DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
All objections/rejections not mentioned in this Office Action have been withdrawn by the Examiner.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 15 December 2025 has been entered.
Response to Amendments
Applicant’s amendment filed on 15 December 2025 has been entered.
In view of the amendment to the claim(s), the amendment of claim(s) 20, 29, and 38 have been acknowledged and entered.
In view of the amendment to claim(s) 20, 29, and 38, the rejection of claims 20-39 under 35 U.S.C. §103 is/are withdrawn.
In light of the amended claims, new grounds for rejection under 35 U.S.C. §103 are provided in the action below.
Response to Arguments
Applicant’s arguments regarding the prior art rejections under 35 U.S.C. §103, see pages 7-10 of the Response to Final Office Action dated 01 October 2025, which was received on 15 December 2025 (hereinafter Response and Office Action, respectively), have been fully considered.
With respect to the rejection(s) of claim(s) 20, 29, and 38 under 35 U.S.C. §103 in light of Lee (U.S. Pat. App. Pub. No. 2016/0180742, hereinafter Lee) in view of Asthana (U.S. Pat. App. Pub. No. 2019/0370385, hereinafter Asthana), applicant asserts (1) Lee does not disclose or suggest a model trained on “the first level entities,” and (2) that Lee and Asthana fail to teach or suggest “identifying composite entities using a second machine learning model trained using numeric vectors including the POS information and the first level entities.” The arguments are addressed individually below.
Regarding the first argument, though applicant concludes that “the 'texts having grammatical errors' thus cannot be the claimed 'first level entities',” it is unclear how this conclusion is being drawn. In support of this contention, applicant asserts that “the 'texts having grammatical errors' should have pattern information” and that “the ‘texts having grammatical errors’ thus cannot be the claimed ‘first level entities’, as ‘patterns representing a structure of the input text' are the identified composite entities, as alleged.” However, applicant’s assertion regarding the “texts having grammatical errors,” even if assumed, arguendo, to be true, does not create the asserted mutual exclusivity with “the ‘patterns representing a structure of the input text' are the identified composite entities.”
During patent examination, the pending claims must be “given their broadest reasonable interpretation (BRI) consistent with the specification.” (MPEP 2111). In determining the BRI, “the meaning given to a claim term must be consistent with the ordinary and customary meaning of the term (unless the term has been given a special definition in the specification), and must be consistent with the use of the claim term in the specification and drawings.” (Id.) First, we look to the specification for special definitions of the word “pattern.” As the word pattern has not been explicitly redefined, examiner turns to the ordinary and customary meaning of the term. As such, the broadest reasonable interpretation of the word “pattern” is a regularly occurring, repeating, or orderly sequence in nature, data, or physical systems that reveals underlying rules or relationships.
In light of the BRI, it is noted that whatever patterns exist within a data set, necessarily exist before their discovery or elucidation. The pattern itself, as distinguished from the inference made by a machine or human with relation to the pattern, is a mathematical relationship that exists regardless whether it is observed or not. Thus, whatever pattern information is available in a set of data is present in the data from the moment the data is generated. The prior existence of “pattern information” which is understood as information about a pattern, does not preclude later observance of “patterns representing a structure of the input text.” Even so, “pattern information” is not a claim limitation in the claims as currently amended. As well, the concept of “pattern information” is not tantamount to “patterns representing a structure of the input text.” As a result, applicant's assertion that the “text having grammatical errors” should have “pattern information,” and downstream events argued to exist as a result, cannot be properly applied to other limitations recited in the claim.
With regards to the claim limitation “create and learn linguistics patterns,” the understood meaning of this phrase in the context of the claim is as follows. It is noted that this limitation was not present in the claims as filed, either in the instant application or in the parent application, App. No. 16/721,452. As such, it is not entitled to the same level of deference as originally filed claims. Though the phrase “create and learn linguistic patterns” is presented in the specification twice, it is presented in different contexts than the claim limitations (i.e., the second machine learning model is trained “using numeric vectors” but the equivalent “memory based entity recognizer module 214” learns from “base entities 445” which is “meaningful text with a clear identification of entities in the text” received after “the output of a machine learning model (which would be in the form of a vector)” is reconverted by the “devectorizer 444”) and with no further explanation. (See Instant Application at [0053]). In light of the ordinary and customary meaning of the phrase and the available usage in the specification, “learn[ing] linguistic patterns” is understood as linguistic patterns which are elucidated in the received data. However, these direct recitations fail to further clarify the intended meaning of “create… linguistic patterns.” As such, we look to the usage in the remaining portions of the specification and the descriptions of the claim part to which the phrase applies.
At other portions of the specification, such as [0026], the memory based entity recognizer module (i.e., the “second machine learning model”) is described as learning and storing the linguistic patterns. More generally, the “memory learning based entity recognizer 446” is described as “Memory Based Learning (MBL),” which is distinguished from “Statistical Machine Learning (SML)” of the “machine learning (ML) based entity recognizer 441”, where the “corpus builder 218 uses memory based learning to add new entities to the corpus over time with additional training.” (See Instant Application at [0030], [0049], [0053], and [0059]), As such, the phrase “create… linguistic patterns” is understood as describing the storage of the learned linguistic patterns (e.g., the storage of “new entities” in “the corpus”). (Id.)
