a ratio can an original reader be expected to have discerned? The lack of clarity regarding these sorts of presuppositional issues make it difficult to accept Kwong’s claims regarding the foregrounding effect of marked orders as currently described (Runge, 4).

Assuming this approach is valid and readers–both modern and ancient–are indeed able to identify such marked orders within the discourse, exactly how small a ratio can one reasonably expect a reader to identify? Only variation within primary clauses, or can it include secondary clauses, as defined by OpenText.org? How about embedded clauses, are they too granular? Based on some of Kwong’s claims, even the Masoretes would be hard pressed to have found what he claims is marked, and they were pretty skilled at such things.

Corpus-defined statistical significance:
One more presupposition needs to be mentioned, one that removes whatever plausibility might remain in this approach. Kwong notes the need for developing data for each author in preparation for discerning their particular use of marked forms by demonstrating how different authors have different default orderings in their discourse, statistically speaking. If this is true (and his statistics show it is), then a reader would need to potentially develop a new statistical framework for each author read! In other words, since Luke uses a different ordering than Paul or Mark, the reader would need to formulate new patterns for recognizing foregrounded clauses in Paul compared to Luke. But wait, there’s more! Depending on the ordering in Luke vs. Acts, or in 1 Corinthians vs. Philippians, one might even be compelled to develop a brand new framework for each book. No discussion was included about the implications of his claims. He seems to presuppose that it all works out. Again, this just seems highly implausible.

SFL’s statistically-based approach to syntax seems to be unique to New Testament studies. I have not seen another such study in the general field of linguistics. This study was published in the JSNTSS with virtually no discussion of the theoretical framework or the presuppositions associated with it. The main presupposition seems to be that the approach works.

Do you know where your presuppositions are? If you are keen to enter the field of discourse analysis, as it seems a whole gaggle of folks are, then chose your methodology carefully. Critically examine the presupposition that it entails, the foundation upon which it is built. Houses built on sand will not withstand the storms for very long.