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 .
Response to Arguments
Applicant's arguments filed 1/21/2026 have been fully considered but they are not persuasive.
The applicant contends Akabe fails to disclose the limitation “render, on an electronic display, both the inferencing task result and a graphical representation of the word-wise attention scores as an explanation of the inferencing task result”, recited in claim 17, since Akabe merely displays artifacts, not internal reasoning signals of a classifier.
The examiner disagrees. The limitation, indicated above, is interpreted in the broadest reasonable interpretation in light of the specification without reading the specification into the claim. As such, the recited limitation merely recites rendering both the inferencing result and graphical representation of word wise attention score as an explanation of the inferencing results. Such indicates “the rendering of inferencing task result and a graphical representation of the word-wise attention scores” is an explanation of the inferencing results. The limitation does not require “internal reasoning signals as an explanation nor does the recited limitation require the explanation as recited to include reasoning or explanation of how or why the inferencing result was generated. Due to the breath of the claimed language, the limitation, as indicated above, is interpreted as “the rendering …” is “an explanation of inferencing task result”.
The applicant contends correlation of Akabe’s word highlighting to “extraction of hidden layer attention that causally influences inference improperly collapses fundamentally different architectural concepts and ignores the express claim requirement that the inferencing result is generated “based in part” on those internal attention scores.
The examiner disagrees. The applicant’s remarks seem to reasoning seems to indicate the following:
Word highlighting as disclose by Akabe et al is equivalent to extracted hidden layer attention resulting in fundamentally different architectural concepts and
The office action ignores the express claimed requirement of “generation of the inferencing task result is based at least in part on word-wise attention scores produced by the hidden attention layer”.
The office action, Akabe et al in view of Costa et al, does not suggest or indicate word highlighting as “equivalent to extraction of hidden layer attention”, but Akabe et al does disclose the inferencing result or conclusion reached on the basis of evidence and reasoning such as classification label based on the scores of the words (paragraph 14). Akabe et al’s disclosure of word score-based classification label generation suggests combination with a secondary reference teaching a classifier based machine learning model with attention or word wise attention scores as recited in the newly amended claimed language, presented in response filed on 1/21/2026 and indicated in 1) above. The office action below provides explanation of Akabe et al in view of Vu et al regarding the limitation “wherein the first machine learning model comprises a hidden attention layer and wherein generation of the inferencing task result is based at least in part on word-wise attention scores produced by the hidden attention layer” as per claim 17.
Regarding 2), the remark is related to newly presented limitations and is considered in this office action. Please see below.
The applicant contends “the Office's proposed combination would require a person of ordinary skill not only to merge Akabe's word highlighting with Costa's internal attention mechanisms, but to further redesign Costa's internal architecture so that the same hidden attention signals that causally drive inference are extracted and rendered as an explanation of that inference.
The examiner disagrees. Akabe et al discloses a word classification unit while Costa discloses word classification via a machine learning model with scores or word wise attention scores. Both reference discloses classification of words in different manners or methods/apparatuses, hence as indicated in the office action, it would be obvious to one skilled in the art to combine via substitution of one well known element with another well known element. Although the office action below no longer includes Costa and introduces Vu et al, as result of the change in scope via amendments to the claim, the grounds of combination for 35 USC 103 of substituting one well known element with another well known element demonstrates the recited limitations are disclosed by Akabe et al in view of Vu et al. Please see the office action below.
The applicant contends “Moreover, the Office's rationale effectively characterizes the claimed extraction and rendering of hidden attention scores as a mere "presentation" or "display" choice. That characterization is incompatible with the claims as written. The claims do not recite simply displaying information; they require that the attention scores are internal to the model, hidden from direct output, and causally relied upon to generate the inferencing result. Features that alter how a computer system internally operates, particularly how it arrives at a result, cannot be dismissed as mere output formatting or presentation. See McRO, Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299, 1315 (Fed. Cir. 2016) (distinguishing claimed rules that govern how results are generated from mere presentation of results).
