DETAILED ACTION
Status of Application
This action is a Non-Final Rejection. This action is in response to the application filed on April 19, 2024.
Claims 1-20 are pending and rejected.
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . 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 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.
Claim Rejections - 35 USC § 112(b)
The following is a quotation 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, or for pre-AIA the applicant regards as the invention.
Claim 1 recites “tokens” in the first step. Claim 5 recites “b) converting words in the natural language text into tokens and inserting the tokens into a token vector.” It is unclear whether the antecedent basis for “the tokens” in claim 5 is found in claim 1 or in claim 5. Claims 7, 8, 13-15, and 18-20 have a similar issue.
Claim Rejections - 35 USC § 101
35 U.S.C. § 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. § 101 as being directed to non-statutory subject matter because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Does the Claim Fall within a Statutory Category? (see MPEP 2106.03)
Yes, with respect to claims 1-8, which recite a method and, therefore, are directed to the statutory class of process.
Yes, with respect to claims 9-15, which recite a system that comprises “at least one processor” and, therefore, are directed to the statutory class of machine or manufacture.
No, with respect to claims 16-20, which recite a transformer that has an encoder, a decoder, and an attention layer. The Specification does not define the transformer as hardware. Instead, the transformer appears to be a model, which is data. Therefore, claims 16-20 are directed to software per se. Because software is not a statutory category, these claims are ineligible. To overcome this rejection, the claims should positively recite hardware. Although these claims are ineligible at step 1, they are being analyzed below with respect to the other steps.
Step 2A, Prong One: Is a Judicial Exception Recited? (see MPEP 2106.04(a))
The following claims (Claims 1-8 are representative) identify the limitations that recite the abstract idea in regular text and that recite additional elements in bold:
1. A computer-implemented method for identifying a semantic frame of a target word in a natural language text, comprising:
receiving, into a transformer, a token vector, wherein the token vector contains tokens representing words in a natural language input text;
generating one or more potential substitute words for the target word;
generating, for each potential substitute word, a paraphrased text, that is a paraphrase of the input text with each potential substitute word;
comparing each paraphrased text to the input text to determine whether the potential substitute word is a valid substitute word;
identifying one or more valid substitute words for the target word, from the one or more potential substitute words.
2. The method of claim 1, wherein the generated potential substitute words are ordered by an output score.
3. The method of claim 2, wherein the identified valid substitute words are the top k valid substitutes.
4. The method of claim 1 further comprising:
identifying the semantic frame most in common among the valid substitute words.
5. The method of claim 1 further comprising:
before receiving, into a transformer as input, a token vector:
a) receiving, as input, a natural language text;
b) converting words in the natural language text into tokens and inserting the tokens into a token vector.
6. The method of claim 5, wherein converting words in the natural language text into tokens includes populating, with the value of zero, any tokens in the vector that do not correspond to a word.
7. The method of claim 1 further comprising:
before receiving, into a transformer as input, a token vector:
a) receiving, as input, a natural language text;
b) pre-processing the natural language text to identify a target word;
c) converting words in the natural language text into tokens and inserting the tokens into a token vector.
8. The method of claim 1 further comprising:
before receiving, into a transformer as input, a token vector:
a) receiving, as input, a natural language text;
b) pre-processing the natural language text to identify a target word and features of the text;
c) converting words in the natural language text into tokens and inserting the tokens into a token vector.
Yes. But for the recited additional elements as shown above in bold, the remaining limitations of the claims recite mental processes. The claims are directed to identifying substitute words. The step of “receiving…a token vector…” includes observation. The steps of “generating one or more potential substitute words…,” “generating…a paraphrased text…,” and “comparing each paraphrased text to the input text…” include evaluation and judgment. The step of “identifying one or more valid substitute words for the target word…” includes judgment or opinion. Thus, the claims recite an abstract idea.
Step 2A, Prong Two: Is the Abstract Idea Integrated into a Practical Application? (see MPEP 2106.04(d))
No. The claims as a whole merely use a computer as a tool to perform the abstract idea. The computing components (i.e., additional elements that are in bold above) are recited at a high level of generality and are merely invoked as a tool to implement the steps. For example, only a programmed general purpose computing device is needed to implement the claimed process. Simply implementing the abstract idea on a generic computer is not a practical application of the abstract idea. Furthermore, the abstract idea is merely being linked to a particular technological environment, i.e., a transformer/model/computing environment. Employing existing technology within a transformer/model/computing environment to execute the abstract idea, even when limiting the use of the abstract idea to this environment, does not integrate the exception into a practical application or add significantly more. Additionally, there is no improvement to the functioning of a computer or technology. Therefore, the abstract idea is not integrated into a practical application.
