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
Information Disclosure Statement
No IDS.
Priority
No Priority. The instant application, filed 10/23/2023.
Drawings
The drawings submitted on 10/23/2023. have been considered and accepted.
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, 2, 4-10, 11, 12-14 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim(s) 1, 13, and 14 the limitation(s) of “obtaining”, “extracting”, “creating” a “applying”, and “adding”, as drafted, are processes that, under broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components. Receiving dataset are just collecting data. More specifically, the mental process of a human can separate bi gram and tri-gram words from specific official language, and make a word conversion table to convert dialect to official language. Using specific language word conversion table , human can create common corpus which contain domain specific dialect word and official language word, Further, adding this hybrid dataset in to domain corpus preparing specific dataset that will be used for learning/training for later time. The claim recites augmented dataset will be used for training, however claim does not recite training steps or how model will be trained. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the --Mental.
This judicial exception is not integrated into a practical application because the recitation of “memory” and “processor” in claim 13 and 14, reads to generalized computer components, based upon the claim interpretation wherein the structure is interpreted using [0030] in the specification. Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim(s) is/are directed to an abstract idea.
The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional element of using generalized computer components to “obtaining”, “extracting”, “creating” a “applying”, and “adding”, and indicate amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim(s) is/are not patent eligible.
With respect to claim(s) 2, the claim(s) recite(s) “The method of training the text classification model for intent detection in the language dialect according to claim 1, further comprising: fine tuning a pre-trained language model using domain specific sentences adapted to the official language and the dialect language using the dialect transformation table.” The claim recites training but lack of training steps. Human can adjust a model based on changing variable, parameter. No additional elements are present in the claim.
.
With respect to claim(s) 4, the claim(s) recite(s) “wherein the official language is Modern Standard Arabic and the dialect language is a dialect of Arabic.” The claim is defining/narrowing Arabic dialect will be converted to official arabic language. No additional elements are present in the claim.
With respect to claim(s) 5, recites “wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Sudanese Arabic, Peninsular Arabic, Maghrebi Arabic, or Levantine Arabic.” This claim further defining Arabic dialect will be at least one of the dialect below: Palestinian, Egyptian, Mesopotamian, Sudanese, Peninsular, Maghrebi, or Levantine Arabic . No additional elements are present in the claim.
With respect to claim(s) 6, “wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Sudanese Arabic, Peninsular Arabic, Maghrebi Arabic, and Levantine Arabic.”. This claim further defining Arabic dialect will be every single dialect below: Palestinian, Egyptian, Mesopotamian, Sudanese, Peninsular, Maghrebi, or Levantine Arabic . No additional elements are present in the claim..
With respect to claim(s) 7, recites, “wherein each n-gram has a 2-, 3-, or 4-word combinations occurring side-by-side in a text of the domain dataset.” Human recognize character arranging word as a bi-gram or tri-gram in chain of word. The claim does not have any additional element. No additional elements are present in the claim.
With respect to claim(s) 8, “wherein when the domain dataset comprises n-grams in the language dialect, extracting a plurality of n-grams in the language dialect from the domain dataset providing an extracted plurality of n-grams in the language dialect.” Human recognize character arranging word as a bi-gram or tri-gram in chain of word and can separate those words. No additional elements are present.
With respect to claim(s) 9, recites, “wherein creating the dialect transformation table further comprises providing equivalent official language words for words in each n-gram of the extracted plurality of n-grams in the language dialect.”. Human can make a dialect to official language conversion table, and bi-gram and tri-gram words can be included in the table. No additional elements are present.
