TL;DR
This paper introduces the first Universal Dependencies (UD) treebank for Ancash Quechua, a low-resource language. By annotating Ancash Quechua, the authors aim to improve its NLP capabilities, enable cross-linguistic comparisons, and test the adaptability of the UD schema to agglutinative languages with rich morphology.
What Problem Does This Paper Address?
The field of Natural Language Processing (NLP) has historically focused on high-resource languages, leaving many of the world's languages, like Ancash Quechua, with limited digital resources and NLP tools. Ancash Quechua, a variety of the Quechua language family spoken in Peru, is particularly under-resourced, hindering systematic linguistic study and the development of computational tools. Furthermore, agglutinative languages, which build words by adding multiple suffixes, present unique challenges for universal annotation schemes like UD, which are often designed with more isolating languages in mind. This paper addresses these gaps by creating an annotated corpus for Ancash Quechua to foster NLP development and to critically examine the UD framework's applicability to morphologically rich languages.
Methods
The core of this work involves creating a treebank for Ancash Quechua annotated according to the Universal Dependencies (UD) schema. This process involved:
- Data Collection and Digitization: Gathering existing linguistic resources, including bilingual (Quechua-Spanish) chronicles, a novel, a dictionary, and published tales. These were digitized and sentence-aligned.
- Standardization: Standardizing the spelling of the corpora according to the Peruvian Ministry of Education's guidelines and the official Quechua alphabet.
- Annotation Guidelines Development: Adapting the UD schema to Ancash Quechua's specific features. This included defining lexical categories, developing a word segmentation strategy that systematically segments case markers and independent suffixes (labeling them as ADP and PART respectively), and proposing solutions for encoding verbal and derivational morphemes as word features. A novel notation was introduced in the MISC column to track complex morphological decompositions of inflected forms.
- Morphosyntactic Feature Annotation: Developing specific annotation strategies for various features, including nominal suffixes (case markers, derivational suffixes), verbal suffixes (aspect), pronominal core arguments, discourse markers, evidentials, non-verbal predication, negation, and complex clauses (relativization).
Datasets
The paper utilizes several sources for its corpus:
- Cuentos y relatos en el quechua de Huaraz (CRH): Approximately 4427 sentence pairs from autobiographical chronicles and tales.
- Apu Kolki Hirka (AKH): 434 sentences from the first short novel in Ancash Quechua.
- Ancash Quechua dictionary (DIC-MINEDU-2005): 879 entries.
- Cuentos Pintados del Perú project (CP): 300 sentences from traditional tales.
The authors have made an initial corpus of around fifteen sentences freely available and are preparing a larger corpus of over 6,000 sentences for participatory annotation.
Key Insights and Ideas
- UD Schema Adaptability: The paper highlights that while the UD schema is largely applicable, its handling of agglutinative languages with rich morphology requires careful adaptation. Specifically, the decision to segment case markers and independent suffixes as separate tokens (ADP/PART) rather than solely as features is a key divergence from some standard UD practices for less agglutinative languages.
- Morphological Richness: Ancash Quechua's extensive use of suffixes for grammatical functions (case, aspect, evidentiality, discourse marking) necessitates detailed annotation strategies to capture this complexity. The authors' approach of using the MISC column for detailed morphological glossing is crucial for preserving this information.
- Low-Resource Language Empowerment: The creation of this treebank is a significant step towards enabling more systematic linguistic studies and the development of NLP tools for Ancash Quechua, a language that has been largely overlooked in NLP research.
- Cross-Linguistic Comparison Potential: By annotating Ancash Quechua within the UD framework, the paper facilitates direct comparisons with other Quechua varieties and other agglutinative languages, potentially refining linguistic classifications and theories.
Key Results
- The paper presents a preliminary set of annotation guidelines for Ancash Quechua within the UD framework.
- An initial corpus of approximately fifteen sentences, annotated according to these guidelines, is made available.
- The authors have identified specific challenges in applying UD to Ancash Quechua, particularly concerning the segmentation of morphemes and the annotation of evidentiality and discourse markers, and have proposed solutions.
- A larger corpus of over 6,000 sentences is being prepared for annotation and is partially available for participatory annotation.
Limitations and Future Work
- Limited Initial Corpus: The initial annotated corpus is small (around fifteen sentences), serving as a proof of concept.
- Ongoing Annotation: The authors acknowledge that this is a work in progress, with other morphosyntactic features still being studied and annotation guidelines requiring refinement.
- Future Work: The authors plan to continue studying other morphosyntactic features, refine the annotation guidelines, and expand the corpus. They are also preparing to implement a morphological analyzer to automate segmentation and pre-annotation, which will aid in developing NLP tools. The long-term goal is to develop an Ancash Quechua syntactic parser.
Summary updated