Introduction
What is the concept behind the Greek New Testament Grammatical Commentary (GNTGC)?
A. How did GNTGC get started?
During typical studies of the Greek New Testament and its commentaries, the ideas for GNTGC were born from the desire to discover new tools to help settle old exegetical, textual and translational questions. And equally, they were born from the desire to set the descriptions of all the grammatical relations and constructions in the GNT in good and consistent order. Students like to have a big framework to hang the details on. How could the language patterns in the GNT be identified, described and put at the ready disposal of GNT researchers?
B. A model for GNTGC
A good model for a classification system is Linnaeus. He wanted the classification of plants and animals to be scientific and defensible. He wanted the living kingdoms to be arranged logically for the sake of a comprehensive, orderly picture. And he wanted that set of terminology to be available to all the students of living species as an aid to good communication, research and comprehension. Linnaeus, like the scientists of his day, was interested in discovering the laws of the Lord’s creation, not in inventing theories or making a name for himself.
Like the scientists of his day, he wanted the simplest explanations consistent with all the observations. He believed that when all the phenomena and their descriptions are set forth in an orderly whole, a consistent picture emerges in which the parts provide controls to make themselves fit the whole and give it a beauty and integrity, because that is the character of God and his creation.
C. History of Syntax Ebooks
Such a system for the GNT is called a grammar if the syntax half of the equation is properly developed. Dionysios Thrax started such a classification system for the Greek language about 200 BC. His parts of speech were both morphological and functional. Many Greek grammars have been written since then, treating both morphology and syntax. In the age of computers, such a book should be interactive and comprehensive, identifying all the instances in the GNT for the phenomena—cavernous actually compared to paper books. Such ebooks exist for morphology (lexemes and parsing), but in the syntax area they are rudimentary. There is no GNT ebook that enumerates the language patterns and their verses beyond an elementary stage. It is an opportunity waiting for development.
D. History of GNTGC
Gradually, lots of attributes were added to the basic GNT text and coded as interest arose. A big advantage of a GNT coded for syntax attributes (not merely morphology) is that it is available to develop syntax rules (both general and absolute). The rules can be subjected to rigorous programmatic quality control over all the thousands of words in the whole GNT. A coded syntax data base is a factory for testable hypotheses and for the improvement of the rules and their generalizations and consistency. If it is done right, it can be the basis for discoveries and GNT grammatical improvements for generations to come.
The syntax coding for GNTGC has been an organic development stretching over many years. It now consists of hundreds of relational files and a couple thousand attribute columns. The head-dependent syntax relations and their grammatical relations have been coded for a sample of about a quarter of the GNT. The morphology, originally supplied by Gramcord, has been significantly extended, for the whole GNT. The future plans are huge. The whole planned work is barely begun, but even at this stage it is already able to supply evidence for deciding many questions that arise in the studies of the GNT verses.
E. The Use of Language Patterns to Add Evidence for Exegesis and Translation
Many translation differences result from lexical or theological considerations. For example, when one translation has “servant” and another has “child,” readers are puzzled, because they do not imagine that the two English words could refer to the same Greek original, but students of Greek understand that παῖς may be translated either way. Syntax evidence is more complex.
The evidence from the syntax coding often takes the form of determining the original Greek syntax underlying each of the differing interpretations and then making a statistical search for the precedents for such constructions. If one of the interpretations follows from a syntactical construction that occurs often and the other occurs rarely or never, this is taken as a factor weighing in favor of the normal language pattern and against the exceptional pattern. This statistical evidence often brings a new tool into the debate that was previously unavailable and may tip the balance. In some cases, where one of the interpretations depends on a construction that never occurs elsewhere, there is a question whether the interpretation even represents good Greek.
F. An Illustration of GNTGC Use Applied to an Exegetical Question
To illustrate a common use of GNTGC, let’s take a case. Suppose the student is reading Hebrews 1:6 and looks at a multi-version display, say of Bible Hub, for confirmation of the correct interpretation. It then it becomes apparent that the translators have connected the word πάλιν (“again”) at the beginning of 1:6 differently. One translation says, “And again, when He brings his firstborn into the world, he says,” and another translation says, “when He again brings his firstborn into the world, he says.” We think they cannot both be right, so we look at the commentary. Some commentators think that the passage cannot mean what it says because God is not bringing his firstborn again. One respected commentator tells us that there is no way to be sure which is correct. This conclusion puts the decision in the translator’s hands, right where we do not want it. Other commentators are pretty sure that the normal interpretation is right and we should not alter it to fit our theology. Where can we find objective evidence to help decide and transfer the problem from the arbitrary to the evidentiary sphere?
