# US 2019/0378492 A1 · Kernel-based verbal phrase splitting devices and methods

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

- Number: US 2019/0378492 A1
- Document ID: US20190378492A1
- Status: Published application
- Kind: A1
- Application number: 16/505,485
- Filed: 2019-07-08
- Published: 2019-12-12
- Earliest claimed: 2015-02-26
- Assignees: Nantmobile, LLC
- Applicants: Nantmobile, LLC
- Inventors: Demitrios L. Master; Farzad Ehsani
- Web page: https://patentgrep.com/patent/US20190378492A1

## CPC

- G10L 15/04: Segmentation; Word boundary detection
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue
- G10L 15/197: Probabilistic grammars, e.g. word n-grams
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A device capable of splitting user input into phrases is presented. The disclosed device leverages multiple phrase splitting models to generate one or more possible split locations. The possible split locations can be derived based on leveraging multiple phrase splitting models. Each model contributes its suggested split locations to the set of possible split locations according to an implementation of a phrase splitting kernel algorithm that weights each model's suggestions.

## Claims

### Claim at position 1

1-27. (canceled)

### Claim 28

A computing device comprising:
a memory configured to store a plurality of phrase splitting models wherein said plurality of phrase splitting models comprises at least a first phrase splitting model having a first process to provide a list of possible split locations and a second phrase splitting model having a second process different from the first process to provide a list of possible split locations; and
a processor coupled with the memory and that executes, based on software instructions stored in the memory, operations comprising:
receiving a digital representation of an utterance comprising multiple words;
for each phrase splitting model in the plurality of phrase splitting models, generating a suggested set of split locations between representations of words in the digital representation as a function of a corresponding phrase splitting model in the plurality of phrase splitting models;
compiling a set of phrase split locations according to a phrase splitting kernel algorithm implementation as a function of the suggested set of split locations from each of the phrase splitting models in the plurality of phrase splitting models based, at least in part, on weights associated with each of the phrase splitting models wherein the weights are capable of adjustment;
splitting the digital representation into a set of tokens comprising words according to the set of phrase split locations; and
causing content related to at least some of the tokens in the set of tokens to be rendered on a display.

### Claim 29 (depends on 28)

The device of claim 28, wherein the operations further comprise querying a content database storing the content as a function of the set of tokens.

### Claim 30 (depends on 29)

The device of claim 29, further comprising the content database stored in the memory.

### Claim 31 (depends on 29)

The device of claim 29, wherein the operations further comprise mapping at least some of the tokens to meaning invariant facets.

### Claim 32 (depends on 31)

The device of claim 31, wherein the meaning invariant facets comprises at least one of the following: a brand, a type, a unit of measure, a preparation, an ingredient, a nutritional value, and a health factor.

### Claim 33 (depends on 31)

The device of claim 31, wherein the operations further comprise generating the query according to hierarchical rules based on the meaning invariant facets.

### Claim 34 (depends on 31)

The device of claim 31, wherein the content is indexed in the content database according to the meaning invariant facets.

### Claim 35 (depends on 28)

The device of claim 28, wherein the plurality of phrase splitting models comprises domain-specific models with respect to a specific domain.

### Claim 36 (depends on 35)

The device of claim 35, wherein the specific domain comprises a food domain.

### Claim 37 (depends on 36)

The device of claim 36, wherein the content comprises digital content associated with the food domain.

### Claim 38 (depends on 37)

The device of claim 37, wherein the content comprises digital content associated with at least one of the following food-related meaning invariant facets: a brand, a type, a nutrition, a preparation, an ingredient, a food, and a unit.

### Claim 39 (depends on 28)

The device of claim 28, wherein the content comprises at least one of the following: object information, an image, a video, a metric, a classification, audio, a coupon, a web site, a news article, a study, and a promotion.

### Claim 40 (depends on 28)

The device of claim 28, wherein the digital representation of the utterance comprises an audio format.

### Claim 41 (depends on 28)

The device of claim 28, wherein the digital representation of the utterance comprises a text representation of the utterance.

### Claim 42 (depends on 28)

The device of claim 28, wherein the digital representation of the utterance comprises a tokenized representation of the utterance.

### Claim 43 (depends on 42)

The device of claim 42, wherein the tokenized representation comprises a normalized representation.

### Claim 44 (depends on 28)

The device of claim 28, wherein at least one of the phrase splitting models comprises a mechanical turk phrase splitting model.

### Claim 45 (depends on 28)

The device of claim 28, wherein at least one of the phrase splitting models comprises a phrase chunking model.

### Claim 46 (depends on 28)

The device of claim 28, wherein at least one of the phrase splitting models comprises a domain-specific token matching model.

### Claim 47 (depends on 28)

The device of claim 28, wherein at least one of the phrase splitting models comprises a split-word matching model.

### Claim 48 (depends on 28)

The device of claim 28, wherein at least one of the phrase splitting models comprises a consecutive word entropy model.

### Claim 49 (depends on 28)

The device of claim 28, wherein at least one of the phrase splitting models comprises an exception model.

### Claim 50 (depends on 28)

The device of claim 28, wherein at least one of the suggested set of split locations from the phrase splitting models in the plurality of phrase splitting models comprises a NULL set.

### Claim 51 (depends on 28)

The device of claim 28, wherein the weights comprises at least one of a frequency and an entropy measure.

### Claim 52 (depends on 28)

The device of claim 28, further comprising a mobile device housing the processor and the memory, the housing further comprising a display.

### Claim 53 (depends on 52)

The device of claim 52, wherein the mobile device comprises at least one of the following: a smart phone, a tablet, and a phablet.

### Claim 54 (depends on 28)

The device of claim 28 wherein the weights are determined based, at least in part, on empirical validation.

### Claim 55 (depends on 28)

The device of claim 28 wherein the weights are adjusted, at least in part, as a function of at least one of the following: user interaction, user input and determination by at least one Mechanical Turk.

### Claim 56

A computer-based language recognition method using at least one processor coupled with a memory storing software instructions, the method comprising:
receiving a digital representation of an utterance comprising multiple words;
generating a suggested set of split locations between representations of words in the digital representation as a function of at least a first phrase splitting model and a second phrase splitting model;
compiling a set of phrase split locations according to a phrase splitting kernel algorithm implementation as a function of the suggested set of split locations from each of the first phrase splitting model and the second phrase splitting model based, at least in part, on adjustable weighted sums associated with each of the first phrase splitting model and the second phrase splitting model;
splitting the digital representation into a set of tokens comprising words according to the set of phrase split locations; and
causing content related to at least some of the tokens in the set of tokens to be rendered on a display.

### Claim 57

A non-transitory computer readable medium comprising instructions executable by a computer processor to execute processing comprising:
receiving a digital representation of an utterance comprising multiple words;
generating a suggested set of split locations between representations of words in the digital representation as a function of at least a first phrase splitting model and a second phrase splitting model;
compiling a set of phrase split locations according to a phrase splitting kernel algorithm implementation as a function of the suggested set of split locations from each of the first phrase splitting model and the second phrase splitting model based, at least in part, on adjustable weighted sums associated with each of the first phrase splitting model and the second phrase splitting model;
splitting the digital representation into a set of tokens comprising words according to the set of phrase split locations; and
causing content related to at least some of the tokens in the set of tokens to be rendered on a display.