Regarding the second argument, applicant’s arguments in light of the amended claims are persuasive. Therefore, the rejection of claims 20, 29, and 38 is withdrawn.
Applicant further argues that the rejection(s) of dependent claims 21-28, 30-37, and 39 should be withdrawn for at least the same reasons as independent claims 20, 29, and 38. Applicant’s arguments in light of the amended claims are not persuasive for at least the same reasons described above with relation to claims 20, 29, and 38. As such, the rejections of claims 21-28, 30-37, and 39 under 35 U.S.C. §103 is withdrawn.
However, upon further consideration, new ground(s) of rejection under 35 U.S.C. §103 are made in light of combinations of Lee, Asthana, Xu (U.S. Pat. App. Pub. No. 2018/0365211, hereinafter Xu), and Mathias (U.S. Pat. App. Pub. No. 2017/0278514, hereinafter Mathias), and newly cited reference Almosallam (U.S. Pat. App. Pub. No. 2017/0235721, hereinafter Almosallam).
The Applicant has not provided any further statement and therefore, the Examiner directs the Applicant to the below rationale.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 20-39 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Regarding claims 20, and mutatis mutandis claims 29 and 38, the specification does not describe the “second machine learning model” as “using numeric vectors.” First, we look to the cited specification support, at paragraphs [0025], and [0028]-[0029] provided in the Response. However, the cited paragraphs fail to provide clear support for the amendments to the claims 20, 29, and 38. Of note, the phrases “first machine learning model”, “second machine learning model”, etc. do not occur in the specification as filed. As such, we look to function and the associated claim part to determine the described parts which correspond with the newly numerically described machine learning models. In this case, the claim limitation provides clear guidance to the corresponding part in the specification. As recited in amended claim 1, “the second machine learning model is used to create and learn linguistics patterns from the first level entities.” The specification at para. [0053] explains that “The present system 200 uses memory based learning to create and learn computational linguistics based patterns.” (emphasis added). Further disclosed in the specification at paragraph [0026], the “memory based entity recognizer module 214 has the capability to learn linguistic patterns and store the linguistic patterns… in memory for future entity recognition processes.” As such, the “second machine learning model” is understood to correspond to the memory learning based entity recognizer (e.g., “memory based entity recognizer module 214,” memory based entity recognizer 332,” the “memory learning based entity recognizer 446”, and “memory learning based entity recognizer 565”), referred to herein as “memory recognizer,” for brevity and clarity.
Regarding paragraph [0025], this paragraph is directed to the rules based entity recognizer, not the memory learning based entity recognizer 213. As the rules based entity recognizer 213 is not described in the specification as performing the functions of the “second machine learning model” as described in claims 20, 29, and 38, this claim part is not considered relevant to specification support for the claim amendments directed to the memory recognizer. Paragraphs [0028] and [0029] are directed to the vectorizer 215 and the vectorization of output from the rules based entity recognizer. However, the vectors produced by the vectorizer 215 are not described as being received by or otherwise used by the memory recognizer.
With regards to the memory recognizer, the memory recognizer is not described as receiving numeric vectors for any purpose, training or otherwise. As explained at [0026], “Memory based entity recognizer 214 uses the first level entities,” which are identified by “the rules based entity recognizer 213 and machine learning based entity recognizer 220,” to “recognize the composite entities that include a base entity with a linguistic pattern.” In turn, the identified first level entities are a result of the de-vectorization process and are not described as vectors of any kind. Regarding the use of vectors in the overall process, as described in the specification of the Instant Application, the vectorizer “preserves the POS information of individual words in a sentence, the number of occurrences of any POS in the sentence, and the entity information recognized by the rules based entity recognizer 213.” (See Instant Application at [0028], [0052]). The numerical vectors in this context are understood as a result of vectorization of the input text, and incorporating the results of the rules based entity recognizer, which is then received by the “machine learning model”. (See Instant Application at [0028], [0052]-[0053]). Then, devectorizer receives the output from the machine learning model and “performs the opposite process of the vectorizer” to “reconvert the output of a machine learning model (which would be in the form of a vector) to meaningful text with a clear identification of entities in the text (e.g. names of persons, organizations, locations, expressions of times, quantities, etc.),” where the “identified entities,” understood as the identified first level entities, and identified entities are fed “into memory learning based entity recognizer 446 to recognize composite entities.” (See Instant Application at [0030], [0053]). As such, the specification fails to provide clear support for training of the memory recognizer using numerical vectors.
Of note, applicant has taken special care in the specification to distinguish “machine learning” and “memory learning.” (See Instant Application at [0026], [0039], [0053], [0055]-[0056], and [0059]). As such, though referred to in claims 20, 29, and 38 as “second machine learning model,” the memory recognizer (i.e., the “second machine learning model”) is understood as distinct from the “machine learning model” as described in specification at least at paragraph [0053], not least of which because there is a separate explicit reference to the “memory learning based entity recognizer 446,” receiving the devectorized results.