G. GNTGC Sectioning of the Example Verse
Here is the verse as sectioned by GNTGC. Sectioning is a method, based on the syntax diagram, of grouping the words into hierarchical verbal units, one unit per verb (shaded). A tree diagram would also accompany the sentence. GNTGC has favored one of the two syntactical constructions, the one that puts πάλιν inside the when-clause.
5 Τίνι γὰρ εἶπέν ποτε τῶν ἀγγέλων·
υἱός μου εἶ σύ,
ἐγὼ σήμερον γεγέννηκά σε;
καὶ πάλιν εἶπέν·
ἐγὼ ἔσομαι αὐτῷ εἰς πατέρα,
καὶ αὐτὸς ἔσται μοι εἰς υἱόν;
6 ὅταν δὲ πάλιν εἰσαγάγῃ τὸν πρωτότοκον εἰς τὴν οἰκουμένην,
λέγει· καὶ προσκυνησάτωσαν αὐτῷ πάντες ἄγγελοι θεοῦ.
For to which of his angels did he ever say,
“You are My Son;
today I have begotten You”?
And again he says,
“I shall be a Father to him
and he shall be a Son to me”?
But when he again brings his firstborn into the world,
he says, “and let all God’s angels worship him.”
H. The Interpretation Options
We assume that the student is comfortable looking at what the text actually says first, rather than correcting it to what the commentator thinks it must mean. The first step in the investigation is to identify the two underlying Greek syntactical constructions. In the GNTGC option 1, πάλιν modifies εἰσαγάγῃ in the when-clause. In the option 2, πάλιν modifies λέγει in the main clause. Option 2 looks wrong right off because in order to get it into English, we have to transpose some words. However, Greek word order frequently differs from English. How do we know where and when order can be altered? Besides, we have some experts telling us that transposing is OK here. Who makes the Greek rules? Can experts make them? Experts have a huge degree of authority for students.
I. The GNTGC Method of Developing Language Patterns as Evidence for the Options
The GNTGC method of bringing new syntax evidence to bear on a disputed interpretation is to identify the underlying constructions and see which one is regular and which one is exceptional or unprecedented. However, there are thousands of constructions, and many of them are ad hoc. And the constructions themselves are based on GNTGC attribute and coding choices. Why would the student trust GNTGC syntax coding any more than the claims of experts?
The first answer to this legitimate question is that GNTGC coding is based on hundreds of programmatically enforced rules, which themselves are based on a large population. They automatically develop some internal consistency. And they tend to be based on principles. The level of consistency will exceed that of the unaided mind.
J. Greek Specific Theoretical Support for the Language Patterns as Evidence
The reader should consider that over 85% of GNT words are inflected, and that inflected words can signal their mutual relations regardless of order. That is where the Greek order flexibility reputation comes from. The uninflected 12.5% of the words need juxtaposition to establish connections. Uninflected words in Greek tend to be as position-dependent as in order-driven English. We begin to suspect that the experts do not have as much latitude for choice as they think they do for πάλιν.
To follow the next bit of evidence, we need to discuss how GNTGC creates word structures within sentences. First, each (dependent) word is associated with a head, if possible. Half of the words have no dependent, and 7% of the words have no head. This process establishes a set of structures within structures. A structure is a word and all its subordinates down to the terminal words. A verb phrase (VP) is a structure headed by a verb.
Scientific research is experimental. It starts as a search for evidence that supports a hypothesis. Often, the hypothesis is the one that the researcher guesses should be disproven. The researcher is not biased in favor of a particular hunch and will be glad to find strong statistical evidence for any hypothesis that bears on the question. Often, the researcher will need to test multiple hypotheses to find the best discriminator between the options. Making an inconclusive start does not invalidate the process. It is not important to pick the right starting point. Perseverance counts. Notice that GNTGC itself does not supply the hypotheses, but does supply the tools to test them.
K. First Try
We like to make our rules as general as possible to have a broad scope and a large sample population. Suppose we start with the hypothesis that adverbials should modify the nearest verb to the left or right. If we could show that that is a grammar rule, that would cinch option 1 because option 2 has πάλιν modifying the distal λέγει instead of the proximal εἰσαγάγῃ.
We find 7516 adverbials to verbs in the coded sample. However, testing our initial hypothesis, we find 984 exception constructions where the adverbial modifies a distal verb with some other verb intervening. This is a 13% exception rate—not even good enough for a general rule.