Step 2B: Does the Claim Provide an Inventive Concept? (see MPEP 2106.05)
No. As discussed with respect to Step 2A, Prong 2, the additional elements in the claims, both individually and in combination, amount to no more than tools to perform the abstract idea. Merely performing the abstract idea using a computer cannot provide an inventive concept. Therefore, the claims do not provide an inventive concept.
As such, the claims are not patent eligible.
Claim Rejections - 35 USC § 103
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.
Claims 1-5 and 7-20 are rejected under 35 U.S.C. 103 as being unpatentable over Barborak et al., U.S. Patent Application Publication No. 2017/0371861 A1 and McCann et al., U.S. Patent Application Publication Number 2019/0251168 A1.
Claim 1:
Barborak teaches:
receiving…a token vector, wherein the token vector contains tokens representing words in a natural language input text (see at least Barborak, paragraph 0245 (“One component of the knowledge induction engine 140 is a word sense disambiguator 1306 that may be executed to disambiguate word senses. Given a sentence, clause, or other text string, the word sense disambiguator 1306 identifies the senses of nouns, verbs, adjectives, adverbs, and prepositions. In the case of nouns, for example, the word sense disambiguator 1306 may differentiate between the word “ball” as either a formal dance or a piece of sports equipment, or the word “bat” as either a flying mammal or another piece of sports equipment. The disambiguator 1306 may use sense-annotated resources compiled in various ways including, for example, training data, unambiguous word senses in large text corpora 410, and sample word-senses derived from running algorithms on the large text corpora 410. In other implementations, the word sense disambiguator 1306 may further access existing third-party sense inventories, such as WordNet for nouns, verbs, adjectives, and adverbs, or a publicly available preposition sense inventory.”); paragraph 0256 (“FIG. 14 shows a process 1400 that is executed by the knowledge induction engine 140 to provide probable candidates for senses and relations of words/phrases in the story to aid the semantic inferences being made by the knowledge integration engine 136. The process 1400 is described with reference to the system architecture 100 and knowledge induction engine 140 of FIGS. 1, 4, 5, and 13. At 1402, queries for analyzing words/phrases found in the story (or text string) are received. In the architecture 100, the knowledge integration engine 136 may submit queries for words provided in the story that have been or will be semantically processed. The queries may include the words or phrases, tokenized versions of the words/phrases, or other representations of words/phrases.”)).
generating one or more potential substitute words for the target word (see at least Barborak, paragraph 0256 (“FIG. 14 shows a process 1400 that is executed by the knowledge induction engine 140 to provide probable candidates for senses and relations of words/phrases in the story to aid the semantic inferences being made by the knowledge integration engine 136. The process 1400 is described with reference to the system architecture 100 and knowledge induction engine 140 of FIGS. 1, 4, 5, and 13. At 1402, queries for analyzing words/phrases found in the story (or text string) are received. In the architecture 100, the knowledge integration engine 136 may submit queries for words provided in the story that have been or will be semantically processed. The queries may include the words or phrases, tokenized versions of the words/phrases, or other representations of words/phrases.”); paragraph 0258 (“At 1406, word sense analysis is performed on the words/phrases to determine possible senses. For each word/phrase, different vector representations are created using sense definitions and possible senses are calculated as a function of those vector representations. More particularly, one implementation of the word sense analysis 1406 is shown as acts 1406(1)-(4). At 1406(1), a sense vector is created for each word/phrase. The sense vector is calculated by first parsing sense definitions corresponding to the word/phrase to produce syntactic tokens of each sense definition. These sense definitions may be maintained in a rules or definitions datastore that may be part of the corpora 410 or induced knowledge resources 412. Afterwards, the tokens for each sense definition are algorithmically processed to produce corresponding vectors, and these vectors are summed to produce a sense vector.”)).
