| Title | : | Cognitive Parsing: Theory and Applications: DIAMETERS: A Cognitive Parsing based NLP System |
| Speaker | : | C E Veni Madhavan (IISc, & CRRao Institute) |
| Details | : | Mon, 16 Sep, 2019 2:00 PM @ ALC-MR1 |
| Abstract: | : | We propose a cognitive parsing framework for processing natural language, in a manner similar to that of human processing of language. This framework captures the mental heuristics of the language processing apparatus in the human brain together with the features of linguistic nuances. The heuristics are based on several notional tags, which we term as cognitive tags. The embodiments of these tags are in the form of meta-phrases, which we call some- tags. These tags are pointers to the cognitive chunks extracted from the language utterances. We build a computational algorithmic schema for processing the cognitive representations using this system of tags and various statistical, linguistics and NLP techniques. We integrate the syntactic information from POS tags, parse trees and the governor-dependent relationships of typed dependencies with our cognitive tags into holistic representations. We demonstrate the power of our approach, by conguring several major NLP tasks of dialogs, metaphors, expansion, rewriting, translation, summa- rization under an umbrella programme termed DIAMETERS. We show a few example tasks from our ongoing work in the area of processing text from standard corpora for handling tasks such as textual entailment. Thus our ap- proach is quite dierent from the large corpora driven contemporary machine learning approaches. Our approach is an alternative to contemporary, ma- jor NLP systems of articial intelligence based on frameworks of wordtovec (W2V), tensorow (TF), bidirectional encoders, representations and trans- formers (BERT), pre-training (GPT), general adversarial learning (GAN) and tools of PyTorch. |