With respect to claim(s) 10, the claim(s) recite(s) “The method of training the text classification model for intent detection in the language dialect according to claim 1, wherein the text classification model is a Bidirectional Encoder Representations from Transformers (BERT) model.” BERT is well-known model used in this area. Ridha Mezzi teaches (“It treats the specific use case of Arabic-speaking patients using a combination of MINI and an adapted BERT model for mental health intent recognition …”) Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model
With respect to claim(s) 11, the claim(s) recite(s) “The method of training the text classification model for intent detection in the language dialect according to claim 2, wherein the text classification model is the fine-tuned pre- trained language model.” The claim recites training a model but lack of steps of training. Mukhtar teaches Figure 1 illustrates the steps we followed to fine-tune BERT for this task. (“In this paper, we describe the process of collecting Sudanese Dialect Data, pre-training BERT transformer model for Sudanese Arabic Data. We evaluate our model on two Arabic Natural Language Understanding downstream tasks that are different tasks I) Sentiment Analysis II) Named Entity Recognition.” Page 1, column 2, 2nd para.) (“1) Sentence classification: Before feeding the text into BERT, the [CLS] token is prepended to each sentence to work as a sentence representation. In order to fine-tune BERT for sentence classification, we inserted a classifier layer on top of the final hidden state corresponding to the [CLS] token. So, the model should learn to encode all information it needs in that hidden state. Figure 1 illustrates the steps we followed to fine-tune BERT for this task.” Page 2, second column. Section D) by Mukhtar et al. (“SudaBERT: A Pre-trained Encoder Representation For Sudanese Arabic Dialect”)
These claims further do not remedy the judicial exception being integrated into a practical application and further fail to include additional elements that are sufficient to amount to significantly more than the judicial exception.
With respect to claim(s) 12, the claim(s) recite(s) “A text classification model for intent detection in a language dialect, wherein the text classification model for intent detection is a text classification model trained according to the method of claim 1.” Human can group text and classify text under that group. Also, Claim steps are missing The claim(s) is/are directed to an abstract idea.
Claims 12 are rejected under 35 U.S.C. 101 because the claims appear to be directed to a software embodiment and not to hardware embodiment, where a machine claim is directed towards a system, apparatus, or arrangement. The claim appears to be directed towards a software embodiment. US 20250131211 A1 of the Published Specification describes the elements of the system being implemented as software alone actualizing the embodiments of the invention. The claimed limitations are capable of being performed as software as described in the above paragraphs, alone since no hardware component is being claimed. Software, alone, are not physical components and thus are not statutory since software do not define any structural and functional interrelationships between the computer programs and other claimed elements of a computer, which permit the computer' s program functionality to be realized. Hence, the stated functions comprise software and is thus not directed to a hardware embodiment. Data structures not claimed as embodied in computer readable media are descriptive material per se and are not statutory because they are not capable of causing functional change in the computer. See e.g., Warmerdam, 33 F.3d at 1361, 31, USPQ2d at 1760 (claim to a data structure per se held nonstatutory). Such claimed data structures do not define any structural and functional interrelationships between data and other claimed aspects of the invention, which permit the data structure' s functionality to be realized. In contrast, a claimed computer readable medium encoded with a data structure defines structural and functional interrelationships between the data structure and the computer software and hardware components which permit the data structure' s functionality to be realized, and is thus statutory.
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.
Claim 1, 2, 11, 4, 7, 8, 9, 10, 12, and 13, 14, is/are rejected under 35 U.S.C. 103 as being unpatentable over Ridha Mezzi et al. Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model in view of Wadhawan (“Dialect Identification in Nuanced Arabic Tweets Using Farasa Segmentation and AraBERT”) and in view of Gamal et al. (“Implementation of Machine Learning Algorithms in Arabic Sentiment Analysis Using N-Gram Features”) and further in view of TARAKJI AHMAD BISHER et al. KR 20180114781 A and further in view of Alkadri et al.(“ Enhancing Detection of Arabic Social Spam Using Data Augmentation and Machine Learning”)
Alkadri et al.(“ Enhancing Detection of Arabic Social Spam Using Data Augmentation and Machine Learning”)
Regarding Claim 1, Ridha Mezzi teaches:
A method of training a text classification model for intent detection in a language dialect, comprising: Ridha Mezzi teaches (“It treats the specific use case of Arabic-speaking patients using a combination of MINI and an adapted BERT model for mental health intent recognition by taking the written or spoken input of a patient and classifying it according to a defined diagnosis, which leads to accurate results with the use of a very unique and first-of-its-kind dataset, built from scratch using the Tunisian dialect phrases with Arabic letters.”) (“4.4. Intent Recognition and Text Classification with BERT Our main goal from using the BERT model is to classify the speech of the patient among three classes which are the diagnosis class (depressed, suicidal, etc.), the other class, and the nothing class, as depicted in Figure 8. “) by Ridha Mezzi et al. Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model
Ridha Mezzi does not explicitly teach the domain dataset is substantially in an official language.