generating, for each potential substitute word, a paraphrased text, that is a paraphrase of the input text with each potential substitute word (see at least Barborak, paragraph 0247 (“The sense calculator 1310 is provided to estimate a prior for each sense from frequency information, such as that found in training data like the large language corpora 410. The sense calculator 1310 derives a sense for the word/phrase as a function of the sense vector, the context vector, and the prior. In one implementation, the sense calculator 1310 may apply a cosine-similarity function for the sense vector and context vector and weight each of the three inputs—sense vector, context vector, and prior.”); paragraph 0248 (“The knowledge induction service engine 1304 may further include a paraphrase detector 1312 to find and recognize paraphrases in the sentence or text string. A paraphrase of a word or phrase is another word or phrase that is written differently but roughly has the same meaning For example, the phrase “crowd erupted” is approximately the same as another phrase “applause in the stands”. The paraphrase detector 1308 uses background knowledge from the large language corpora 410 and other sources to recognize similar phrases.”); paragraph 0260 (“At 1408, the words/phrases in the queries may be analyzed to detect paraphrases, relations, and/or entity types. Background knowledge from tagged resources, like large corpora 410 or other sources may be examined to identify one or more paraphrases, relations, and/or entity types that might apply to the words/phrases under analysis. At 1410, one or more other services—text entailment detection, missing text generation, scene analysis, text embedding, text matcher, etc.—may be performed.”)).
comparing each paraphrased text to the input text to determine whether the potential substitute word is a valid substitute word (see at least Barborak, paragraph 0211 (“In FIG. 9, the belief generation components 402 may further include an episodic frame inference module 918 that implements the frame inference process to retrieve one or more uninstantiated frames 914(1)-(F) from the current world model 138 that may be relevant to the current beliefs of the story. The frames are inferred in part based on the subset of GSP structure instances 916 that are identified to represent a possible semantic interpretation of the text. The episodic frame inference module 918 proposes probability distributions over episodic frames, aligns entities in the story to roles in the frame, and new beliefs are inferred from the frames. This process may occur iteratively to discover the frame hierarchy that explains the story. In the Ben and Ava story, for example, a restaurant frame 920, for example, may provide the implicit information that Ben and Ava are in a restaurant, thereby providing a theme or context within which to better understand the explicit language in the story. The restaurant frame 920 may be selected for this purpose as having a higher likelihood of being relevant to a story that involves people, a menu, food, and tables, as compared to possible other frames for hotels or pubs.”); paragraph 0261 (“At 1412, semantic primitives are predicted from background knowledge sources, such as large language corpora. As a background process, the induction engine 140 may analyze corpora to examine various subject-verb-object (or PAS) combinations as to what other combinations might be relevant to them. Values of relevance may be computed based on how related these combinations tend to be in large corpora. At 1414, to the extent not otherwise produced through the analyses at 1404, probabilities are calculated to help rank the multiple interpretation candidates discovered by the analysis. The probabilities may be passed back in response to the queries and used by the knowledge integration engine 136 to select appropriate interpretations when inferring semantic and frame level understanding.”)).
identifying one or more valid substitute words for the target word, from the one or more potential substitute words (see at least Barborak, paragraph 0261 (“At 1412, semantic primitives are predicted from background knowledge sources, such as large language corpora. As a background process, the induction engine 140 may analyze corpora to examine various subject-verb-object (or PAS) combinations as to what other combinations might be relevant to them. Values of relevance may be computed based on how related these combinations tend to be in large corpora. At 1414, to the extent not otherwise produced through the analyses at 1404, probabilities are calculated to help rank the multiple interpretation candidates discovered by the analysis. The probabilities may be passed back in response to the queries and used by the knowledge integration engine 136 to select appropriate interpretations when inferring semantic and frame level understanding.”)).