Therefore, claims 20, 29, and 38 contain subject matter which fails for lack of specification support and the claims are rejected.
Regarding claims 21-28, 30-37, and 39, claims 21-28, 30-37, and 39 depend from claims 20, 29, and 38 and incorporate all limitations therefrom. As such, claims 21-28, 30-37, and 39 are rejected for at least the same reasons as described with reference to claims 20, 29, and 38.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim 20-22, 24-27, 29-31, 33-36, and 38-39 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee in view of Asthana.
Regarding claim 20, Lee discloses A computer-implemented method comprising (Systems and methods described with reference to preposition error correction; Lee, ¶ [0098]): receiving an input sentence (The system receives input text, such as from a user, which “may include any type of text or document” such as an input sentence; Lee, ¶ [0102]); tagging parts of speech (POS) of the input sentence (“text normalization part 110 may normalize the input text by tagging words constituting the input text based on part-of-speech information {tagging parts of speech (POS)} of the words {of the input sentence}.”; Lee, ¶ [0103]); identifying first level entities using a first... model and POS information (“The text normalization part 110 may further include a time normalization module 111 and a place normalization module 113”, each of which, in combination or independently, are the first model, where the “time normalization module 111 may substitute words having temporal meaning in the tagged input text {using... POS information} with time-type information (i.e., time-type tags) based on a pre-constructed text dictionary” and the “place normalization module 113 may substitute words having place implications in the tagged input text with place-type information (i.e., place-type tags) based on named entity recognition.” wherein the word having temporal meaning or place implication {first level entities} is identified “based on a pre-constructed text dictionary” and/or “named entity recognition” {using a first... model}”; Lee, ¶ [0105]-[0109]); and identifying composite entities using a second machine learning model (“The pattern extraction part 120 may extract patterns representing a structure of the input text with reference to prepositions included in the normalized input text” where the pattern extraction part 120 determines “A plurality of patterns representing a structure of the input text may be extracted” based on the words having temporal meaning or place implication “prior to or subsequent to a preposition included in the normalized input text”, where the words having temporal meaning and/or place implication and the associated preposition(s), which is also referred to in Lee as “patterns representing a structure of the input text” are the identified composite entities; Lee, ¶ [0111]-[0113]) trained using... the POS information and the first level entities, (“A plurality of patterns extracted based on the plurality of word sequences may be pre-constructed as an error pattern database 130 through verification... by comparing a pre-constructed grammatical error corpus and the plurality of patterns, and the pattern verified as having preposition errors may be recorded into the error pattern database 130,” which is machine learning in the context of the present application (e.g., “memory based learning”); Lee, ¶ [0114]) wherein: the second machine learning model is used to create and learn linguistics patterns from the first level entities (“the reason of the verification is for recording only valid patterns having preposition errors into the error pattern database 130 among a large number of patterns extracted by using the word sequences” where the extraction of “a large number of patterns” related to “preposition errors” is learning linguistic patterns from the first level entities, and the “recording only valid patterns having preposition errors” selected from the extracted patterns is the creating of linguistics patterns from the first level entities; Lee, ¶ [0115]), each composite entity includes a first level entity with a linguistic pattern (the words having temporal meaning and/or place implication {first level entity} and the associated preposition(s), which is also referred to in Lee as “patterns representing a structure of the input text” are the identified composite entities, and grammatical errors and the associated grammatical correction are linguistic patterns which are associated with the words having temporal meaning or place implication.; Lee, ¶ [0048]-[0049], [0042]-[0043]), and the identified composite entities include a first composite entity and a second composite entity that have a common first level entity but different linguistic patterns (Discloses generating “A plurality of patterns extracted based on the plurality of word sequences” based on “words having temporal meaning or place implications” associated with “a preposition” being “located before or after a noun or a pronoun” which, using a generated example for purposes of clarification in the form of <preposition><noun><preposition>, such as “On the river in Austin”, the normalized text of “on river” {a first composite entity} and “river in” {a second composite entity} are detected, based on being “located before or after a noun”, which share a common first level entity but include different linguistic patterns.; Lee, ¶ [0111], [0114], [0092]-[0095]; FIG. 5). However, Lee fails to expressly recite identifying first level entities using a first machine learning model and the second machine learning model trained using numeric vectors including the POS information and the first level entities.
Asthana teaches “a domain-specific type system for a machine learning model.” (Asthana, ¶ [0001]). Regarding claim 20, Asthana teaches identifying first level entities using a first machine learning model and the POS information (Discloses an entity type identifier as part of a machine learning model 101, where “The entity type identifier 204 is configured to perform a cluster analysis on each conceptual text to identify potential entity types”; Asthana, ¶ [0031]-[0032]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction system of Lee to incorporate the teachings of Asthana to include identifying first level entities using a first machine learning model. The generation system of Asthana “the type system to be formed is based on the document corpus in a specific domain,” and “thus the type system is relevant to the specific domain” which avoids “multiple iterations of generating and refining the type system to adapt to the specific domain,” which increases ease and convenience of the developer as well as expediting the availability of the ML model, as recognized by Asthana. (Asthana, ¶ [0004], [0062]). However, Lee and Asthana fail to expressly recite the second machine learning model trained using numeric vectors including the POS information and the first level entities.