Wadhawan teaches :
obtaining a domain dataset for training the text classification model, wherein the domain dataset is substantially in an official language, Wadhawan teaches (“The WANLP-2021 Shared Task 1 (Abdul-Mageed et al., 2021) is based on a multi-class classification problem where the aim is to recognize which country or province an Arabic tweet in the form of modern standard Arabic or dialect belongs to.” Page 2, section 2: Task Definition, column 1, first para) (“The NADI 2021 task promotes efforts made towards distinguishing both modern standard Arabic (MSA) and dialects (DA) according to their geographical origin, focusing on fine-grained dialects with new datasets. … … Subtask 1.1: Country-level MSA identification and … … Subtask 2.1: Province-level MSA identification”) page 2, column 1, para1,) (“The training dataset has a total of 21,000 tweet, validation and test datasets have 5,000 tweets each. Every example belongs to one of 100 provinces of 21 Arab countries. Additional 10M unlabeled tweets are provided that can be used in developing the systems for either or both of the tasks. F-score, Accuracy, Precision and Recall are the evaluation metrics.”) by Wadhawan (“Dialect Identification in Nuanced Arabic Tweets Using Farasa Segmentation and AraBERT”)
Wadhawan is considered to be analogous to the claimed invention because paper presents our approach to address the EACL WANLP-2021 Shared Task 1: Nuanced Arabic Dialect Identification (NADI).
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi to incorporate the teachings of Wadhawan in order to invoking text classification model is the fine- tuned pre- trained language model.
One could have been motivated to do so because the device performs all other models for subtasks and produces the best results on the validation set. (“[0182] … AraBERTv2-large out performs all other models for subtasks 1.1 and 1.2. For subtasks 2.1 and 2.2, AraBERTv0.2-base produces the best results on the validation set.” Page 4, column 2, pont 3. …”) by Wadhawan (“Dialect Identification in Nuanced Arabic Tweets Using Farasa Segmentation and AraBERT”)
The combination does not explicitly teach wherein the domain dataset contains a plurality of n-grams in the official language.
Gamal further teaches:
wherein the domain dataset contains a plurality of n-grams in the official language, and Fig. 1, table 1. Gamal teaches (“section 3 This section introduces a methodology for SA and explains how it handles the reprocessing and classification of Arabic tweets. …”page 334) (“The polarity data set utilized is a set of tweet sentences, which have been collected and labeled automatically as a positive or a negative opinion. The tweet sentences are gathered using different Arabic dialects phrases in querying tweets using Tweepy Application Programming Interface (API)17. 438,931 positive and negative tweets are collected. …”) section 3., page 334) 1 (“3.3 Generating N-gram features N-gram as the feature extractor method is utilized. The N-gram features are frequently applied in textual content classification task19. The n-gram features could be divided into letter n-gram features and word n-gram features as illustrated in table 1. Unigram refers to n-gram of length 1, bigram refers to n-gram of length 2, trigram refers to n-gram of length 3, and so on.” Page 335) Table 1 Example of N-gram (“3.4 Training model using ML classifiers” page 335) (“The accuracy is described as the rate of correctly classified tweets. It is characterized by the formula (1).” Page 336) (“One of the most standard and popular methods of validating ML classifiers is the K-fold cross validation where K is an integer value. A comparison of the performances using different feature sets on the ten different ML algorithms is done using the 10-fold cross validation. The comparative evaluation and analysis based on the acquired accuracy, precision, recall and f-measure results utilizing n-gram features are shown in the tables (2)-(5).” Page 336, section 4) (“It is noticed that using unigram outperformed using bigram or trigram with different ML algorithms on an Arabic tweets dataset. The maximum accuracy of 99.96% is obtained for unigram using PA and RR. On expanding the feature extraction procedure for bigram and trigram, it can be concluded that bigram gives preferable and better accuracy and precision than trigram. …”page 339, lines 7-10) by Gamal et al. (“Implementation of Machine Learning Algorithms in Arabic Sentiment Analysis Using N-Gram Features”)
extracting the plurality of n- grams in the official language from the domain dataset providing an extracted plurality of n-grams in the official language; Gamal teaches (“The N-gram and diverse weighting scheme were utilized to extract the most traditional features.”) (“Using Trigram as a feature extractor achieved the lowest accuracy among different ML algorithms compared with Unigram and Bigram as presented in Fig. 4.” Page 339, line 3-4, ) (“It is noticed that using unigram outperformed using bigram or trigram with different ML algorithms on an Arabic tweets dataset. The maximum accuracy of 99.96% is obtained for unigram using PA and RR. On expanding the feature extraction procedure for bigram and trigram, it can be concluded that bigram gives preferable and better accuracy and precision than trigram. …”page 339, lines 7-10) by Gamal et al. (“Implementation of Machine Learning Algorithms in Arabic Sentiment Analysis Using N-Gram Features”)