The examiner disagrees. The recited limitation merely recites “render, …, both the inferencing task result and a graphical representation of the word-wise attention scores as an explanation of the inferencing task result”. Such does not specify what constitutes as a “rending” or “explanation of the inferencing task result” nor provides further limitations describing “rendering” or “explanation of the inferencing task result”. Instead, the recited limitation is directed towards actions, render, performed on the output of the hidden layer, word-wise attention scores, and inferencing task result, as opposed to the manner in which the word wise attention scores are generated as suggested in the applicant’s remarks. The definition of the term “render” is “provide or give (aid, charity, a service, etc.)”, where displaying of a graphical representation word-wise attention scores, such as highlighting a word, and inferencing task result, such as classification result, is a manner or method of providing or giving information to a user. (https://www.dictionary.com/browse/render) As a result of the breath of the claimed language and interpretation of the limitation in broadest reasonable interpretation in light of the specification without reading the specification into the claim, “display” or “presentation” of both the inferencing task result and a graphical representation of the word-wise attention scores as equivalent to the limitation of “rendering …”.
The applicant contends “Finally, the rejection fails to address the express claim requirement that the rendered attention scores correspond to the classifier's reliance on the words of the input sentence. This limitation requires semantic alignment between internal attention weights and the inferencing decision itself. Neither Akabe nor Costa teaches or suggests such correspondence, nor does the Office identify where this requirement is met in the proposed combination. The absence of this teaching is fatal to the rejection. See Inre Rijckaert, 9 F.3d 1531, 1533 (Fed. Cir. 1993) (obviousness cannot be established where a claim limitation is missing and not suggested by the prior art)”.
The recited limitations are currently amended and changes the scope of the claim. Such change in scope is considered below and interpreted in the broadest reasonable interpretation in light of the specification without reading the specification into the claim. Please see the office action below.
Claim Rejections - 35 USC § 112
The following is a quotation of 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 9,17 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 applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 9 recites the limitation "rendering … the word-wise attention scores". The highlighted portion is ambiguous, hence unclear and indefinite as to which word-wise attention scores are referenced. Is the word-wise attention scores extracted from a hidden attention layer of the first machine learning model or the assigned word-wise attention scores? For these reasons, the claimed language fails to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor regards as the invention.
Claim 17 recites the limitation "render … the word-wise attention scores". The highlighted portion is ambiguous, hence unclear and indefinite as to which word-wise attention scores are referenced. Is the word-wise attention scores extracted from a hidden attention layer of the first machine learning model as recited in limitation “extract …” or the word-wise attention scores produced by the hidden attention layer recited in limitation “generate, …, wherein the first machine learning model …”? For these reasons, the claimed language fails to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor regards as the invention.
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(s) 1-3,5-7,9-11,13-15,17-19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Akabe et al (US Publication No.: 20220375246) in view of Vu et al (Title: A Label Attention Model for ICD Coding from Clinical Text).
Claim 1, Akabe et al discloses
a processor (Fig. 11, label processor module) that executes computer-executable components stored in a non-transitory computer-readable memory (Fig. 11, label memory, storage. Paragraph 25), wherein the computer-executable components comprise:
an access component (Fig. 2, label 220) that accesses a plain text clinical sentence (Fig. 2, label 220 as accesses a document, wherein a document includes sentences. Paragraph 58 discloses sentences in the document. Fig. 4 shows the sentence in the document as a clinical sentence.);
an assertion component (Fig. 2, label 234,236) that generates an assertion status classification label (Paragraph 53 discloses word classification unit classifies and identifies whether the word is a selection target word, a non-selection-target word, or an indeterminate word. Paragraph 54 discloses 236 estimates, for each word segmented from a text sentence in a document, a label.) for a word of interest in the plain text clinical sentence (Paragraph 52 discloses label 232 segments a sentence into a plurality of words, wherein any of the words can be considered a word of interest.), and
a display component (Fig. 2, label 250) that extracts a word-wise attention scores (Paragraph 14 discloses “a word classification unit which calculates a predetermined score for each of the plurality of segmented words based on the database and classifies the segmented word into one of a selection target word, a non-selection-target word, or an indeterminate word according to the calculated predetermined score.” The predetermined score or calculated score for each segmented score is considered a word-wise attention score.) corresponding to the plain text clinical sentence (Paragraphs 52-54 discloses the word classification unit, label 234 classifies segmented words from a sentence of the document. This indicates the score corresponds to the sentence.) and renders, on an electronic display, both the assertion status classification label and a graphical representation of the word-wise attention scores (Fig. 2, label 250, Fig. 4 shows a rendition of the assertion status classification label (words are highlighted) and graphical representation of the word-wise attention scores (highlighted words indicated words with predetermined scores meeting the classification requirement and “a score of the document, related keywords, and the like are additionally indicated on the right side of the browser screen 400” (paragraph 59).).) corresponding to the classifier’s reliance on the words of the plain text clinical sentence (Paragraph 59 discloses graphical representation of word-wise attention scores (score as explained above) and assertion status classification label (words highlighted) correspond to the classifier’s reliance on words of the plain text clinical sentence aka selected target words highlighted as per paragraph 59.).