L. First Modification
Suppose we speculate that the ratio of exceptions might be different for finite verb clauses, since both verbs in Hebrews 1:6 are clausal. Therefore we modify our test constructions. We first find 5765 adverbials modifying clauses. There are 338 exceptions where the adverbial skips the proximal finite verb. This is about a 6% rate, better than the general 13%, but still not a very good discriminator.
M. Second Modification
We notice that a lot of the adverbials are verbs. We consider restricting them to class adverbs, fitting Hebrews 1:6. We bear in mind that any sustained process of adding conditions will narrow down the sample size and subject the result to accident or insignificance if the sample size shrinks too much. We first determine that there are 526 such adverbial adverbs modifying finite verbs, a respectable sample. And we find only six exceptions, where the adverb skips over the proximal finite verb to modify the distal one. This proposed general rule brings the exception rate down to about 1%, a lot better than the previous 6%. This would qualify as evidence in favor of option 1, although without disproving option 2.
N. Third Modification Supporting Option 1
On examining the six exception verses for syntactical patterns, we notice that all the intervening proximal clauses are subordinate to the distal head of the adverb. We observe that this is not the case in option 2. So of course we can add this restriction to the hypothesis and get an absolute statistical rule in our sample that option 2 does not happen. There are no instances in our sample of a construction like option 2 where an adverb skips over the nearest finite verb to modify a distal one and where the proximal verb is not a subordinate of the distal one.
Here we have achieved our initial ideal goal of discovering an absolute discriminator construction between option 1 and 2. Although it is adequate, it bothers us a little that the “rule” is not as intuitive as we would like. We would like children learning the language to be able to get it.
O. Branching Out in a New Direction
So far we have been thinking along the lines of component order. We want to branch out and think along other lines. We notice that πάλιν is inside the εἰσαγάγῃ-clause but not of it (not a subordinate of it) in option 2. We think of reforming our test hypothesis to say something like that the adverb that is physically in a VP must be a subordinate to the VP. That would be intuitive and would serve to exclude option 2.
At this point we transition from the exegetical arena to a syntactical perspective that is probably not especially familiar to students. It is part of GNTGC. εἰσαγάγῃ defines a VP. Each VP has boundaries. The initial word of the VP is clearly the subordinator ὅταν (GNTGC has validated rules to prove this). Since πάλιν is between the initial word and εἰσαγάγῃ, the head, it is clearly inside the VP physically. The GNTGC tree diagram shows that πάλιν is not inside the VP syntactically, i.e., not a subordinate of εἰσαγάγῃ, in option 2. In sectioning this is called a gap in the verbal unit. This gap does not show up in the sectioning above because that is based on option 1.
We have to be careful with this line of speculation. Clearly the postpositive δὲ does not follow our proposed rule. It is in the subordinator clause but not of it. Maybe πάλιν can just attach itself to δὲ and permit option 2. So we exclude postpositives in our new hypothesis.
P. The revised Hypothesis in Favor of Option 1
The revised hypothesis is that a non-postpositive inside a VP must be subordinate to the verb. We have reverted to verbs in general instead of just finite verbs in order to generalize and expand the sample size. On testing the instances of non-postpositives that fall inside a VP but are not of it (not subordinates of the head), we find 222 exceptions. This is too large a number, but we notice that only 3 of the exceptions are adverbs. This would make a respectable general rule.
However, on examination of the three exceptions, we discover that the containing VP does not intervene between the adverb and its head. So we add the intervening condition. The final “rule” is that an adverb cannot be in an intervening VP as a gap and modify a distal verb. This is true for all 702 coded adverbs that modify verbs adverbially. There are no exceptions. We prefer this form of evidence over the original result as being more intuitive. It goes along with the insight that Greek speakers, like English speakers, must rely on proximity to make the connections for uninflected words like πάλιν. Otherwise, the language fails to communicate and degenerates into guesswork. Note that the SQL to find the list of these cases depends on GNTGC feature called structure architecture rather than merely on order attributes.
To summarize, πάλιν is in the εἰσαγάγῃ-clause physically, to the right of εἰσαγάγῃ. This is a fact because the subordinator ὅταν is also in the clause. Therefore, the reader would have no way of knowing that πάλιν modified the distal λέγει, as option 2 would have it. However, if πάλιν were to the left of ὅταν and to the right of the εἰσαγάγῃ-clause but not in the clause, it would still be possible for πάλιν to modify λέγει, as examples demonstrate. Option 2 is not a confirmed language pattern, but that one is.
