LESA: Linguistic Encapsulation and Semantic Amalgamation Based Generalised Claim Detection from Online Content

The conceptualization of a claim lies at the core of argument mining. The\nsegregation of claims is complex, owing to the divergence in textual syntax and\ncontext across different distributions. Another pressing issue is the\nunavailability of labeled unstructured text for experimentation. In this paper,\nwe propose LESA, a framework which aims at advancing headfirst into expunging\nthe former issue by assembling a source-independent generalized model that\ncaptures syntactic features through part-of-speech and dependency embeddings,\nas well as contextual features through a fine-tuned language model. We resolve\nthe latter issue by annotating a Twitter dataset which aims at providing a\ntesting ground on a large unstructured dataset. Experimental results show that\nLESA improves upon the state-of-the-art performance across six benchmark claim\ndatasets by an average of 3 claim-F1 points for in-domain experiments and by 2\nclaim-F1 points for general-domain experiments. On our dataset too, LESA\noutperforms existing baselines by 1 claim-F1 point on the in-domain experiments\nand 2 claim-F1 points on the general-domain experiments. We also release\ncomprehensive data annotation guidelines compiled during the annotation phase\n(which was missing in the current literature).\n