Barborak does not explicitly teach, but McCann, however, does teach:
into a transformer (see at least McCann, Figure 3 and associated text; Figure 5 and associated text; Abstract; paragraph 0036 (“FIG. 5 is a simplified diagram of a layer 500 for an attention-based transformer network according to some embodiments. According to some embodiments, each transformer layer 351 and/or 352 of system 300 is consistent with layer 500. As shown in FIG. 5, layer 500 includes an encoding layer 510 and a decoding layer 520.”); paragraph 0037 (“Encoding layer 510 receives layer input (e.g., from an input network for a first layer in an encoding stack or from layer output of a next lowest layer for all other layers of the encoding stack) and provides it to all three (q, k, and v) inputs of a multi-head attention layer 511, thus multi-head attention layer 511 is configured as a self-attention network. Each head of multi-head attention layer 511 is consistent with attention network 400. In some examples, multi-head attention layer 511 includes three heads, however, other numbers of heads such as two or more than three are possible. In some examples, each attention layer has a dimension of 200 and a hidden size of 128. The output of multi-head attention layer 511 is provided to a feed forward network 512 with both the input and output of feed forward network 512 being provided to an addition and normalization module 513, which generates the layer output for encoding layer 510. In some examples, feed forward network 512 is a two-layer perceptron network, which implements Equation 11 where γ is the input to feed forward network 512 and M.sub.i and b.sub.i are the weights and biases respectively of each of the layers in the perceptron network. In some examples, addition and normalization module 513 is substantially similar to addition and normalization module 450.”)).
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 incorporate McCann’s self-attention based transformer that includes an encoder and decoder with Barborak’s syntactic analysis and revision of natural language input. One of ordinary skill in the art would have been motivated to incorporate this feature for the purpose of understanding context and relationships of text so that substitute or revised text can accurately be identified.
Claim 2:
Barborak further teaches:
wherein the generated potential substitute words are ordered by an output score (see at least Barborak, paragraph 0081 (“What choice below uses the word “bring” most similarly to the sentence above? The computing system 102 also offers multiple options, and provides those options in a second dialog box 128(b) that is attributed to the collaborator's response (as represented by the user icon). The system can rank the answer choices in the generated question based on its internal confidence of each of the choices. In this example, there are three options from which the collaborator can choose.”); paragraph 0092 (“Among its various applied techniques, the knowledge induction engine 140 disambiguates the word sense of a word (e.g., the word “ball” may be a formal dance or a piece of sports equipment). The induction engine 140 can find/recognize paraphrases, where words and phrases can be rewritten but have roughly the same meaning (e.g., is “crowd erupted” approximately the same as “applause in the stands”?). The knowledge induction engine 140 may further detect relations among words or phrases (e.g., in the phrase “OPEC ramped up the price of oil”, the phrase “ramped up” has a relation of increasing an amount). The knowledge induction engine 140 may perform other forms of word and phrase analysis to detect and unlock other related knowledge. In each of these cases, the knowledge induction engine 140 returns to the knowledge integration engine 136 a ranked list of candidates with associated inference probabilities that are used by the knowledge integration engine 136 to select proper word senses and build accurate semantic structures and frames that infuse meaning into the story model 132. More detailed discussion of the knowledge integration engine 136 and the knowledge induction engine 140 is provided below with references at least to FIGS. 4, 9, and 13.”)).
Claim 3:
Barborak further teaches:
wherein the identified valid substitute words are the top k valid substitutes (see at least Barborak, paragraph 0251 (“The knowledge induction service engine 1304 may further include a scene analyzer 1328 to predict what type of scenes may be inferred from texts. The scene analyzer 1328 explores known corpora 410 and other sources to identify the most popular phrases under particular scenes. As one example, suppose a text reads, “I ordered some food and then drank coffee.” The scene analyzer 1328 may explore background knowledge sources to detect scenes that contain the words/phrases such as “food”, “coffee”, “ordered”, “ordered some food” and “drank coffee.” In this example, the scene analyzer 1328 may return a ranked list of possible scenes such as “coffee house”, “diner”, “commissary”, “crib”, “verandah”, “café” and “patio.””); paragraph 0252 (“The knowledge induction service engine 1304 further has a background informed, corpus-based inference module 1330 that trains on corpora 410 and other sources (e.g., non-constrained sources like Wikipedia) to predict future semantic primitives from the background knowledge. As one example for discussion purposes, the inference module 1330 examines subject-verb-object (or PAS) combinations in the corpora 410 and explores what other combinations most closely resemble the target combination. Resemblance may be determined in various ways, such as by a scoring algorithm that computes relevance or likelihood of proximity in a text. For instance, suppose the subject-verb-structure contained “Dave eat food”. Other results may include, in ranked order, “Dave gain weight” (score of 4), “Dave lose weight” (score of 3), “Dave take criticism” (score of 3), “Dave lose pound” (score of 2.5), “Dave drink wine” (score of 2.5), and “Dave conquer tension” (score of 2.5). As another example, suppose the subject-verb-structure contained “I buy car”. Other results may include, in ranked order, “I give dollar” (score of 6), “repeat buy car” (score of 6), “I give deposit” (score of 6), “I pay proof” (score of 5), and “I take car” (score of 4.5).”)).