Akabe et al discloses word classification with word selection model (Fig. 2, label 234,236), but fails to disclose classification is executed via a first machine learning model, the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model, wherein the first machine learning model comprises a hidden layer that assigns word-wise attention scores to words of the plain text clinical sentence and where the assertion status classification label is generated based at least in part on the word-wise attention scores.
Vu et al discloses classification is executed via a first machine learning model (Fig. 1), the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model (Fig. 1, label label attention layer), wherein the first machine learning model (Fig. 1) comprises a hidden layer (label label attention layer) that assigns word-wise attention scores (Fig. 1, label attention. Section Attention layer discloses hidden state vectors of words output from BILSTM shown in Fig. 1 and output L label specific vectors where A refers to a weight vector regarding the ith label in L label specific vectors.) to words of the plain text clinical sentence (Fig. 1, label w1-wn as the word tokens in a clinical document, where the document includes words of the plain text clinical sentence.) and where the assertion status classification label is generated based at least in part on the word-wise attention scores (Section 3.1 discloses “Each classifier users a single feed-forward network to predict whether a certain ICD code is assigned to the input text or not”. Fig. 1, label output layer for generating a label based at least in part on the word-wise attention scores, label attention and label attention layer.).
Akabe et al discloses a word classification unit classifying words of input text using scores and outputting classification word classification (Fig. 5, label 234) and Vu et al discloses clinical text classification including classification based on the word-wise attention (Fig. 1), hence it would be obvious to one skilled in the art before the effective filing date of the application to simply substitute one well known element of Akabe et al’s word classification unit with another well-known element of a word classifier performing classification using a neural network applying attention weights or scores to words as disclosed by Vu et al so to yield predictable results of classification for a word and improve text processing ability to emphasize words of importance, hence improving the user’s knowledge by providing words of importance.
Claim 2, Vu et al discloses an entity component that identifies the word of interest via named entity recognition. (Section 3.1 discloses “each classifier users a single feed-forward network to predict whether a certain ICD code”, where named entity is the text as a ICD code, “is assigned to the input text or not”, where identifying whether a text is a certain ICD code indicates a word of interest or not. This indicates named entity recognition is performed to id words of interest.)
Claim 3, Vu et al discloses an embedding layer of the first machine learning model generates an embedding vector for each word of the plain text clinical sentence, thereby yielding a set of embedding vectors. (Fig. 1, label embedding layer, w1-wn indicates word tokens of a clinical document D.)
Claim 5, Akabe et al discloses the assertion status classification label indicates that the plain text clinical sentence belongs to an uncertain class, a present class, an absent class, a conditional class or an historical class. (Paragraph 53 discloses word classification unit classifies and identifies whether the word is a selection target word, a non-selection-target word, or an indeterminate word, wherein depending on the classification of the word in the sentence, this indicates the sentence classification. For example, when the classification of the sentence is dependent on the classification of the word, the word classification is indeterminate word would indicate an uncertain classification of the sentence.).
Claim 6, Akabe et al discloses the graphical representation of the word-wise attention scores visually indicates which words of the plain text clinical sentence the first machine learning model focused on when generating the assertion status classification label (Fig. 4 shows the highlighted terms in the sentence which indicates emphasis of focus of the machine learning model.).
Claim 7, Akabe et al discloses a reliability component (Fig. 2, label 236) that determines, via execution of a second machine learning model (Paragraph 54), a reliability score for the assertion status classification label (Paragraph 61 discloses “The word selection model 236 which has been applied with machine learning for calculating the degree of similarity of the context with respect to a predetermined word in the document of a specific field by using the registered selection-target word set and non-selection-target word set.” Paragraph 57 discloses 236 estimates the labels for words classified by 234, wherein the label can be the same as the label provided by the classifier or different depending on 236’s findings. This indicates a reliability score for the assertion status classification label.) based on the word wise attention scores (Per paragraph 57,61, since 236 determines or estimates a label for the classified words output by 234, the estimated label is based on the word wise attention scores determined at 234. The score or degree or similarity is based on the predetermined word using the classified words from label 234. This indicates the score is based on the word wise attention scores used to classify words at 234.), and wherein the display component (Fig. 2, label 250) renders the reliability score on the electronic display (Fig. 4 shows highlighted or boxed words indicating the words with the labels as asserted by label 234,236. Such indicates renders the reliability score since the words highlighted are the words with specific labels indicated by label 234,236.).
