Computer-Assisted Morphological Analysis of Ancient Greek

David W. PackardCOLING 1973

TL;DR

This paper introduces a computer-assisted system for the morphological analysis of Ancient Greek, aiming to accelerate the learning process for students by providing automated lexical and grammatical analysis of texts. The system combines computational analysis with human verification, offering a more efficient way to process and understand complex Greek morphology.

What Problem Does This Paper Address?

Traditional methods for teaching Ancient Greek often require students to spend a significant amount of time (at least a year) on grammar before they can engage with substantial literary texts. This paper addresses the challenge of this steep learning curve by proposing a computer-assisted approach to morphological analysis. The authors' conviction is that by focusing initial instruction on the grammatical features that actually appear in early readings, and by leveraging computational tools to handle the complexity of Greek morphology, students can begin reading literature much earlier in their first year. The core problem is the sheer complexity of Ancient Greek morphology, which acts as a major obstacle to learning the language, and the paper aims to alleviate this by automating a significant portion of the analysis.

Methods

The system employs a combination of computer analysis and subsequent editorial verification. The core program analyzes Greek words in the order they appear in the text, without requiring prior editing. It first checks if a word is in an "indeclinable list" of approximately 800 entries, which includes prepositions, adverbs, particles, and highly irregular forms. If a word is not in this list, the program attempts to segment it into a stem and an ending based on Greek morphology rules.

A key aspect of the method is handling phonological changes that obscure the juncture between stem and ending, such as vowel contractions in Attic Greek. The system treats these contracted forms as having separate conjugations and endings, creating "pseudo-stems" for analytical purposes. For nouns of the third declension, where sound changes can be extensive, the approach prioritizes expediency by placing problematic forms in the indeclinable list rather than attempting complex reconstruction.

The analysis also accounts for verb morphology, including the six principal parts and the use of prefixes and augments. The system attempts to remove prepositional prefixes and then analyze the remainder. It also identifies verbal augments (like the initial 'e-' for imperfect and aorist tenses) and can handle elided final short vowels by hypothetically restoring them. Ambiguity is handled by presenting the most likely analysis with a warning of potential error, allowing editors to correct mistakes.

Datasets

The paper mentions analyzing 40,000 words of Ancient Greek selected from texts suitable for first-year students. Specific texts analyzed include Plato's Apology, Crito, and Lysis, Xenophon's Constitution of the Spartans, Euripides' Medea, and selections from Thucydides and the New Testament. The tables and dictionary used by the program were developed entirely from these texts.

Key Insights and Ideas

  • Pragmatic Approach to Learning: The paper's fundamental insight is that a practical, text-driven approach to morphology, supported by computational tools, can significantly accelerate language acquisition.
  • "Pseudo-Stems" for Efficiency: The use of "pseudo-stems" to handle phonological changes (like vowel contractions) is a clever pragmatic solution that simplifies the analytical process, even if it deviates from strict linguistic reconstruction.
  • Indecclinable List as a Catch-All: The "indecclinable list" serves as a crucial component, efficiently handling a large percentage of words (around 50%) that are either function words or highly irregular, thus streamlining the analysis of more complex forms.
  • Ambiguity Management: The strategy of providing the most likely analysis with a "doubtful" flag, rather than overwhelming the user with all possibilities, is a practical way to manage the inherent ambiguities in language analysis.
  • Textbook Design Implications: The statistical analysis of word frequencies and morphological features in texts like Plato's Apology directly informs the design of a more effective textbook, prioritizing common structures earlier.

Key Results

  • The system can analyze approximately 2000 words per second.
  • The computer charges for analyzing Plato's Apology were under one dollar.
  • Manual editing to check analyzed text requires about one hour per 500 words, or roughly four hours of human time for every second of machine analysis. This highlights that the cost of editing is significantly higher than the initial computer analysis.
  • The developed tables appear to cover a high proportion of forms needed for analyzing a variety of additional texts, suggesting good generalizability.
  • The system's dictionary can accommodate 4000 stems, 2000 endings, and 1000 indeclinables within 300K bytes of memory.

Limitations and Future Work

  • Ambiguity: While the system provides the most likely analysis, ambiguity remains a challenge. The authors acknowledge that some ambiguities, particularly between verbal and nominal forms, are resolved by preferring the nominal form, and others (like gender or voice) are marked as doubtful for editor correction.
  • Dialectal Variation: The paper notes that the existence of multiple literary dialects with varying morphology (e.g., Attic, Doric, Ionic, Homeric) presents a problem, requiring separate tables and dictionaries for different texts. The authors are currently working on tables for Homeric Greek.
  • Dictionary Size Limitations: The current implementation has a limitation on dictionary size (4000 stems). While this is sufficient for many texts, expanding to a wider range of texts will eventually necessitate exceeding this limit. For machines with limited memory, processing the text multiple times with different dictionary subsets might be necessary.
  • Future Work: The authors express a desire to expand the system to include dictionaries for a wider range of texts. They also mention that the program could be modified to operate on other systems with relative ease. The ongoing improvement of the dictionary and program is expected to reduce the time required for correcting computer analysis in the future.

Summary updated

BibTeX
@inproceedings{packard1973computer,
  author = {David W. Packard},
  title = {{Computer-Assisted Morphological Analysis of Ancient Greek}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1973},
}
Figure