Gamal is considered to be analogous to the claimed invention because it relates to analyze the collected twitter posts in different Arabic Dialects and a comparison between the various algorithms used for SA with various n-gram as a feature extraction method.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi and Wadhawan to incorporate the teachings of Gamal in order to include n-gram feature extractor in Arabic dialect.
One could have been motivated to do so because the device can have Arabic dialect accuracy in bi gram extraction. (“…it can be concluded that bigram gives preferable and better accuracy and precision than trigram.” Page 339, line (11-15)”) by Gamal et al. (“Implementation of Machine Learning Algorithms in Arabic Sentiment Analysis Using N-Gram Features”)
The combination does not explicitly teach creating a dialect language transformation table by providing equivalent dialect words in the language dialect for words in each n-gram of the extracted plurality of n- grams in the official language;
TARAKJI AHMAD BISHER teaches:
creating a dialect language transformation table by providing equivalent dialect words in the language dialect for words in each n-gram of the extracted plurality of n- grams in the official language; TARAKJI AHMAD BISHER teaches (“… . The device can obtain a common set by combining a set of standard language words and a set of dialect words based on common words used in the standard language and the dialect.”) (“Meanwhile, the device may generate the translation mapping table by matching the second vector values and the third vector values. The device can continue to learn and update the generated translation mapping table. For example, the device may acquire a set of new nomenclature words and a set of tongue words, and repeat the process described above. The translation mapping table can learn new information based on previously stored information. Since the common words do not need to be learned separately in the translation mapping table, the unnecessary learning process can be reduced.” Page 6, last 2 lines and page 7 first 4 lines) (“The processor 710 may generate the translation mapping table by matching the second vector values and the third vector values. Processor 710 may continue to learn and update the generated translation mapping table. For example, the processor 710 may obtain a set of new nomenclature words and a set of tongue words, and repeat the process described above.” Page 13, 2nd paragraph from bottom page) (“… . The device can obtain a common set by combining a set of standard language words and a set of dialect words based on common words used in the standard language and the dialect.” Page 10, lines 18-21) by TARAKJI AHMAD BISHER et al. KR 20180114781
applying the dialect language transformation table to the domain dataset to transform the domain dataset to create a hybrid dataset; TARAKJI AHMAD BISHER teaches the device can obtain a common set by combining a set of standard language words and a set of dialect words based on common words used in the standard language and the dialect. Common word set (i.e. hybrid dataset). ( “In step 550, the device may generate a standard-language dialing mapping table. The standard language-dialect mapping table may be the translation mapping table described with reference to FIG. The standard word-dialect mapping table generated by the steps described with reference to Fig. 5 can be used to obtain one or more vector values corresponding to words included in a dialect sentence in the process of converting a dialect to a standard word.”) (“The system 100 receives the standard sentence 20 and converts the received standard sentence 20 into dialect sentences 10 or 15. For example, as shown in FIG. 1, the system 100 receives "How are things?" As the standard word sentence 20 and "What's?", Which is the English dialect sentence 10, goin 'on? " Or "What's the craic?", An Irish dialect (15).”) (“Referring to FIG. 5, in step 505, the device may obtain a set of standard language words (standard word corpus). Also, at step 510, the device may obtain a set of dialect words (dialect corpus). A corpus can mean a set of extracted samples of a language for machine translation.”) (“For example, the device may extract common words that are used equally in the standard language and dialect of a set of standard language words and a set of tongue words.”) (“The device can generate a single corpus based on the relationship between extracted common words and a set of standard word words and words included in a set of dialect words. The device can obtain a common set by combining a set of standard language words and a set of dialect words based on common words used in the standard language and the dialect.”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because it relates to a method and apparatus for converting a dialect to a standard word.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan and Gamal to incorporate the teachings of TARAKJI AHMAD BISHER in order to include dialect conversion table to official language.