Claim 9, Akabe et al discloses
Accessing, by a device operatively coupled to a processor (Fig. 2, label 220, Fig. 1, label 110 is operatively coupled to 120, wherein 110,120 includes a processor.), a plain text clinical sentence (Fig. 2, label 220 as accesses a document, wherein a document includes sentences. Paragraph 58 discloses sentences in the document. Fig. 4 shows the sentence in the document as a clinical sentence.);
Generating, by the device and via execution by a model (Fig. 2, label 234,236, Fig. 1, label 110,120. Paragraph 85 discloses 234 is implemented using a statistical model and paragraph 54 discloses 236 is implemented using a sequence labeling model.) an assertion status classification label (Paragraph 53 discloses word classification unit classifies and identifies whether the word is a selection target word, a non-selection-target word, or an indeterminate word. Paragraph 54 discloses 236 estimates, for each word segmented from a text sentence in a document, a label.) for a word of interest in the plain text clinical sentence (Paragraph 52 discloses label 232 segments a sentence into a plurality of words, wherein any of the words can be considered a word of interest.); and
Extracting, by the device and from the model (Fig. 2, label 234, Fig. 1, label 110,120), word-wise attention scores (Paragraph 14 discloses “a word classification unit which calculates a predetermined score for each of the plurality of segmented words based on the database and classifies the segmented word into one of a selection target word, a non-selection-target word, or an indeterminate word according to the calculated predetermined score.” The predetermined score or calculated score for each segmented score is considered a word-wise attention score.) that contributed to generation of the assertion status classification label (paragraph 14 discloses the word-wise attention score or calculated score contributes to generation of the label.); and
rendering, by the device and on an electronic display (Fig. 2, label 250, Fig. 1, label 110,120), both the assertion status classification label and a graphical representation of the word-wise attention scores (Fig. 2, label 250, Fig. 4 shows a rendition of the assertion status classification label (words are highlighted) and graphical representation of the word-wise attention scores (highlighted words indicated words with predetermined scores meeting the classification requirement and “a score of the document, related keywords, and the like are additionally indicated on the right side of the browser screen 400” (paragraph 59).).) as an explanation of the assertion status classification level (Such indicates intended use of displaying label and scores, hence such limitation is a result of displaying label and scores. Fig. 4, paragraph 59 displaying of the results indicates an explanation of the assertion status classification level.).
Akabe et al discloses word classification with word selection model (Fig. 2, label 234,236), but fails to disclose classification is executed via a first machine learning model, the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model, wherein the first machine learning model assigns a word-wise attention scores to words of the plain text clinical sentence during generation of the assertion status classification label and the word-wise attention scores contribute to generation of the assertion status classification label.
Vu et al discloses classification is executed via a first machine learning model (Fig. 1), the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model (Fig. 1, label label attention layer), wherein the first machine learning model (Fig. 1) assigns a word-wise attention scores (Fig. 1, label attention of label attention layer. Section Attention layer discloses hidden state vectors of words output from BILSTM shown in Fig. 1 and output L label specific vectors where A refers to a weight vector regarding the ith label in L label specific vectors.) to words of the plain text clinical sentence (Fig. 1, label w1-wn as the word tokens in a clinical document, where the document includes words of the plain text clinical sentence.) during generation of the assertion status classification label (Fig. 1 shows the process of generating the assertion status classification label or label of the input text, where label attention layer performs actions during the generation of the label at label output layer.) and where the assertion status classification label is generated based at least in part on the word-wise attention scores (Section 3.1 discloses “Each classifier users a single feed-forward network to predict whether a certain ICD code is assigned to the input text or not”. Fig. 1, label output layer for generating a label based at least in part on the word-wise attention scores, label attention and label attention layer.).
Akabe et al discloses a word classification unit classifying words of input text using scores and outputting classification word classification (Fig. 5, label 234) and Vu et al discloses clinical text classification including classification based on the word-wise attention (Fig. 1), hence it would be obvious to one skilled in the art before the effective filing date of the application to simply substitute one well known element of Akabe et al’s word classification unit with another well-known element of a word classifier performing classification using a neural network applying attention weights or scores to words as disclosed by Vu et al so to yield predictable results of classification for a word and improve text processing ability to emphasize words of importance, hence improving the user’s knowledge by providing words of importance.
Claim 10, Akabe et al discloses identification by the device (Fig. 2, label 220)< but fails to disclose identifying the word of interest via named entity recognition.