Almosallam teaches “automatically detecting semantic errors in a text using an ANN trained on linguistic properties of words in the text.” (Almosallam, ¶ [0001]). Regarding claim 20, Almosallam teaches trained using numeric vectors including the POS information and the first level entities, (Discloses an “ANN model 110” which is “trained by passing samples generated from concatenating a window of features centered on a word in question to learn the mapping to a corresponding label” where, in one example, “consider a sentence like ‘In case of fire use the rear exist’ generates a learning sample for the word “case” by concatenating the features of the words “<S>“, “In”, “case”, “of” and “fire” next to each other in that order and the corresponding label is set to 1 when a window size is set to 5,” where the example includes both POS information and first level entities, and the input is based on “an N-gram feature vector for each word in the text”; Almosallam, ¶ [0033], [0043]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction systems of Lee, as modified by the domain-specific type systems of Asthana, to incorporate the teachings of Almosallam to include trained using numeric vectors including the POS information and the first level entities. Almosallam discloses an “improved method and system for automatically detecting semantic errors in a text” which “evaluates the semantic validity of a word given its surrounding context,” which overcomes the problems of dictionary or corpus based systems related to large set sizes and different languages, thus improving contextual correction of text and proper attribution of named entities, as recognized by Almosallam. (Almosallam, ¶ [0003], [0005]-[0007]).
Regarding claim 21, the rejection of claim 20 is incorporated. Lee and Asthana disclose all of the elements of the current invention as stated above. Lee further discloses wherein the first level entities include one or more of Company, Name, Currency, City, Social Security Number, State, E-Mail Address, Product, Contact, and Postal Index Number (Pin) Code (Discloses the word having temporal meaning or place implication as being determined based on “named entity recognition” and including “a word corresponding to one of a person, a location, and an organization”; Lee, ¶ [0042], [0047]).
Regarding claim 22, the rejection of claim 20 is incorporated. Lee and Asthana disclose all of the elements of the current invention as stated above. Lee further discloses wherein the composite entities include one or more of To Date, From Date, Promo Amount, and Payment Account (Discloses detecting prepositions in conjunction with words having temporal meaning and/or place implication, where the words having temporal meaning including the types of “<DATE>“ and “<MONTH>“ and where to and from are both prepositions.; Lee, ¶ [0042], [0044]).
Regarding claim 24, the rejection of claim 20 is incorporated. Lee and Asthana disclose all of the elements of the current invention as stated above. Lee further discloses wherein identifying the composite entities uses one or more of memory based learning, computational linguistics, and custom rules (the system matches “between the error patterns included in the pre-constructed error pattern database 130 and the extracted patterns” using “machine learning” trained based on “texts having grammatical errors” where “patterns having a preposition error can be extracted among the plurality of patterns through... machine-learning on the plurality of patterns.”; Lee, ¶ [0058]-[0059], [0064], [0097]).
Regarding claim 25, the rejection of claim 20 is incorporated. Lee and Asthana disclose all of the elements of the current invention as stated above. Lee further discloses further comprising storing one or more of the learned linguistic patterns, information of the first level entities, keywords, and relative proximity information of one of the keywords to one of the first level entities (Discloses that “the pattern verified as having preposition errors may be recorded into the error pattern database 130.”; Lee, ¶ [0114]).
Regarding claim 26, the rejection of claim 20 is incorporated. Lee and Asthana disclose all of the elements of the current invention as stated above. However, Lee fails to expressly recite further comprising training the first and second machine learning models based on domain specific knowledge.
The relevance of Asthana is described above with relation to claim 20. Regarding claim 26, Asthana teaches further comprising training the first and second machine learning models based on domain specific knowledge (“document corpus 103, related to a particular domain, is uploaded into the machine learning model 101 by a facility user (for example, an annotation process manager), so that the generation system 102 can generate a domain specific type system based on the document corpus 103.” where each of the “entity type identifier 204... and relation type identifier 207” are incorporated into the machine learning model 101; Asthana, ¶ [0031]-[0032]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction system of Lee to incorporate the teachings of Asthana to include further comprising training the first and second machine learning models based on domain specific knowledge. The generation system of Asthana “the type system to be formed is based on the document corpus in a specific domain,” and “thus the type system is relevant to the specific domain” which avoids “multiple iterations of generating and refining the type system to adapt to the specific domain,” which increases ease and convenience of the developer as well as expediting the availability of the ML model, as recognized by Asthana. (Asthana, ¶ [0004], [0062]).
