Corpus linguistics is the study of languages on the basis of a collection, or corpus, of texts. Classical linguistic analysis supported its theories with selective examples. In the days before computers, it was almost impossible to present an exhaustive survey of a phenomenon in a body of texts and quantify the distribution of alternative constuctions. Compiling a word-level concordance of a body of text (such as the Hebrew Old Testament) was the work of a lifetime, and exhaustively exploring phenomena such as co-occurrence of words or syntactical constructions was only a dream.
The advent of digital computers and high-capacity storage has made that dream possible. For more than 30 years, Francis Andersen (a Semitic philologist) and Dean Forbes (a physicist) have pursued it together in the domain of biblical Hebrew. By the early 1980s, they had encoded the surface form of the text and begun morphological analysis. This volume is a report of the syntactic annotation that they have applied to the text. Their encoded text is broadly available through a commercial product, the Logos Bible software, and one motive for this volume is to document the coding offered in the Logos product.
Their account gives a vivid picture of the exploratory nature of such an undertaking. They frankly describe early decisions that were motivated by the limitations of computers available three decades ago, initial coding conventions that have proven inadequate and are being revised, and grammatical uncertainties that they are still in the process of resolving. They frequently remind the reader of inconsistencies in the encoding that they are discovering, promise updated versions of the database, and look forward to another level of encoding that they hope to add at the level of discourse structures (to be accompanied by a successor to this volume).
Morphological and semantic encodings of texts for machine processing are fairly straightforward because one can simply attach codes to each segment of text (informally, each word). Syntactic coding, the focus of this volume, is more complicated, since it involves capturing the relations among text segments. The authors’ approach is to represent syntactic structures as directed graphs (in most cases, trees), in which each node above the level of a segment is composed of one or more lower-level segments. Each nonleaf node is annotated with its grammatical function or semantic role and a licensing relation justifying its composition from lower-level nodes. The ambiguity between grammatical function and semantic role is one of the inconsistencies that the authors have targeted for future revision. The resulting structures are presented as tree-like structures, accounting for the word “visualized” in the book’s title.
The book focuses at the clause level (roughly, the unit of predication), and the kinds of constituents that can make up a clause (clause immediate constituents, or CICs). The first three chapters introduce the book’s objectives, define basic terminology, and outline the classification used for individual segments. Chapter 4 introduces the structural language for describing syntax (“phrase markers”), and chapters 5 and 6 introduce basic and more complex phrases. Chapters 7 and 8 discuss main and embedded clauses. All of this exposition is preparatory for the discussion of CICs in chapters 9 through 11.
A fundamental hypothesis of the book is that the relative prominence of different CICs in clauses with verbs depends on the verb itself. Chapters 12 through 15 present CIC censuses for four common Hebrew verbs. Then, chapter 16 reviews the incidence of different CICs across the entire corpus. Chapter 17, supported with technical detail in appendix 5, explores the hypothesis that the distributional difference between CIC types for two verbs (measured using the Aitchison metric) reflects their semantic proximity, suggesting that one might be able to define semantic categories automatically using clustering. The hypothesis is intriguing, and supported by other work in computational linguistics, but the relatively small size of the Biblical Hebrew corpus (about 300,000 segments) means that only a few verbs are common enough to yield reliable statistics.
Chapters 18 through 20 discuss complicating factors, including quasiverbals, verbless clauses, and phrase markers that deviate from the strict conditions for a tree. Chapter 21 anticipates the next level of encoding, at the discourse level. Seven appendices provide technical details on different philological, linguistic, and mathematical issues. A helpful glossary gathers together the various specialized terms that the authors use to describe their approach, followed by a full bibliography and indices sorted by author, Hebrew sentence analyzed, and topics.
Traditionally, scientists present the results of their research after stripping away the scaffolding and passing over the false starts. This volume, by contrast, presents a refreshing and energizing view of science in the making. It contains many new contributions of great scientific value in their own right, but in addition it helps the reader appreciate the history of those contributions, and shows the continuity of that effort reaching toward even higher-level insights.