Vu et al discloses identifying the word of interest via named entity recognition. (Section 3.1 discloses “each classifier users a single feed-forward network to predict whether a certain ICD code”, where named entity is the text as a ICD code, “is assigned to the input text or not”, where identifying whether a text is a certain ICD code indicates a word of interest or not. This indicates named entity recognition is performed to id words of interest.)
Akabe et al discloses a word classification unit classifying words of input text using scores and outputting classification word classification (Fig. 5, label 234) and Vu et al discloses clinical text classification including classification based on the word-wise attention (Fig. 1), hence it would be obvious to one skilled in the art before the effective filing date of the application to simply substitute one well known element of Akabe et al’s word classification unit with another well-known element of a word classifier performing classification using a neural network applying attention weights or scores to words and generate embeddings in a embedding layer as disclosed by Vu et al so to yield predictable results of classification for a word and improve text processing ability to emphasize words of importance, hence improving the user’s knowledge by providing words of importance.
Claim 11, Vu et al discloses an embedding layer of the first machine learning model generates an embedding vector for each word of the plain text clinical sentence, thereby yielding a set of embedding vectors. (Fig. 1, label embedding layer, w1-wn indicates word tokens of a clinical document D.)
Claim 13, Akabe et al discloses the assertion status classification label indicates that the plain text clinical sentence belongs to an uncertain class, a present class, an absent class, a conditional class or an historical class. (Paragraph 53 discloses word classification unit classifies and identifies whether the word is a selection target word, a non-selection-target word, or an indeterminate word, wherein depending on the classification of the word in the sentence, this indicates the sentence classification. For example, when the classification of the sentence is dependent on the classification of the word, the word classification is indeterminate word would indicate an uncertain classification of the sentence.).
Claim 14, Akabe et al discloses the graphical representation of the word-wise attention scores visually indicates which words of the plain text clinical sentence the first machine learning model focused on when generating the assertion status classification label (Fig. 4 shows the highlighted terms in the sentence which indicates emphasis of focus of the machine learning model.).
Claim 15, Akabe et al discloses determining, by the device (Fig. 1, label 110,120) and via execution of a second machine learning model (Fig. 2, label 236,Paragraph 54), a reliability score for the assertion status classification label (Paragraph 61 discloses “The word selection model 236 which has been applied with machine learning for calculating the degree of similarity of the context with respect to a predetermined word in the document of a specific field by using the registered selection-target word set and non-selection-target word set.” Paragraph 57 discloses 236 estimates the labels for words classified by 234, wherein the label can be the same as the label provided by the classifier or different depending on 236’s findings. This indicates a reliability score for the assertion status classification label.) based on the word wise attention scores (Per paragraph 57,61, since 236 determines or estimates a label for the classified words output by 234, the estimated label is based on the word wise attention scores determined at 234. The score or degree or similarity is based on the predetermined word using the classified words from label 234. This indicates the score is based on the word wise attention scores used to classify words at 234.), and
rendering, by the device (Fig. 1, label 110,120), the reliability score on the electronic display (Fig. 2, label 250, Fig. 4 shows highlighted or boxed words indicating the words with the labels as asserted by label 234,236. Such indicates renders the reliability score since the words highlighted are the words with specific labels indicated by label 234,236.).
Claim 17, Akabe et al discloses
Access a plain text clinical sentence (Fig. 2, label 220 as accesses a document, wherein a document includes sentences. Paragraph 58 discloses sentences in the document. Fig. 4 shows the sentence in the document as a clinical sentence.);
Generate, via execution by a model (Fig. 2, label 234,236, Fig. 1, label 110,120. Paragraph 85 discloses 234 is implemented using a statistical model and paragraph 54 discloses 236 is implemented using a sequence labeling model.), an inferencing task result (Paragraph 53 discloses word classification unit classifies and identifies whether the word is a selection target word, a non-selection-target word, or an indeterminate word (inferencing task result). Paragraph 54 discloses 236 estimates, for each word segmented from a text sentence in a document, a label.) based on the plain text clinical sentence (Paragraph 52 discloses label 232 segments a sentence into a plurality of words, wherein any of the words can be considered a word of interest.); and
Extract, from the first machine learning model (Fig. 2, label 234, Fig. 1, label 110,120), a word-wise attention scores (Paragraph 14 discloses “a word classification unit which calculates a predetermined score for each of the plurality of segmented words based on the database and classifies the segmented word into one of a selection target word, a non-selection-target word, or an indeterminate word according to the calculated predetermined score.” The predetermined score or calculated score for each segmented score is considered a word-wise attention score.) based on the plain text clinical sentence (Paragraphs 52-54 discloses the word classification unit, label 234 classifies segmented words from a sentence of the document. This indicates the score corresponds to the sentence.); and
render, on an electronic display (Fig. 2, label 250, Fig. 1, label 110,120), both the inferencing task result and a graphical representation of the word-wise attention scores (Fig. 2, label 250, Fig. 4 shows a rendition of the assertion status classification label (words are highlighted) (inferencing task result) and graphical representation of the word-wise attention scores (highlighted words indicated words with predetermined scores meeting the classification requirement and “a score of the document, related keywords, and the like are additionally indicated on the right side of the browser screen 400” (paragraph 59).).) as an explanation of the inferencing task result (Such limitation indicates intended use of displaying labels and scores as explained above, hence the displaying of such data provides an explanation of the inferencing task result.).