Claim 4:
Barborak further teaches:
identifying the semantic frame most in common among the valid substitute words (see at least Barborak, paragraph 0052 (“The computing system maintains and continuously updates a current world model that contains its beliefs about what is true about the world. The current world model can be made of a collection of frames, where each frame is a collection of propositions, such as GSPs, that are likely to be true in some common context. For example, in the Enzo and Zoe story, frame structures may provide what generally occurs during a race. The computing system constructs the story model drawing upon the knowledge in its current world model and on knowledge induced automatically from large language corpora.”)).
Claim 5:
Barborak further teaches:
before receiving, into a transformer as input, a token vector: a) receiving, as input, a natural language text (see at least Barborak, paragraph 0089 (“A knowledge integration engine 136 receives as input the story and the linguistic analysis results from the story parsing engine 134. The knowledge integration engine 136 builds an initial, probabilistic semantic representation of the story that makes sense with respect to the system's current knowledge about the world, which is maintained in a current world model 138. The initial semantic representation forms the first version of the story model 132 that is then evolved over time by the knowledge integration engine 136 through use of human interaction and previously acquired knowledge resources. In a story, information is often left unsaid as it is assumed; unfortunately, this also can result in ambiguous meanings. As will be described below in more detail, the knowledge integration engine 136 infers relevant semantic structures that effectively predict what is likely unsaid, so the system can form better knowledge models and ask more intelligent questions of the human students or collaborators. When the story says, “Ben brought the food to Ava”, the knowledge integration engine 136 assesses what information may be missing, like: “Is Ben a waiter?”, “Is Ava in a restaurant?” and so on.”); paragraph 0105 (“With reference to FIG. 2A, at 202, a story is received at the learning and understanding computing system 102. The story is formed of multiple text sentences, as exemplified by the short “Ben and Ava” story 106(1) and the short “Enzo and Zoe” story 106(T). The story is ingested in, or converted to, a digital format by the system 102. The same story is received by a user device 114/118, at 204, and presented to the user for him or her to read the story, at 206. The story may be displayed and/or converted to an audio output for the user to consume. The story may be accompanied by a number of reading-comprehension questions.”)).
b) converting words in the natural language text into tokens and inserting the tokens into a token vector (see at least Barborak, paragraph 0127 (“With joint inference, the knowledge integration engine 136 combines levels of interpretation. That is, the engine 136 interprets text at various levels of conceptual richness. Higher levels of interpretation are more powerful, but also more implicit and therefore harder to infer. In one implementation, the interpretation levels include a first or base level which is essentially the natural language text, perhaps expressed in a sequence of words (or sometimes generically called “tokens”). The second or next interpretation level involves a linguistic analysis of the natural language text. This linguistic analysis may be performed to provide a grammatical parse and statistical word similarity (e.g., embeddings). This second level of interpretation is provided by the story parsing engine 134.”); paragraph 0246 (“In one implementation, the word sense disambiguator 1306 is embodied as programmatic software modules that include a vector calculator 1308 and a sense calculator 1310. The vector calculator 1308 generates different vector representations for each syntactic token in a sense definition and sums the vectors to produce a sense vector. The vector calculator 1308 further computes a context vector for a word/phrase by treating the sentence without the word/phrase as the context, parse the reduced sentence, and produce a vector representation from the syntactic tokens. In one embodiment, an embedding algorithm is used to create the vectors for each syntactic token, such as word embedding that operates on tokens (rather than words).”)).
