Primary Research: The Root of Meaning

The Root of Meaning: Tree Diagrams Reveal the Complexities of Latin Grammar

By Henry Bauer

“All philosophy is the critique of language.”—Ludwig Wittgenstein

The modern discipline of theoretical grammar seeks to formally describe language, just as theoretical physics seeks mathematical descriptions of nature. The purpose of formal languages in linguistics is often taken to be the clarification of ambiguities in natural language. However, I would argue that encountering these ambiguities should lead us to examine them more deeply in the broader context of the relationship of language to thought. My research attempted in a small way to do just this. My work was unusual in taking as its material not the colloquial speech in terms of which these ambiguities are usually discussed, but an even more complex linguistic document: Latin poetry.

Last fall at the University of Nebraska I assisted Dr. Robert Gorman in a Latin treebanking project. Treebanking is the construction of sentence “trees” like the one below and their compilation in a database. The trees are used for grammar instruction and also historical criticism (that is, criticism which attempts to ascertain the origin and authorship of an ancient text). Treebank statistics measure how often authors employ various grammatical constructions so that unknown authors can be compared to familiar ones. But my interests were in sharpening my Latin skills and investigating grammar. I dealt frequently with complex problems in constructing trees for the sentences of Vergil’s Aeneid, a Latin epic poem composed around 19 B.C.

The grammatical framework I used is called dependency grammar. A dependency tree hangs each sentence element from exactly one other element, with all branches leading back to the main predicate (verb).

Here is a tree for the English sentence, “Hopefully an English example will clearly demonstrate the rules of dependency trees”. Don’t worry about every word for now; just try to see the basic structure. Begin reading the tree from the top, where the main verb (PRED) is connected to [ROOT], indicating that all other elements are ultimately connected to it. Attached to the verb are its two arguments, “example” and “rules”, tagged SBJ for “subject” and OBJ for “object”, respectively.
Here is a tree for the English sentence, “Hopefully an English example will clearly demonstrate the rules of dependency trees”. Don’t worry about every word for now; just try to see the basic structure. Begin reading the tree from the top, where the main verb (PRED) is connected to [ROOT], indicating that all other elements are ultimately connected to it. Attached to the verb are its two arguments, “example” and “rules”, tagged SBJ for “subject” and OBJ for “object”, respectively.

Dependency grammar emphasizes the concept of valency (the number of arguments a predicate controls). An argument is any word dependent on the predicate which cannot be omitted without the sentence becoming ungrammatical. For instance, transitive verbs control two arguments, a subject and direct object. Intransitive verbs control only one argument because these verbs cannot take an object.

Theoretically, the proper framework can effectively capture the logic of language. However, I found it wasn’t always easy. Dependency grammar was designed to deal with languages like Latin which have free word order and a complicated case system (a set of endings which are taken by nouns, pronouns and adjectives to indicate their function in the sentence). In Latin, the connections formed by cases, independent of word order, allow for extreme freedom in omitting words while retaining the sentence’s meaning. This can be problematic: if a word depends syntactically on an omitted word, what should we hang it from in our tree?  The missing word can be supplied; but if the language allows for certain words to be omitted, aren’t we failing to totally capture the efficiency of the grammatical structure if we supply the missing words?

Some semantic properties the tree simply cannot capture. For instance, consider the case of the class of words called predicativa, described by Harm Pinkster in Latin Syntax and Semantics. The predicativum is a word which agrees syntactically with an argument of the predicate and provides information about the action of the predicate, e.g. “I stood awestruck”. Here, “awestruck” describes both the person standing and the manner in which he stands. A Latin predicativum is hung from the appropriate argument and tagged Atv. But if the tags were not present, it would be indistinguishable from an ordinary adjective. The dependencies alone fail to capture its relationship with the predicate.

Now consider when the argument with which a predicativum agrees is omitted, as in this sentence of Vergil. Here the goddess Venus is speaking to her son Aeneas upon his arrival at Carthage:

“Quisquis es, haud, credo, invisus caelestibus auras vitalis carpis, Tyriam qui adveneris urbem.”

This roughly translates as: “Whoever you are, I do not believe that you draw breath as one hated by the gods, you who have come to the Tyrian city”. The Atv here is invisus, since it agrees syntactically—in case and number–with the omitted subject of the main verb (carpis), like an adjective, and also provides information about the main verb, like an adverb. Invisus also agrees in case and number with the pronouns quisquis and qui (“whoever” and “you who”, respectively). But neither of those pronouns is an argument of the main verb; rather, they are the subjects of es (“you are”) and adveneris (“you have come”), respectively. So if we hang the Atv, invisus, from either of quisquis or qui in order to capture its case agreement with them, then carpis wouldn’t be the verb nearest to invisus going up the tree, which would be incorrect because that is the verb about which invisus supplies information. The convention is to hang it from the main verb itself and tag it AtvV as opposed to Atv, indicating that it depends on the verb directly, but now the dependencies alone fail to capture its syntactic relationship, and it is indistinguishable from an ordinary adverb! Plus, an adjective has now been hung directly from a verb, which makes no sense syntactically.

Although we seek to formalize it, language and grammar remain in many respects a landscape of mystery within the human mind.

Here is the conventional tree for this sentence, with invisus hung from carpis and tagged AtvV. Invisus is not connected to any argument of the main verb higher up on its branch; the dependencies do not capture its syntactic relationship, and it is indistinguishable from an adverb.
Here is the conventional tree for this sentence, with invisus hung from carpis and tagged AtvV. Invisus is not connected to any argument of the main verb higher up on its branch; the dependencies do not capture its syntactic relationship, and it is indistinguishable from an adverb.

In summary, I encountered many problems in describing formally the complicated expressions of Latin poetry. But I would argue that they were significant, even if they only served to expose the deficiencies of grammatical frameworks. Although we seek to formalize it, language and grammar remain in many respects a landscape of mystery within the human mind.