{
    "byline": null,
    "dir": null,
    "excerpt": "The B is for bidirectional, and that\u2019s a big deal. It makes it possible to do well on sentence-level (NLI, question answering) and token-level tasks (NER, POS tagging). In a unidirectional model, the word \u201cbank\u201d in a sentence like \u201cI made a bank deposit.\u201d has only \u201cI made a\u201d as its context, keeping useful information from the model.",
    "length": 504,
    "siteName": null,
    "title": "BERT: pre-training of deep bidirectional transformers for language understanding"
}