Akabe et al discloses word classification with word selection model (Fig. 2, label 234,236), but fails to disclose classification is executed via a first machine learning model and the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model.
Akabe et al discloses word classification with word selection model (Fig. 2, label 234,236), but fails to disclose classification is executed via a first machine learning model, the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model, wherein the first machine learning model comprises a hidden attention layer and wherein generation of the inferencing task result is based at least in part on the word-wise attention scores produced by the hidden attention layer.
Vu et al discloses classification is executed via a first machine learning model (Fig. 1), the extracted word-wise attention scores are from a hidden attention layer of the first machine learning model (Fig. 1, label label attention layer), wherein the first machine learning model (Fig. 1) comprises the hidden attention layer (label label attention layer) and generation of the inferencing task result is based at least in part on the word-wise attention scores produced by the hidden attention layer (Fig. 1, label attention. Section Attention layer discloses hidden state vectors of words output from BILSTM shown in Fig. 1 and output L label specific vectors where A refers to a weight vector regarding the ith label in L label specific vectors. Section 3.1 discloses “Each classifier users a single feed-forward network to predict whether a certain ICD code is assigned to the input text or not. Fig. 1, label output layer for generating a label based at least in part on the word-wise attention scores, label attention and label attention layer.).
Akabe et al discloses a word classification unit classifying words of input text using scores and outputting classification word classification (Fig. 5, label 234) and Vu et al discloses clinical text classification including classification based on the word-wise attention (Fig. 1), hence it would be obvious to one skilled in the art before the effective filing date of the application to simply substitute one well known element of Akabe et al’s word classification unit with another well-known element of a word classifier performing classification using a neural network applying attention weights or scores to words as disclosed by Vu et al so to yield predictable results of classification for a word and improve text processing ability to emphasize words of importance, hence improving the user’s knowledge by providing words of importance.
Claim 18, Akabe et al discloses the graphical representation of the word-wise attention scores visually indicates which words of the plain text clinical sentence the first machine learning model focused on when generating the inferencing task result (Fig. 4 shows the highlighted terms in the sentence which indicates emphasis of focus of the machine learning model.).
Claim 19, Akabe et al discloses determine, via execution of a second machine learning model (Fig. 2, label 236,Paragraph 54), a reliability score for the assertion status classification label (Paragraph 61 discloses “The word selection model 236 which has been applied with machine learning for calculating the degree of similarity of the context with respect to a predetermined word in the document of a specific field by using the registered selection-target word set and non-selection-target word set.” Paragraph 57 discloses 236 estimates the labels for words classified by 234, wherein the label can be the same as the label provided by the classifier or different depending on 236’s findings. This indicates a reliability score for the assertion status classification label.) based on the word wise attention scores (Per paragraph 57,61, since 236 determines or estimates a label for the classified words output by 234, the estimated label is based on the word wise attention scores determined at 234. The score or degree or similarity is based on the predetermined word using the classified words from label 234. This indicates the score is based on the word wise attention scores used to classify words at 234.), and
render the reliability score on the electronic display (Fig. 2, label 250, Fig. 4 shows highlighted or boxed words indicating the words with the labels as asserted by label 234,236. Such indicates renders the reliability score since the words highlighted are the words with specific labels indicated by label 234,236.).
Allowable Subject Matter
Claims 4,8,12,16,20 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LINDA WONG whose telephone number is (571)272-6044. The examiner can normally be reached 9-5.
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, Andrew C Flanders can be reached at 571-272-7516. 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.
/LINDA WONG/Primary Examiner, Art Unit 2655