# US 12,118,295 B2 · Text simplification with minimal hallucination

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,118,295 B2
- Document ID: US12118295B2
- Status: Granted
- Kind: B2
- Application number: 18/045,551
- Filed: 2022-10-11
- Published: 2024-04-11
- Granted: 2024-10-15
- Earliest claimed: 2022-10-11
- Assignees: Adobe Inc.
- Applicants: Adobe Inc.
- Inventors: Vinay Aggarwal; Aparna Garimella; Ananya Ganesh; Niyati Himanshu Chhaya; Nandakishore Kambhatla
- Web page: https://patentgrep.com/patent/US12118295B2

## CPC

- G06F 40/166: Editing, e.g. inserting or deleting
- G06N 3/045: Combinations of networks
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G06F 40/30: Semantic analysis
- G06N 3/08: Learning methods
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Systems and methods for text simplification are described. Embodiments of the present disclosure identify a simplified text that includes original information from a complex text and additional information that is not in the complex text. Embodiments then compute an entailment score for each sentence of the simplified text using a neural network, wherein the entailment score indicates whether the sentence of the simplified text includes information from a sentence of the complex text corresponding to the sentence of the simplified text. Then, embodiments generate a modified text based on the entailment score, the simplified text, and the complex text, wherein the modified text includes the original information and excludes the additional information. Embodiments may then present the modified text to a user via a user interface.

## Claims

### Claim 1

A method for language processing, comprising:
identifying a simplified text that includes original information from a complex text and additional information that is not in the complex text;
computing an entailment score for each sentence of the simplified text using a neural network, wherein the entailment score indicates whether the sentence of the simplified text includes information from a sentence of the complex text corresponding to the sentence of the simplified text;
generating a modified text based on the entailment score, the simplified text, and the complex text, wherein the modified text includes the original information and excludes the additional information;
computing a first similarity score based on the complex text and the simplified text;
computing a second similarity score based on the complex text and the modified text;
comparing the first similarity score to the second similarity score; and
providing the modified text based on the comparison.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
generating the simplified text based on the complex text using a text simplification model.

### Claim 3 (depends on 2)

The method of claim 2, further comprising:
generating an alternative simplified text based on the complex text using an alternative text simplification model; and
generating an alternative modified text based on the alternative simplified text.

### Claim 4 (depends on 1)

The method of claim 1, further comprising:
computing a plurality of entailment scores for the sentence of the simplified text, wherein the plurality of entailment scores correspond to a plurality of sentences from the complex text, respectively;
determining whether any of the plurality of entailment scores exceeds an entailment score threshold; and
determining whether to include the sentence of the simplified text in the modified text based on the score determination.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
computing a first hallucination score based on the complex text and the simplified text;
computing a second hallucination score based on the complex text and the modified text;
comparing the first hallucination score to the second hallucination score; and
providing the modified text as an output based on the comparison.

### Claim 6 (depends on 5)

The method of claim 5, further comprising:
identifying a first set of entities in the complex text;
identifying a second set of entities in the simplified text;
comparing the first set of entities to the second set of entities; and
computing the first hallucination score based on the comparison.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
dividing the complex text into a first set of sentences; and
dividing the simplified text into a second set of sentences, wherein the entailment score is computed based on the first set of sentences and the second set of sentences.

### Claim 8

A method for language processing, comprising:
generating a simplified text based on a complex text;
computing an entailment score for each sentence of the simplified text using a neural network, wherein the entailment score indicates whether the sentence of the simplified text includes information from a sentence of the complex text;
filtering sentences of the simplified text based on the entailment score to obtain a modified text;
computing a first hallucination score based on the complex text and the simplified text;
computing a second hallucination score based on the complex text and the modified text;
comparing the first hallucination score to the second hallucination score; and
providing the modified text as an output based on the comparison.

### Claim 9 (depends on 8)

The method of claim 8, further comprising:
identifying a first set of entities in the complex text, a second set of entities in the simplified text, and a third set of entities in the modified text;
comparing the first set of entities to the second set of entities to compute the first hallucination score; and
comparing the first set of entities to the third set of entities to compute the second hallucination score.

### Claim 10 (depends on 8)

The method of claim 8, further comprising:
generating the simplified text based on the complex text using a text simplification model.

### Claim 11 (depends on 10)

The method of claim 10, further comprising:
generating an alternative simplified text based on the complex text using an alternative text simplification model;
generating an alternative modified text based on the alternative simplified text; and
comparing the modified text and the alternative modified text.

### Claim 12 (depends on 10)

The method of claim 10, further comprising:
receiving a text simplification training dataset, wherein the text simplification training dataset includes a plurality of sentence pairs; and
updating parameters of the text simplification model based on the text simplification training dataset.

### Claim 13 (depends on 8)

The method of claim 8, further comprising:
receiving an entailment training dataset including a first set of training sentences, a second set of training sentences, and ground-truth entailment data, wherein the ground-truth entailment data includes entailment relationships between the first set of training sentences and the second set of training sentences; and
updating parameters of the neural network based on the training dataset.

### Claim 14 (depends on 8)

The method of claim 8, further comprising:
computing a plurality of entailment scores for the sentence of the simplified text, wherein the plurality of entailment scores correspond to a plurality of sentences from the complex text, respectively;
determining whether any of the plurality of entailment scores exceeds an entailment score threshold; and
determining whether to include the sentence of the simplified text in the modified text based on the score determination.

### Claim 15

An apparatus for language processing, comprising:
a processor;
a memory storing instructions executable by the processor;
a neural network comprising parameters stored in the memory and configured to compute an entailment score for each sentence of a simplified text, wherein the entailment score indicates whether the sentence of the simplified text includes information from a sentence of a complex text corresponding to the sentence of the simplified text;
a pruning component configured to generate a modified text based on the entailment score, the simplified text, and the complex text; and
a hallucination scoring component configured to compute a first hallucination score based on the complex text and the simplified text and a second hallucination score based on the complex text and the modified text.

### Claim 16 (depends on 15)

The apparatus of claim 15, further comprising:
a text simplification component configured to generate the simplified text based on the complex text, wherein the text simplification component implements one or more text simplification models.

### Claim 17 (depends on 15)

The apparatus of claim 15, further comprising:
a hallucination scoring component configured to compute a hallucination score between the complex text and the simplified text.

### Claim 18 (depends on 15)

The apparatus of claim 15, further comprising:
a semantic similarity component configured to compute a similarity score between the complex text and the simplified text.

### Claim 19 (depends on 15)

The apparatus of claim 15, further comprising:
a user interface configured to receive the complex text and present the modified text and metrics based on the complex text and the modified text.