Claim 7:
Barborak further teaches:
before receiving, into a transformer as input, a token vector: a) receiving, as input, a natural language text (see at least Barborak, paragraph 0089 (“A knowledge integration engine 136 receives as input the story and the linguistic analysis results from the story parsing engine 134. The knowledge integration engine 136 builds an initial, probabilistic semantic representation of the story that makes sense with respect to the system's current knowledge about the world, which is maintained in a current world model 138. The initial semantic representation forms the first version of the story model 132 that is then evolved over time by the knowledge integration engine 136 through use of human interaction and previously acquired knowledge resources. In a story, information is often left unsaid as it is assumed; unfortunately, this also can result in ambiguous meanings. As will be described below in more detail, the knowledge integration engine 136 infers relevant semantic structures that effectively predict what is likely unsaid, so the system can form better knowledge models and ask more intelligent questions of the human students or collaborators. When the story says, “Ben brought the food to Ava”, the knowledge integration engine 136 assesses what information may be missing, like: “Is Ben a waiter?”, “Is Ava in a restaurant?” and so on.”); paragraph 0105 (“With reference to FIG. 2A, at 202, a story is received at the learning and understanding computing system 102. The story is formed of multiple text sentences, as exemplified by the short “Ben and Ava” story 106(1) and the short “Enzo and Zoe” story 106(T). The story is ingested in, or converted to, a digital format by the system 102. The same story is received by a user device 114/118, at 204, and presented to the user for him or her to read the story, at 206. The story may be displayed and/or converted to an audio output for the user to consume. The story may be accompanied by a number of reading-comprehension questions.”)).
b) pre-processing the natural language text to identify a target word (see at least Barborak, paragraph 0181 (“FIG. 7 shows one implementation of the story parsing engine 134, illustrating select components that may be used to process a text string, such as a story. The story parsing engine 134 is configured to propose multiple possible linguistic analysis results, and to pass those results onto the knowledge integration engine, which calculates a joint distribution over the results using joint inference. The story parsing engine 134 ingests a story 106 and passes a digital representation of the story 106 to a linguistic analyzer 702 for natural language processing (NLP). The linguistic analyzer 702 receives the story 106 and breaks the story into digestible segments, such as words, phrases, sentences, or other definable text-strings. The linguistic analyzer 702 has a set of NLP components that perform various language analyses on the text strings. A syntactic parser 704 identifies the parts of speech of words and the grammatical relationships between them in a sentence. In one implementation, the syntactic parser 704 is implemented in part by using the Stanford CoreNLP package for syntactic parsing.”); paragraph 0182 (“In some implementations, the story engine 134 may employ a single parser which outputs multiple possible parses for a sentence or multiple parsers 706 to provide parsing diversity. A parse selector 708 may be used to choose or merge the parse results according to desired applications, with the goal to ultimately improve parse accuracy for the given applications. In other implementations, there may be no parse selector, but rather the multiple parse results will be passed to the knowledge integration engine 136, which will determine the confidence in each parse result jointly with the confidence in the semantic and frame structures, as described below in more detail.”)).
c) converting words in the natural language text into tokens and inserting the tokens into a token vector (see at least Barborak, paragraph 0127 (“With joint inference, the knowledge integration engine 136 combines levels of interpretation. That is, the engine 136 interprets text at various levels of conceptual richness. Higher levels of interpretation are more powerful, but also more implicit and therefore harder to infer. In one implementation, the interpretation levels include a first or base level which is essentially the natural language text, perhaps expressed in a sequence of words (or sometimes generically called “tokens”). The second or next interpretation level involves a linguistic analysis of the natural language text. This linguistic analysis may be performed to provide a grammatical parse and statistical word similarity (e.g., embeddings). This second level of interpretation is provided by the story parsing engine 134.”); paragraph 0246 (“In one implementation, the word sense disambiguator 1306 is embodied as programmatic software modules that include a vector calculator 1308 and a sense calculator 1310. The vector calculator 1308 generates different vector representations for each syntactic token in a sense definition and sums the vectors to produce a sense vector. The vector calculator 1308 further computes a context vector for a word/phrase by treating the sentence without the word/phrase as the context, parse the reduced sentence, and produce a vector representation from the syntactic tokens. In one embodiment, an embedding algorithm is used to create the vectors for each syntactic token, such as word embedding that operates on tokens (rather than words).”)).
