TL;DR
This paper introduces VeLePa, a comprehensive verbal lexicon for Central Pame, a morphologically complex indigenous language. By providing a detailed, computationally usable resource, it aims to advance both theoretical linguistics and natural language processing (NLP) for underresourced languages, highlighting the extreme nature of human morphological diversity.
What Problem Does This Paper Address?
The paper addresses the critical need for computational resources for underresourced languages, particularly those with complex morphological systems. Existing documentation for Central Pame is insufficient, undigitized, and largely unusable for computational tasks. Furthermore, languages like Central Pame, with their high levels of irregularity and intricate systems of inflection distributed across prefixes, stems, tone, and suffixes, present significant challenges for both linguistic theory and NLP applications like reinflection tasks. Such languages are crucial for building more representative models of human language diversity.
Methods
The authors built VeLePa through extensive fieldwork and elicitation with native speakers over four years. They adopted a phonemic orthography to accurately represent linguistic features like vowel nasality, tone, and consonant length, avoiding the limitations of Spanish-based orthographies. The lexicon contains 12,528 inflected forms for 216 verbs, with each form independently elicited and checked for accuracy. The paper also utilizes computational tools like Qumín and Principal Parts Analyzer to analyze the morphological complexity, calculating measures such as conditional entropy and identifying principal parts needed to predict paradigms.
Datasets
The primary dataset introduced is VeLePa, a verbal lexicon of Central Pame. It contains 12,528 inflected forms for 216 verbs, along with their lemmas and morphosyntactic values. Use frequencies for lemmas and morphosyntactic values are also provided, derived from existing Central Pame texts and supplemented with native speaker estimates. This dataset is made freely available online.
Key Insights and Ideas
- Extreme Morphological Complexity: Central Pame exhibits an exceptionally complex inflectional system, exceeding the complexity of even well-studied languages like Latin in its individual subsystems.
- Distributed Exponence: Morphosyntactic information is spread across multiple layers (prefixes, stems, tone, suffixes), each with its own set of irregularities and inflection classes, making the system highly challenging.
- High Irregularity and Suppletion: The language features a high degree of irregularity, with many unique verb behaviors and significant suppletion (using different roots for related forms).
- Unpredictability of Syncretism: Syncretism (identical forms for different meanings) occurs frequently but not systematically, adding another layer of difficulty for prediction.
- Interdependence of Layers: While each inflectional layer is complex, preliminary analysis suggests significant informativeness between layers, which may aid speakers in learning and using the system.
Key Results
- VeLePa contains 12,528 inflected forms for 216 Central Pame verbs.
- The language's inflectional subsystems (prefixes, stems, tone, suffixes) are analyzed, revealing varying degrees of complexity. For instance, tone exhibits the highest average conditional entropy (1.01 bits) and requires the most principal parts (7) for prediction.
- Prefixal inflection, while complex, is found to be lower in complexity than other layers, requiring 5 principal parts for paradigm prediction.
- The analysis of stem morphology reveals that most verbs (96.3%) do not display stem alternation, with the primary action occurring on consonantal onsets.
- Suffixes primarily encode person-number and are non-syllabic, attaching as codas or modifying stem endings.
Limitations and Future Work
- Data Sparsity: The available corpus of Central Pame texts is small (1171 verbal tokens), necessitating supplementation with subjective frequency estimates from native speakers.
- Orthographic Challenges: While a phonemic orthography was adopted, the traditional orthography used in the community has limitations in representing features like tone and vowel nasalization.
- Future Work: The authors suggest that VeLePa can now be used to explore the Paradigm Cell Filling Problem and related challenges in more depth. It also opens avenues for incorporating such typologically diverse languages into NLP tasks like reinflection, making NLP models more representative of global linguistic diversity. The paper also highlights the potential for exploring the interplay between different inflectional layers to understand how speakers manage such complex systems.
Summary updated