Constructional Potentiality: Priscianic grammar as a disambiguation technique in the automatic recognition of Latin syntax

R. R. DyerCOLING 1969

TL;DR

This paper proposes a novel approach to automatically parsing Latin syntax by leveraging Priscian's constructional grammar. By treating Latin's rich inflectional system not as a source of ambiguity but as a set of "constructional potentialities," the authors aim to overcome the limitations of traditional word-order-based parsing methods for highly inflected languages.

What Problem Does This Paper Address?

Traditional natural language processing (NLP) techniques often rely heavily on word order to determine syntactic relationships within a sentence. This works well for languages like English, where word order is relatively fixed. However, highly inflected languages, such as Latin, present a significant challenge. In Latin, word order can be very flexible, with grammatical roles primarily indicated by inflectional endings (e.g., case, number, gender). While this inflectional richness theoretically makes grammatical roles explicit, it also introduces a high degree of ambiguity. Native speakers resolve this ambiguity by combining their understanding of semantic meaning with the learned patterns of inflectional habits. This paper addresses the problem of how to automatically parse Latin syntax in a way that effectively handles this complex interplay of inflection and flexible word order, moving beyond simple word-order-based methods.

Methods

The paper introduces an approach based on "constructional syntax," drawing from the work of the ancient grammarian Priscian. The core idea is to view each word not in isolation but as a member of at least one two-member construction. These constructions represent grammatical relationships, such as verb agreement with a subject or a verb governing an object.

The key innovation lies in the concept of "constructional potentiality." For the first word of a sentence, its inflectional morphemes are analyzed to determine all possible syntactic roles it could play. This "potentiality" is then used in a parsing program. The program iteratively assigns construction modes to possible constructional pairs, working towards a surface-level structure that forms a well-formed "A-structure" (a type of syntax tree). This approach aims to disambiguate by considering the inherent grammatical possibilities encoded in the inflections, rather than solely relying on linear word order.

Datasets

Not discussed in the paper. The paper mentions the existence of "rather large parsing dictionaries" developed by institutes involved in automatic parsing of Latin, but does not detail their contents or any specific datasets used for training or evaluation of the proposed method.

Key Insights and Ideas

  • Inflection as a Feature, Not Just Noise: The paper reframes Latin's complex inflectional system from a source of ambiguity to a rich set of signals that define "constructional potentialities." This is a departure from methods that might struggle with the flexibility it introduces.
  • Constructional Grammar for Highly Inflected Languages: The application of constructional grammar, inspired by Priscian, offers a promising framework for languages where word order is less dominant than morphology.
  • Deep Structure Representation: The proposed deep structure of a Latin sentence is conceptualized as a left side (presentences, labels, topics) and a right side (moving towards the speaker's comment), forming a syntax tree through constructional membership.
  • Probability in Parsing: The paper acknowledges that the "probability factor" for constructional potentiality differs depending on a word's position in the sentence, suggesting a nuanced approach to parsing.

Key Results

The paper describes the approach and its theoretical underpinnings rather than presenting specific quantitative results from an implemented parser. It outlines how a parsing program would assign construction modes and arrive at a well-formed A-structure. The authors suggest that by incorporating the constructional potentiality of basic verbs into the dictionary, one can assign values and semantic features to words involved in transitive and bi-transitive constructions.

Limitations and Future Work

  • Ambiguity in Morphemes: The authors acknowledge that despite the explicitness of inflectional morphemes, there is still a "high degree of ambiguity" within them, and native speakers rely on semantic understanding.
  • Parser Output: The paper notes that existing parsers (from the institutes mentioned) do not, to their knowledge, produce tree structures or deep-structure representations. This implies that the proposed constructional approach aims to fill this gap.
  • Future Work: The paper implicitly suggests future work in developing and evaluating a parser that fully implements this constructional potentiality framework. It also hints at extending the list of constructions to include relationships like the genitive and ablative.

Summary updated

BibTeX
@inproceedings{dyer1969constructional,
  author = {R. R. Dyer},
  title = {{Constructional Potentiality: Priscianic grammar as a disambiguation technique in the automatic recognition of Latin syntax}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1969},
}
Figure