Claim 8:
Barborak further teaches:
before receiving, into a transformer as input, a token vector: a) receiving, as input, a natural language text (see at least Barborak, paragraph 0089 (“A knowledge integration engine 136 receives as input the story and the linguistic analysis results from the story parsing engine 134. The knowledge integration engine 136 builds an initial, probabilistic semantic representation of the story that makes sense with respect to the system's current knowledge about the world, which is maintained in a current world model 138. The initial semantic representation forms the first version of the story model 132 that is then evolved over time by the knowledge integration engine 136 through use of human interaction and previously acquired knowledge resources. In a story, information is often left unsaid as it is assumed; unfortunately, this also can result in ambiguous meanings. As will be described below in more detail, the knowledge integration engine 136 infers relevant semantic structures that effectively predict what is likely unsaid, so the system can form better knowledge models and ask more intelligent questions of the human students or collaborators. When the story says, “Ben brought the food to Ava”, the knowledge integration engine 136 assesses what information may be missing, like: “Is Ben a waiter?”, “Is Ava in a restaurant?” and so on.”); paragraph 0105 (“With reference to FIG. 2A, at 202, a story is received at the learning and understanding computing system 102. The story is formed of multiple text sentences, as exemplified by the short “Ben and Ava” story 106(1) and the short “Enzo and Zoe” story 106(T). The story is ingested in, or converted to, a digital format by the system 102. The same story is received by a user device 114/118, at 204, and presented to the user for him or her to read the story, at 206. The story may be displayed and/or converted to an audio output for the user to consume. The story may be accompanied by a number of reading-comprehension questions.”)).
b) pre-processing the natural language text to identify a target word and features of the text (see at least Barborak, paragraph 0181 (“FIG. 7 shows one implementation of the story parsing engine 134, illustrating select components that may be used to process a text string, such as a story. The story parsing engine 134 is configured to propose multiple possible linguistic analysis results, and to pass those results onto the knowledge integration engine, which calculates a joint distribution over the results using joint inference. The story parsing engine 134 ingests a story 106 and passes a digital representation of the story 106 to a linguistic analyzer 702 for natural language processing (NLP). The linguistic analyzer 702 receives the story 106 and breaks the story into digestible segments, such as words, phrases, sentences, or other definable text-strings. The linguistic analyzer 702 has a set of NLP components that perform various language analyses on the text strings. A syntactic parser 704 identifies the parts of speech of words and the grammatical relationships between them in a sentence. In one implementation, the syntactic parser 704 is implemented in part by using the Stanford CoreNLP package for syntactic parsing.”); paragraph 0182 (“In some implementations, the story engine 134 may employ a single parser which outputs multiple possible parses for a sentence or multiple parsers 706 to provide parsing diversity. A parse selector 708 may be used to choose or merge the parse results according to desired applications, with the goal to ultimately improve parse accuracy for the given applications. In other implementations, there may be no parse selector, but rather the multiple parse results will be passed to the knowledge integration engine 136, which will determine the confidence in each parse result jointly with the confidence in the semantic and frame structures, as described below in more detail.”))
c) converting words in the natural language text into tokens and inserting the tokens into a token vector (see at least Barborak, paragraph 0127 (“With joint inference, the knowledge integration engine 136 combines levels of interpretation. That is, the engine 136 interprets text at various levels of conceptual richness. Higher levels of interpretation are more powerful, but also more implicit and therefore harder to infer. In one implementation, the interpretation levels include a first or base level which is essentially the natural language text, perhaps expressed in a sequence of words (or sometimes generically called “tokens”). The second or next interpretation level involves a linguistic analysis of the natural language text. This linguistic analysis may be performed to provide a grammatical parse and statistical word similarity (e.g., embeddings). This second level of interpretation is provided by the story parsing engine 134.”); paragraph 0246 (“In one implementation, the word sense disambiguator 1306 is embodied as programmatic software modules that include a vector calculator 1308 and a sense calculator 1310. The vector calculator 1308 generates different vector representations for each syntactic token in a sense definition and sums the vectors to produce a sense vector. The vector calculator 1308 further computes a context vector for a word/phrase by treating the sentence without the word/phrase as the context, parse the reduced sentence, and produce a vector representation from the syntactic tokens. In one embodiment, an embedding algorithm is used to create the vectors for each syntactic token, such as word embedding that operates on tokens (rather than words).”)).
Claim 9:
Claim 9 is rejected using the same rationale that was used for the rejection of claim 1.
Claim 10:
Claim 10 is rejected using the same rationale that was used for the rejection of claim 2.
Claim 11:
Claim 11 is rejected using the same rationale that was used for the rejection of claim 3.
