Automatic detection of context switch triggers

Patent metadata

Field	Value
Publication number	US-11295080-B2
Application number	US-201916430427-A
Country	US
Kind code	B2
Filing date	Jun 4, 2019
Priority date	Jun 4, 2019
Publication date	Apr 5, 2022
Grant date	Apr 5, 2022

Abstract

A method, system, and computer program product include providing a list of triggers, training the natural language processor with the list of triggers, providing to the natural language processor a text including one trigger, selecting nodes in the text to create an original potential span, predicting whether the original potential span includes another trigger, and adjusting, in response to predicting that the original potential span includes another trigger, the original potential span to exclude the another trigger to create a new potential span.

First claim

What is claimed is:

1. A method of using a natural language processor, the method comprising:
providing a list of hypothetical triggers that indicate hypothetical spans;
providing a list of confirm triggers that indicate factual spans;
training the natural language processor with the list of hypothetical triggers and the list of confirm triggers;
providing to the natural language processor a text including a plurality of confirm triggers, wherein the plurality of confirm triggers includes a first trigger that is not included in the list of confirm triggers;
selecting, iteratively, a plurality of nodes in the text to create a first original potential span, wherein the first original potential span includes one of the plurality of confirm triggers;
predicting, during the one of the iterations that includes the first trigger, whether the first trigger is a confirm trigger that indicates a context switch to a factual span;
adjusting, in response to predicting that the first trigger is a confirm trigger, the original potential span to exclude the first trigger to create a new potential span and a new confirmed span, wherein the new potential span includes some of the selected plurality of nodes and excludes the first trigger, and wherein the new confirmed span includes some of the selected plurality of nodes and the first trigger;
selecting the new confirmed span for retraining of the natural language processor and not selecting the new potential span for retraining of the natural language processor; and
retraining the natural language processor using the new confirmed span that includes the first trigger.

2. The method of claim 1 , wherein the new potential span includes a second trigger, the method further comprising:
predicting whether the second trigger is a hypothetical trigger; and
adjusting, in response to predicting that the second trigger is a hypothetical trigger, the new potential span to exclude the second trigger.

3. The method of claim 2 , wherein the second trigger is not included in the list of hypothetical triggers.

4. The method of claim 1 , wherein predicting whether the first trigger is a confirm trigger comprises:
finding a probability that the first trigger is a confirm trigger; and
comparing the probability to a threshold.

5. The method of claim 4 , wherein finding the probability that the first trigger is a confirm trigger is based on a word/phrase similarity between the first trigger and at least one of the list of confirm triggers.

6. The method of claim 1 , wherein training and retraining the natural language processor involves mapping words/phrases to vectors and analyzing them using machine learning techniques selected from the group consisting of: Word2Vec (two-layer neural net), cosine similarity, and/or GloVe (global vectors).

7. The method of claim 1 , wherein retraining the natural language processor using the new potential span that includes the first trigger comprises:
adding the first trigger to the list of confirm triggers.

8. A system comprising:
a computing processor; and
a memory coupled to the computing processor, wherein the memory comprises instructions which, when executed by the computing processor, specifically configures the computing processor and causes the computing processor to:
provide a list of hypothetical triggers that indicate hypothetical spans;
provide a list of confirm triggers that indicate factual spans;
train a natural language processor with the list of hypothetical triggers and the list of confirm triggers;
provide to the natural language processor a text including a plurality of confirm triggers, wherein the plurality of confirm triggers includes a first trigger that is not included in the list of confirm triggers;
select, iteratively, a plurality of nodes in the text to create a first original potential span, wherein the first original potential span includes one of the plurality of confirm triggers;
predict, during the one of the iterations that includes the first trigger, whether the first trigger is a confirm trigger that indicates a context switch to a factual span;
adjust, in response to predicting that the first trigger is a confirm trigger, the original potential span to exclude the first trigger to create a new potential span and a new confirmed span, wherein the new potential span includes some of the selected plurality of nodes and excludes the first trigger, and wherein the new confirmed span includes some of the selected plurality of nodes and the first trigger;

select the new confirmed span for retraining of the natural language processor and not selecting the new potential span for retraining of the natural language processor; and
retrain the natural language processor using the new confirmed span that includes the first trigger.

9. The system of claim 8 , further comprising:
a hypothetical span analyzer;
a hypothetical dictionary data structure that includes the list of hypothetical triggers; and
a factual dictionary data structure that includes the list of confirm triggers.

10. The system of claim 8 , wherein the memory further comprises instructions which, when executed by the computing processor, specifically configures the computing processor and causes the computing processor to:
predict whether the second trigger is a hypothetical trigger; and
adjust, in response to predicting that the second trigger is a hypothetical trigger, the new potential span to exclude the second trigger.

11. The system of claim 10 , wherein the second trigger is not included in the list of hypothetical triggers.

12. The system of claim 8 , wherein training and retraining the natural language processor involves mapping words/phrases to vectors and analyzing them using machine learning techniques selected from the group consisting of: Word2Vec (two-layer neural net), cosine similarity, and/or GloVe (global vectors).

13. The system of claim 8 , wherein retraining the natural language processor using the new potential span that includes the first trigger comprises:
adding the first trigger to the list of confirm triggers.

14. The system of claim 8 , wherein predicting whether the first trigger is a confirm trigger comprises:
finding a probability that the first trigger is a confirm trigger is based on a word/phrase similarity between the first trigger and at least one of the list of confirm triggers.

15. A computer program product comprising a non-transitory computer-readable storage medium having a computer readable program stored therein to process natural language text, wherein the computer readable program, when executed on a computing device, specifically configures the computing device, and causes the computing device to:
provide a list of hypothetical triggers that indicate hypothetical spans;
provide a list of confirm triggers that indicate factual spans;
train a natural language processor with the list of hypothetical triggers and the list of confirm triggers;
provide to the natural language processor a text including a plurality of confirm triggers, wherein the plurality of confirm triggers includes a first trigger that is not included in the list of confirm triggers;
select, iteratively, a plurality of nodes in the text to create a first original potential span, wherein the first original potential span includes one of the plurality of confirm triggers;
predict, during the one of the iterations that includes the first trigger, whether the first trigger is a confirm trigger that indicates a context switch to a factual span;
adjust, in response to predicting that the first trigger is a confirm trigger

Assignees

• IBM

Inventors

• Allen Corville O
• Delima Roberto
• Contreras David
• Mahajan Krishna

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• G06F40/279 Primary

  Recognition of textual entities · CPC title