Regarding claim 27, the rejection of claim 20 is incorporated. Lee and Asthana disclose all of the elements of the current invention as stated above. However, Lee fails to expressly recite further comprising training the first and second machine learning models based on feedback systems.
The relevance of Asthana is described above with relation to claim 20. Regarding claim 27, Asthana teaches further comprising training the first and second machine learning models based on feedback systems (“the machine learning model user can revise or update any entity types and relation types generated by the generation system” and the “machine learning model user can adjust the hierarchy of entity types and relation types formed by the generation system” using the revisions or updates.; Asthana, ¶ [0049]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction system of Lee to incorporate the teachings of Asthana to include further comprising training the first and second machine learning models based on feedback systems. The generation system of Asthana “the type system to be formed is based on the document corpus in a specific domain,” and “thus the type system is relevant to the specific domain” which avoids “multiple iterations of generating and refining the type system to adapt to the specific domain,” which increases ease and convenience of the developer as well as expediting the availability of the ML model, as recognized by Asthana. (Asthana, ¶ [0004], [0062]).
Regarding claim 29, Lee discloses A system comprising (Systems and methods described with reference to preposition error correction; Lee, ¶ [0098]): a processor; a memory in communication with the processor (discloses “a user terminal” for implementing the systems and methods, examples of which include “a smartphone, a tablet PC, a Personal Digital Assistant (PDA), a laptop computer, or a computer” where all of said devices includes a processor and memory in communication with said processor; Lee, ¶ [0101]) and comprising instructions which, when executed by the processor, program the processor to (Discloses “a method for correcting preposition errors, performed in an information processing apparatus” which is understood as software, as stored in memory, comprising instructions to cause the processor to perform the functions described.; Lee, ¶ [0037], [0101]): receive an input sentence (The system receives input text, such as from a user, which “may include any type of text or document” such as an input sentence; Lee, ¶ [0102]); tag parts of speech (POS) of the input sentence (“text normalization part 110 may normalize the input text by tagging words constituting the input text based on part-of-speech information {tagging parts of speech (POS)} of the words {of the input sentence}.”; Lee, ¶ [0103]); identify first level entities using a first... model and POS information (“The text normalization part 110 may further include a time normalization module 111 and a place normalization module 113”, each of which, in combination or independently, are the first model, where the “time normalization module 111 may substitute words having temporal meaning in the tagged input text {using... POS information} with time-type information (i.e., time-type tags) based on a pre-constructed text dictionary” and the “place normalization module 113 may substitute words having place implications in the tagged input text with place-type information (i.e., place-type tags) based on named entity recognition.” wherein the word having temporal meaning or place implication {first level entities} is identified “based on a pre-constructed text dictionary” and/or “named entity recognition” {using a first... model}”; Lee, ¶ [0105]-[0109]); and identifying composite entities using a second machine learning model (“The pattern extraction part 120 may extract patterns representing a structure of the input text with reference to prepositions included in the normalized input text” where the pattern extraction part 120 determines “A plurality of patterns representing a structure of the input text may be extracted” based on the words having temporal meaning or place implication “prior to or subsequent to a preposition included in the normalized input text”, where the words having temporal meaning and/or place implication and the associated preposition(s), which is also referred to in Lee as “patterns representing a structure of the input text” are the identified composite entities; Lee, ¶ [0111]-[0113]) trained using... the POS information and the first level entities, (“A plurality of patterns extracted based on the plurality of word sequences may be pre-constructed as an error pattern database 130 through verification... by comparing a pre-constructed grammatical error corpus and the plurality of patterns, and the pattern verified as having preposition errors may be recorded into the error pattern database 130,” which is machine learning in the context of the present application (e.g., “memory based learning”); Lee, ¶ [0114]), wherein: the second machine learning model is used to create and learn linguistics patterns from the first level entities (“the reason of the verification is for recording only valid patterns having preposition errors into the error pattern database 130 among a large number of patterns extracted by using the word sequences” where the extraction of “a large number of patterns” related to “preposition errors” is learning linguistic patterns from the first level entities, and the “recording only valid patterns having preposition errors” selected from the extracted patterns is the creating of linguistics patterns from the first level entities; Lee, ¶ [0115]), each composite entity includes a first level entity with a linguistic pattern (the words having temporal meaning and/or place implication {first level entity} and the associated preposition(s), which is also referred to in Lee as “patterns representing a structure of the input text” are the identified composite entities, and grammatical errors and the associated grammatical correction are linguistic patterns which are associated with the words having temporal meaning or place implication.; Lee, ¶ [0048]-[0049], [0042]-[0043]), and the identified composite entities include a first composite entity and a second composite entity that have a common first level entity but different linguistic patterns (Discloses generating “A plurality of patterns extracted based on the plurality of word sequences” based on “words having temporal meaning or place implications” associated with “a preposition” being “located before or after a noun or a pronoun” which, using a generated example for purposes of clarification in the form of <preposition><noun><preposition>, such as “On the river in Austin”, the normalized text of “on river” {a first composite entity} and “river in” {a second composite entity} are detected, based on being “located before or after a noun”, which share a common first level entity but include different linguistic patterns.; Lee, ¶ [0111], [0114], [0092]-[0095]; FIG. 5). However, Lee fails to expressly recite identify first level entities using a first machine learning model and POS information, and the second machine learning model trained using numeric vectors including the POS information and the first level entities.