Claim 12:
Claim 12 is rejected using the same rationale that was used for the rejection of claim 4.
Claim 13:
Claim 13 is rejected using the same rationale that was used for the rejection of claim 5.
Claim 14:
Claim 14 is rejected using the same rationale that was used for the rejection of claim 7.
Claim 15:
Claim 15 is rejected using the same rationale that was used for the rejection of claim 8.
Claim 16:
Claim 16 is rejected using the same rationale that was used for the rejection of claim 1. Note: The McCann citation used in the rejection of claim 1 also teaches the transformer, encoder, decoder, and attention layer recited in claim 16.
Claim 17:
Claim 17 is rejected using the same rationale that was used for the rejection of claim 4.
Claim 18:
Claim 18 is rejected using the same rationale that was used for the rejection of claim 5.
Claim 19:
Claim 19 is rejected using the same rationale that was used for the rejection of claim 7.
Claim 20:
Claim 20 is rejected using the same rationale that was used for the rejection of claim 8.
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Barborak et al., U.S. Patent Application Publication No. 2017/0371861 A1; McCann et al., U.S. Patent Application Publication Number 2019/0251168 A1; and Zhou et al., U.S. Patent Application Publication Number 2021/0319188 A1.
Claim 6:
Barborak does not explicitly teach, but Zhou, however, does teach:
wherein converting words in the natural language text into tokens includes populating, with the value of zero, any tokens in the vector that do not correspond to a word (see at least Zhou, paragraph 0061 (“The output of the classifier module for each word may be a probability distribution over two values. For example, the two values may be 0 and 1, with 1 denoting that the word is determined to be likely to lead to a grammatically correct output, and 0 denoting that the word is unlikely to lead to a grammatically correct output. From the classifier module's collective output for all candidate words there can be inferred a vocabulary-length binary vector B. The vector B may include all of the words output by the classifier module 304 with the numerical value 1, i.e. those words determined to be likely to lead to a grammatically correct output.”)).
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 incorporate this feature with Barborak’s syntactic analysis and revision of natural language input. One of ordinary skill in the art would have been motivated to incorporate this feature for the purpose of providing the advantages of standardized data structure size, thereby increasing the speed of calculations as well as providing a fast check for determining how many viable options are available.
Relevant Prior Art
The following references are relevant to Applicant’s invention:
Anslow, U.S. Patent Number 11,853,695 B2. This reference teaches inserting substitute words based on target characteristics.
Zhang et al., U.S. Patent Application Publication Number 2021/0294972 A1. This reference teaches a data processing method that includes determining a context word set and a candidate substitute word set.
Volkovs et al., U.S. Patent Application Publication Number 2021/0255862 A1. This reference teaches an online system that trains a transformer.
Alloh et al., U.S. Patent Application Publication Number 2021/0124803 A1. This reference teaches user-customized computer-automated translation and specifically a word swapper module. See paragraphs 0073 and 0077.
Kargiannakis et al., U.S. Patent Application Publication Number 2020/0265185 A1. This reference teaches transformation rules for transforming a body of text.
Beller et al., U.S. Patent Application Publication Number 2020/0159825 A1. This reference teaches creating a variant message by substituting words based on a target tone vector.
Zuo et al., U.S. Patent Application Publication Number 2018/0018308 A1. This reference teaches text editing that includes word substitution.
Zamora, U.S. Patent Number 4,773,039. This reference teaches compaction and replacement of phrases.
J. Qiang, Y. Li, Y. Zhu, Y. Yuan, Y. Shi and X. Wu, "LSBert: Lexical Simplification Based on BERT," in IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 3064-3076, 2021, doi: 10.1109/TASLP.2021.3111589. This reference teaches lexical simplification based on BERT.
Email Communications
Per MPEP 502.03, Applicant may authorize email communications by filing Form PTO/SB/439, available at https://www.uspto.gov/sites/default/files/documents/sb0439.pdf, via the USPTO patent electronic filing system.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ELIZABETH H ROSEN whose telephone number is (571) 270-1850 and email address is elizabeth.rosen@uspto.gov. The examiner can normally be reached Monday-Friday, 10 AM ET - 7 PM ET.
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, Michael Anderson, can be reached at 571-270-0508. 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.
/ELIZABETH H ROSEN/Primary Examiner, 3693