Chris Alberti

TL;DR: The authors presented a text-to-text (seq2seq) paradigm to predict mentions and links jointly, which achieved state-of-the-art performance on the CoNLL-2012 datasets.

TL;DR: The authors proposed two approaches for generating large parallel datasets for grammatical error correction using publicly available Wikipedia data, one method extracts source-target pairs from Wikipedia edit histories with minimal filtration heuristics, while the second method introduces noise into Wikipedia sentences via round-trip translation through bridge languages.

TL;DR: The EfficientQA competition as discussed by the authors focused on open-domain question answering (QA), where systems take natural language questions as input and return natural language answers, and the aim of the competition was to build systems that can predict correct answers while also satisfying strict on-disk memory budgets.

TL;DR: The authors derive example-level scores on large pretraining data based on a smaller, higher-quality dataset and use delta-log-perplexity, a type of example scoring, into a training schedule for grammatical error correction.

Abstract: Systems and methods for identifying competitors using content extensions in content items in content items associated with their content placement campaigns are described. A processor identifies one or more competing entities from auctions in which the first entity places a bid. The competing entities are associated with content items having a first type of content extension that received impressions in at least one identified auction. The processor computes an overlap rate based on a number of auctions in which both a content item having the first type of content extension of the identified competing entity and a content item of the first entity received impressions and a number of auctions in which a content item having the first type of content extension of the identified competing entity received an impression and the first entity competed. The processor ranks the competing entities based on the computed overlap rate.