The relevance of Asthana is described above with relation to claim 20. Regarding claim 29, Asthana teaches identify first level entities using a first machine learning model and POS information (Discloses an entity type identifier as part of a machine learning model 101, where “The entity type identifier 204 is configured to perform a cluster analysis on each conceptual text to identify potential entity types”; Asthana, ¶ [0031]-[0032]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction system of Lee to incorporate the teachings of Asthana to include identify first level entities using a first machine learning model and POS information. The generation system of Asthana “the type system to be formed is based on the document corpus in a specific domain,” and “thus the type system is relevant to the specific domain” which avoids “multiple iterations of generating and refining the type system to adapt to the specific domain,” which increases ease and convenience of the developer as well as expediting the availability of the ML model, as recognized by Asthana. (Asthana, ¶ [0004], [0062]). However, Lee and Asthana fail to expressly recite the second machine learning model trained using numeric vectors including the POS information and the first level entities.
The relevance of Almosallam is described above with relation to claim 20. Regarding claim 29, Almosallam teaches trained using numeric vectors including the POS information and the first level entities, (Discloses an “ANN model 110” which is “trained by passing samples generated from concatenating a window of features centered on a word in question to learn the mapping to a corresponding label” where, in one example, “consider a sentence like ‘In case of fire use the rear exist’ generates a learning sample for the word “case” by concatenating the features of the words “<S>“, “In”, “case”, “of” and “fire” next to each other in that order and the corresponding label is set to 1 when a window size is set to 5,” where the example includes both POS information and first level entities, and the input is based on “an N-gram feature vector for each word in the text”; Almosallam, ¶ [0033], [0043]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction systems of Lee, as modified by the domain-specific type systems of Asthana, to incorporate the teachings of Almosallam to include trained using numeric vectors including the POS information and the first level entities. Almosallam discloses an “improved method and system for automatically detecting semantic errors in a text” which “evaluates the semantic validity of a word given its surrounding context,” which overcomes the problems of dictionary or corpus based systems related to large set sizes and different languages, thus improving contextual correction of text and proper attribution of named entities, as recognized by Almosallam. (Almosallam, ¶ [0003], [0005]-[0007]).
Regarding claim 30, the rejection of claim 29 is incorporated. Claim 30 is substantially the same as claim 21 and is therefore rejected under the same rationale as above.
Regarding claim 31, the rejection of claim 29 is incorporated. Claim 31 is substantially the same as claim 22 and is therefore rejected under the same rationale as above.
Regarding claim 33, the rejection of claim 29 is incorporated. Claim 33 is substantially the same as claim 24 and is therefore rejected under the same rationale as above.
Regarding claim 34, the rejection of claim 29 is incorporated. Claim 34 is substantially the same as claim 25 and is therefore rejected under the same rationale as above.
Regarding claim 35, the rejection of claim 29 is incorporated. Claim 35 is substantially the same as claim 26 and is therefore rejected under the same rationale as above.
Regarding claim 36, the rejection of claim 29 is incorporated. Claim 36 is substantially the same as claim 27 and is therefore rejected under the same rationale as above.
Regarding claim 38, Lee discloses A computer program product comprising a non-transitory computer-readable medium having computer readable program code stored thereon, the computer readable program code configured to (Systems and methods described with reference to preposition error correction, as implemented through “a user terminal” for implementing the systems and methods, examples of which include “a smartphone, a tablet PC, a Personal Digital Assistant (PDA), a laptop computer, or a computer” where all of said devices includes a processor and memory in communication with said processor; Lee, ¶ [0098], [0101]): receive an input sentence (The system receives input text, such as from a user, which “may include any type of text or document” such as an input sentence; Lee, ¶ [0102]); tag parts of speech (POS) of the input sentence (“text normalization part 110 may normalize the input text by tagging words constituting the input text based on part-of-speech information {tagging parts of speech (POS)} of the words {of the input sentence}.”; Lee, ¶ [0103]); identify first level entities using a first... model and POS information (“The text normalization part 110 may further include a time normalization module 111 and a place normalization module 113”, each of which, in combination or independently, are the first model, where the “time normalization module 111 may substitute words having temporal meaning in the tagged input text {using... POS information} with time-type information (i.e., time-type tags) based on a pre-constructed text dictionary” and the “place normalization module 113 may substitute words having place implications in the tagged input text with place-type information (i.e., place-type tags) based on named entity recognition.” wherein the word having temporal meaning or place implication {first level entities} is identified “based on a pre-constructed text dictionary” and/or “named entity recognition” {using a first... model}”; Lee, ¶ [0105]-[0109]); and identifying composite entities using a second machine learning model (“The pattern extraction part 120 may extract patterns representing a structure of the input text with reference to prepositions included in the normalized input text” where the pattern extraction part 120 determines “A plurality of patterns representing a structure of the input text may be extracted” based on the words having temporal meaning or place implication “prior to or subsequent to a preposition included in the normalized input text”, where