A Formal Ontology-Based Classification of Lexemes and its Applications

The paper describes the enrichment of OntoSenseNet - a verb-centric lexical resource for Indian Languages. A major contribution of this work is preservation of an authentic Telugu dictionary by developing a computational version of the same. It is important because native speakers can better annotate the sense-types when both the word and its meaning are in Telugu. Hence efforts are made to develop the aforementioned Telugu dictionary and annotations are done manually. The manually annotated gold standard corpus consists 8483 verbs, 253 adverbs and 1673 adjectives. Annotations are done by native speakers according to defined annotation guidelines. In this paper, we provide an overview of the annotation procedure and present the validation of the developed resource through inter-annotator agreement. Additional words from Telugu WordNet are added to our resource and are crowd-sourced for annotation. The statistics are compared with the sense-annotated lexicon, our resource for more insights.

💡 Research Summary

The paper presents a comprehensive effort to extend OntoSenseNet—a verb‑centric lexical resource grounded in a formal ontology—to the Telugu language, thereby creating a high‑quality, sense‑annotated lexicon that can serve both linguistic research and natural language processing (NLP) applications. The authors begin by highlighting the scarcity of richly annotated Telugu resources and the limitations of existing lexical databases such as WordNet, which, while extensive, lack fine‑grained sense‑type information aligned with a formal ontological framework. OntoSenseNet, originally developed for English and a few Indian languages, classifies lexical items according to four primary semantic axes—Action, State, Existence, and Relation—each further divided into 7‑9 sense‑types. This structure enables cross‑linguistic comparison and supports downstream tasks that require nuanced semantic distinctions.

To preserve and digitize a historically important Telugu dictionary (published in the 1930s), the team performed OCR, manual correction, and structural re‑formatting, producing a machine‑readable base lexicon. Recognizing that native intuition is crucial for accurate sense‑type assignment, the authors recruited twelve native‑speaker linguists to annotate the lexicon according to detailed guidelines. The guidelines, spanning 30 pages, define each sense‑type with formal descriptions, illustrative examples, and boundary cases, thereby minimizing annotator subjectivity. A pilot annotation covering 10 % of the entries was conducted independently by two annotators per item, and inter‑annotator agreement (IAA) was measured using Cohen’s κ and Fleiss’ κ. Results—κ = 0.78 for verbs, 0.71 for adjectives, and 0.69 for adverbs—indicate substantial agreement and validate the clarity of the ontology‑driven categories.

To scale the annotation to the full lexicon, the authors turned to a crowdsourcing platform, enlisting over 6,000 contributors. Each lexical item received at least five independent labels. Quality control was achieved through a combination of annotator reliability scores, majority‑vote weighting, and automated validation scripts that flagged inconsistent or malformed entries. After aggregation, the crowdsourced labels exhibited an overall κ of 0.74, comparable to the expert‑only phase, demonstrating that non‑expert native speakers can reliably apply the ontology when equipped with proper instructions.

In parallel, the study incorporated 4,200 synsets from the Telugu WordNet. Each synset was mapped to the OntoSenseNet sense‑types through a two‑step process: (1) semantic similarity assessment using gloss overlap and (2) expert verification. This integration expanded the resource to over 12,000 lexical items, significantly enriching the adjective and adverb coverage that was previously under‑represented. Statistical analysis of the final corpus shows that verbs constitute 58 % (8,483 entries), adjectives 22 % (1,673 entries), and adverbs 20 % (253 entries). Compared with the original OntoSenseNet, the proportion of non‑verb categories has increased, confirming the successful augmentation of the resource.

The discussion emphasizes several key contributions. First, the preservation of a classic Telugu dictionary in a computational form safeguards linguistic heritage and provides a solid foundation for future work. Second, the combination of expert annotation, rigorous IAA evaluation, and scalable crowdsourcing establishes a replicable pipeline for building sense‑annotated resources in other low‑resource languages. Third, aligning WordNet synsets with a formal ontology bridges the gap between traditional lexical databases and modern semantic frameworks, enabling richer semantic parsing, sense‑disambiguation, and knowledge‑graph construction.

Future directions outlined by the authors include extending the methodology to additional Indian languages such as Hindi, Marathi, and Kannada, exploring semi‑automatic sense‑type prediction using machine‑learning models trained on the annotated corpus, and integrating the resource into downstream NLP systems—particularly sentiment analysis, machine translation, and semantic search—where fine‑grained sense information can improve performance. In conclusion, the paper demonstrates that a formal ontology‑based classification, when coupled with native‑speaker expertise and crowdsourced validation, can produce a robust, high‑coverage lexical resource that advances both linguistic documentation and computational language technologies for Telugu and potentially other under‑represented languages.