The reader may note that normally the statistics in favor of one option and against another are not as decisive as 700 to 0. They might be 20 to 2. That is not decisive, but it is a factor to weigh; it is still interesting information to consider. At other times some construction may be unprecedented in either option. While not contributing to a conclusion, it is still better to know that fact of uniqueness than not to know it.
Q. The Need to Exhibit Examples for Options
Of course, the proponent of option 2 may still say that this is the exception to the rule or that the results depend on the coder’s practice. Proponents may have many theological or semantic thoughts. However, this is not a valid response. Science is based on a testable, falsifiable hypothesis. It is based on devising fair tests and letting the chips fall where they may. So the response to this challenger is that, principles of uninflected order, English order itself, and strong statistical evidence have produced a presumption for option 1 against option 2. Now it is the part of the defender of option 2 to come up with some other counter-examples where the translation favors the method of option 2 (namely, making an adverb that is physically in a subordinate clause modify the main verb).
This is based on structure architecture, which is part of the underlying patterns of language, and probably a tool formerly unavailable. It is probable that the proponent of option 2 will be unable to find counter-examples. If found, the examples will probably also be disputed.
R. Canned and Ad Hoc Constructions
At this point, a word about the constructions in GNTGC is in order. Constructions are the language patterns of Greek. There are two kinds of constructions: canned and ad hoc. The canned constructions are those that are available for the student to pick out of an GNTGC menu to see all the examples. Hundreds of grammatical relations are canned constructions, but thousands of other constructions need to be coded and made available. They are as varied as imagination, but only valid if confirmed by programming. Within the canned constructions there will be many options for variables like lexemes or word class, or parsing or a large variety of other attributes to tailor permutations to specific verse questions. The other kind of construction is ad hoc, for example, the result of a programmer SQL query. This depends on the query constructor’s SQL skill and a knowledge of the GNTGC attributes. This example depends on ad hoc constructions. The repertory of canned constructions will build up over the years of use.
S. How Would Greek Express Option 2?
In order to demonstrate the invalidity of option 2 it is not necessary to broaden the constructions under consideration. However, curiosity about the normal Greek expression for option 2 is a legitimate question: “and again when he brings his firstborn into the world he says.” So we frame it in syntactical terms and pose the question to GNTGC: how many times does an adverb modify a clause where an intervening VP occurs as in option 2? We find 58 constructions with an intervening verb. In all of them, where the intervening VP, like the preceding adverb, modifies the distal main clause, the intervening VP is a circumstantial participle.
We conclude that the normal Greek way to construct the idea expressed in option 2 would be: πάλιν δὲ εἰσαγαγῶν τὸν πρωτότοκον εἰς τὴν οἰκουμένην, λέγει· καὶ προσκυνησάτωσαν αὐτῷ πάντες ἄγγελοι θεοῦ, “And again, having brought the firstborn into the world he says, ‘And let all the angels of God worship him.’”
T. The Normal Greek Construction for Option 2 Merely Reinforces the Conclusion for Option 1.
Knowing how the GNT would probably express option 2 reinforces our idea that the actual form of Hebrews 1:6 probably did not intend option 2, even though it makes more sense to some exegetes than option 1. This knowledge of the Greek language circumstantial patterns merely reinforces the main conclusion. The main conclusion is still based on the fact that πάλιν is physically inside the εἰσαγάγῃ-clause and thus must be subordinate to it if it intervenes.
U. Christ’s Incarnation or Coming Again in Glory: the Reason for Choosing Option 2 May Be Not Knowing the Syntactical Evidence, or Putting Assumptions before Grammar.
In the commentaries, the dispute over which verb πάλιν modifies is often connected with a choice between the first and second coming. The semantic difficulty that some commentators have with option 1 may be because they put their assumption before textual analysis or do not know the syntactical constraints on their imagination. They often assume that εἰσαγάγῃ refers to Christ’s incarnation, and that the angels are possibly the angels in the Luke nativity story. Under that circular assumption, option 2 makes more sense than option 1.
However, the right way to do exegesis is to start with what the text actually says and fit the interpretation to that. If we assume that the text means what it says, that God will bring his firstborn into the world again, εἰσαγάγῃ more naturally refers to his coming again in glory with the angels. Option 2 (apparently started by Martin Luther, not the Vulgate) encounters a severe syntactical obstacle, but option 1 does not encounter any serious objection; so evidence from Greek language patterns provided by GNTGC strongly favors option 1 over option 2.
—Dennis Kenaga