the words having temporal meaning and/or place implication and the associated preposition(s), which is also referred to in Lee as “patterns representing a structure of the input text” are the identified composite entities; Lee, ¶ [0111]-[0113]) trained using... the POS information and the first level entities, (“A plurality of patterns extracted based on the plurality of word sequences may be pre-constructed as an error pattern database 130 through verification... by comparing a pre-constructed grammatical error corpus and the plurality of patterns, and the pattern verified as having preposition errors may be recorded into the error pattern database 130,” which is machine learning in the context of the present application (e.g., “memory based learning”); Lee, ¶ [0114]), wherein: the second machine learning model is used to create and learn linguistics patterns from the first level entities of a same type (“the reason of the verification is for recording only valid patterns having preposition errors into the error pattern database 130 among a large number of patterns extracted by using the word sequences” where the extraction of “a large number of patterns” related to “preposition errors” is learning linguistic patterns from the first level entities, and the “recording only valid patterns having preposition errors” selected from the extracted patterns is the creating of linguistics patterns from the first level entities, and where each of the entities described in the patterns is an example of a first level entity in the “tagged input text” of a same type (e.g., time-type or place-type).; Lee, ¶ [0115], [0042]), each composite entity includes a first level entity with a linguistic pattern (the words having temporal meaning and/or place implication {first level entity} and the associated preposition(s), which is also referred to in Lee as “patterns representing a structure of the input text” are the identified composite entities, and grammatical errors and the associated grammatical correction are linguistic patterns which are associated with the words having temporal meaning or place implication.; Lee, ¶ [0048]-[0049], [0042]-[0043]), and the identified composite entities include a first composite entity and a second composite entity that have a common first level entity but different linguistic patterns (Discloses generating “A plurality of patterns extracted based on the plurality of word sequences” based on “words having temporal meaning or place implications” associated with “a preposition” being “located before or after a noun or a pronoun” which, using a generated example for purposes of clarification in the form of <preposition><noun><preposition>, such as “On the river in Austin”, the normalized text of “on river” {a first composite entity} and “river in” {a second composite entity} are detected, based on being “located before or after a noun”, which share a common first level entity but include different linguistic patterns.; Lee, ¶ [0111], [0114], [0092]-[0095]; FIG. 5).. However, Lee fails to expressly recite identify first level entities using a first machine learning model and POS information.
The relevance of Asthana is described above with relation to claim 20. Regarding claim 38, Asthana teaches identify first level entities using a first machine learning model and POS information (Discloses an entity type identifier as part of a machine learning model 101, where “The entity type identifier 204 is configured to perform a cluster analysis on each conceptual text to identify potential entity types”; Asthana, ¶ [0031]-[0032]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction system of Lee to incorporate the teachings of Asthana to include identify first level entities using a first machine learning model and POS information. The generation system of Asthana “the type system to be formed is based on the document corpus in a specific domain,” and “thus the type system is relevant to the specific domain” which avoids “multiple iterations of generating and refining the type system to adapt to the specific domain,” which increases ease and convenience of the developer as well as expediting the availability of the ML model, as recognized by Asthana. (Asthana, ¶ [0004], [0062]). However, Lee and Asthana fail to expressly recite the second machine learning model trained using numeric vectors including the POS information and the first level entities.
The relevance of Almosallam is described above with relation to claim 20. Regarding claim 38, Almosallam teaches trained using numeric vectors including the POS information and the first level entities, (Discloses an “ANN model 110” which is “trained by passing samples generated from concatenating a window of features centered on a word in question to learn the mapping to a corresponding label” where, in one example, “consider a sentence like ‘In case of fire use the rear exist’ generates a learning sample for the word “case” by concatenating the features of the words “<S>“, “In”, “case”, “of” and “fire” next to each other in that order and the corresponding label is set to 1 when a window size is set to 5,” where the example includes both POS information and first level entities, and the input is based on “an N-gram feature vector for each word in the text”; Almosallam, ¶ [0033], [0043]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction systems of Lee, as modified by the domain-specific type systems of Asthana, to incorporate the teachings of Almosallam to include trained using numeric vectors including the POS information and the first level entities. Almosallam discloses an “improved method and system for automatically detecting semantic errors in a text” which “evaluates the semantic validity of a word given its surrounding context,” which overcomes the problems of dictionary or corpus based systems related to large set sizes and different languages, thus improving contextual correction of text and proper attribution of named entities, as recognized by Almosallam. (Almosallam, ¶ [0003], [0005]-[0007]).
Regarding claim 39, the rejection of claim 38 is incorporated. Claim 39 is substantially the same as claim 26 and is therefore rejected under the same rationale as above.
Claims 23 and 32 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee, Asthana, and Almosallam as applied to claims 20 and 29 above, and further in view of Xu (U.S. Pat. App. Pub. No. 2018/0365211, hereinafter Xu).
