18513981. SYSTEMS AND METHODS FOR TRAINING LANGUAGE MODELS TO REASON OVER TABLES simplified abstract (Google LLC)

From WikiPatents
Jump to navigation Jump to search

SYSTEMS AND METHODS FOR TRAINING LANGUAGE MODELS TO REASON OVER TABLES

Organization Name

Google LLC

Inventor(s)

[[:Category:Thomas M�ller of Dietikon (CH)|Thomas M�ller of Dietikon (CH)]][[Category:Thomas M�ller of Dietikon (CH)]]

Jonathan Herzig of Tel Aviv (IL)

Pawel Nowak of Zurich (CH)

Julian Eisenschlos of Zurich (CH)

Francesco Piccinno of Zurich (CH)

Syrine Krichene of Zurich (CH)

SYSTEMS AND METHODS FOR TRAINING LANGUAGE MODELS TO REASON OVER TABLES - A simplified explanation of the abstract

This abstract first appeared for US patent application 18513981 titled 'SYSTEMS AND METHODS FOR TRAINING LANGUAGE MODELS TO REASON OVER TABLES

Simplified Explanation

The abstract describes systems and methods for pre-training and fine-tuning neural network-based language models to reason directly over tables without generating logical forms. The language model is pre-trained using masked-language modeling tasks generated from tables and further pre-trained using counterfactual statements and data comparison statements from those tables. Fine-tuning is done using examples with a question, an answer, and a table.

  • Pre-training and fine-tuning of neural network-based language models for reasoning over tables
  • Pre-training using masked-language modeling tasks generated from tables
  • Further pre-training using counterfactual statements and data comparison statements from tables
  • Fine-tuning using examples with a question, an answer, and a table

Potential Applications

This technology could be applied in various fields such as natural language processing, data analysis, and information retrieval.

Problems Solved

This technology solves the problem of enabling neural network-based language models to reason directly over tables without the need for generating logical forms.

Benefits

The benefits of this technology include improved accuracy and efficiency in processing and analyzing tabular data using language models.

Potential Commercial Applications

Potential commercial applications of this technology include data analytics software, chatbots for customer service, and information retrieval systems.

Possible Prior Art

One possible prior art could be the use of neural networks for natural language processing tasks such as text generation and sentiment analysis.

What datasets are used for pre-training the language model in this technology?

The datasets used for pre-training the language model include masked-language modeling tasks generated from tables pulled from a knowledge corpus.

How are the language models fine-tuned in this technology?

The language models are fine-tuned using examples that consist of a question, an answer, and a table, allowing for direct harvesting from existing benchmark datasets or synthetic generation.


Original Abstract Submitted

Systems and methods for pre-training and fine-tuning of neural-network-based language models to reason directly over tables without generating logical forms. In some examples, a language model can be pre-trained using masked-language modeling tasks synthetically generated from tables pulled from a knowledge corpus. In some examples, the language model may be further pre-trained using pairs of counterfactual statements generated from those tables, and/or one or more statements that compare selected data from those tables. The language model may then be fine-tuned using examples that include only a question, an answer, and a table, allowing fine-tuning examples to be harvested directly from existing benchmark datasets or synthetically generated.