Regarding claim 23, the rejection of claim 20 is incorporated. Lee, Asthana, and Almosallam disclose all of the elements of the current invention as stated above. Lee further discloses wherein tagging the POS of the input sentence comprises creating …[tags] including at least one or more of nouns, pronouns, adverbs, verbs, and help verbs (Discloses part of speech tagging as including “part-of-speech information such as “personal pronoun (PP)+verb (VB)+ at +definite article (DA)+noun (NN)”; Lee, ¶ [0041]). However, Lee, Asthana, and Almosallam fail to expressly creating a tag set.
Xu teaches “a method and device for recognizing a domain named entity.” (Xu, ¶ [0002]). Regarding claim 23, Xu teaches wherein tagging the POS of the input sentence comprises creating a tag set including at least one or more of nouns, pronouns, adverbs, verbs, and help verbs (“tagging may be conducted on each segmented word in the text to be recognized according to the tag set of the domain corresponding to the text” thus the tag set comprises tags for each segmented word in the text; Xu, ¶ [0035]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction systems of Lee, as modified by the domain-specific type systems of Asthana, and as modified by the neural network based semantic error detection of Almosallam, to incorporate the teachings of Xu to include wherein tagging the POS of the input sentence comprises creating a tag set including at least one or more of nouns, pronouns, adverbs, verbs, and help verbs. The disclosed named entity recognition systems of Xu “accurately locate the boundary of the named entity by using the method of tagging, which effectively reduces the influence of a word segmentation result on the recognition of the domain named entity, and improves the accuracy of named entity recognition,” as recognized by Xu. (Xu, ¶ [0006]).
Regarding claim 32, the rejection of claim 29 is incorporated. Claim 32 is substantially the same as claim 23 and is therefore rejected under the same rationale as above.
Claims 28 and 37 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee, Asthana, and Almosallam as applied to claims 20 and 29 above, and further in view of Mathias (U.S. Pat. App. Pub. No. 2017/0278514, hereinafter Mathias).
Regarding claim 28, the rejection of claim 20 is incorporated. Lee, Asthana, and Almosallam disclose all of the elements of the current invention as stated above. However, Lee fails to expressly recite further comprising training the first machine learning model, wherein the training comprises: preprocessing the input sentence to generate a cleaned input sentence, and converting and translating the cleaned input sentence into numeric vectors to identify the first level entities.
Mathias teaches systems and methods for entity recognition and resolution. (Mathias, ¶ [0025]). Regarding claim 28, Mathias teaches further comprising training the first machine learning model, (“Further, as training data is added to, or otherwise changed, new classifiers/models may be trained to update the classifiers/models as desired.”; Mathias, ¶ [0103]) wherein the training comprises: preprocessing the input sentence to generate a cleaned input sentence (“The lexical analyzer component 502 may receive input text (such as output from an ASR component 250) and may parse and tag the text according to its parts of speech (e.g., identify subject, object, verb, preposition, etc.)” where parsing of the text is preprocessing of the input to generate a cleaned input sentence.; Mathias, ¶ [0067]), and converting and translating the cleaned input sentence into numeric vectors to identify the first level entities (“Both the named entity type and command models are hierarchical models that produce a coarse to fine-grained classification. Each can be trained either as a log-linear model or a support vector machine (or other classifier)” where the system “converts the input spoken form text to a representation {numeric vectors} that can be... used to train the NLU models”; Mathias, ¶ [0067]-[0068]).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the error correction systems of Lee, as modified by the domain-specific type systems of Asthana, and as modified by the neural network based semantic error detection of Almosallam, to incorporate the teachings of Mathias to include further comprising training the first machine learning model, wherein the training comprises: preprocessing the input sentence to generate a cleaned input sentence, and converting and translating the cleaned input sentence into numeric vectors to identify the first level entities. The system of Mathias incorporates training for one or more machine learning model(s) which receive “the input text and traverse[s] the hierarchy of entity types… to arrive at the highest scoring entity type” where “that entity type” is applied “to refer to a list of entities in the knowledge base for ultimate entity resolution,” which avoids the problems of “the parallel multi-domain approach” of prior art systems, including significant waste of “computing resources at runtime, in addition to the resources expended configuring the system to operate across multiple domains,” as recognized by Mathias. (Mathias, ¶ [0024]-[0025]).
Regarding claim 37, the rejection of claim 29 is incorporated. Claim 37 is substantially the same as claim 28 and is therefore rejected under the same rationale as above.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Asthana et al. (U.S. Pat. App. Pub. No. 2020/0160231) discloses a system and method for identifying corpus complexity attributes for the document corpus and annotator qualification attributes for each candidate annotator for matching analysis.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Sean E. Serraguard whose telephone number is (313)446-6627. The examiner can normally be reached 07:00-17:00 M-F.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Daniel C. Washburn can be reached at (571) 272-5551. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Sean E Serraguard/Patent Examiner, Art Unit 2657