One could have been motivated to do so because the device efficient translation can be done. (“…efficient translation We need a way to improve the translation process by using the similarity between the dialects and the standardized words.” Page 2, line (15-17)”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
The combination does not explicitly teach and training a text classification model using the augmented dataset, thereby producing the trained text classification model for intent detection in the language dialect.
Alkadri teaches:
adding the hybrid dataset to the domain dataset to create an augmented dataset;
Alkadri teaches Figure 3. Arabic text data augmentation example. (“3.2. Applying Data Augmentation to Collected Data In this section, the method for dataset augmentation will be explained. …” page 5) (The colors represent the selected words and each word is distinguished by a different color.) (“Using the data augmentation process in Section 3.2, the number of spam tweets were doubled four times, resulting in 1648 spam tweets. Table 5 and Figure 6 show the obtained results for the augmented dataset.” Page 10, section 4.3) (“We believe that various approaches are worth studying for future work on data augmentation for Arabic spam detection tweets. We believe that replacement strategies could be further automated by training a model that predicts which words should be replaced and learning word-specific similarity thresholds rather than a global one. Although the generative method we provided in this research achieved a good improvement over the data augmentation method, we believe that using contextual data augmentation, such as BERT, for data augmentation of Arabic text can be very promising. In addition, it would be interesting to show how the data augmentation strategies we described perform when using deep learning techniques or NLP applications other than the spam detection domain.” Page 13, line 3-13) by Alkadri et al.(“ Enhancing Detection of Arabic Social Spam Using Data Augmentation and Machine Learning”)
Alkadri teaches:
and training a text classification model using the augmented dataset, thereby producing the trained text classification model for intent detection in the language dialect.
Alkadri teaches (“3.2. Applying Data Augmentation to Collected Data.”) … … By increasing the diversity of training samples, the model will be able to learn more fundamental features of the data, leading to a high-quality classifier. First para, last two lines, section 3.2 ” page 5) (“Using our datasets described in the data preparation Section 3.1, we ran a series of classification experiments to identify spam tweets. For the test data sets, the classification results were evaluated using accuracy, precision, recall, and F1-measure for the spam (0)/non-spam (1) classes, as well as AUC (Area Under the receiver operating characteristic Curve (ROC)), and macro-averaged F1, which averages F1-measure for the spam and non-spam classes to account for the class imbalance.” Section 4, 2nd para. Page 9) by Alkadri et al.(“ Enhancing Detection of Arabic Social Spam Using Data Augmentation and Machine Learning
Alkadri is considered to be analogous to the claimed invention because it relates to enhancing detection of Arabic Social Spam Using Data Augmentation and Machine Learning.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal
and TARAKJI AHMAD BISHER to incorporate the teachings of Alkadri in order to include augmentation data to train with.
One could have been motivated to do so because the device can have effective data augmentation technique, bringing a consistent improvement to the baseline classifiers. (“… Unlike the non-augmented dataset experiments, the confusion matrix of LinearSVC for the augmented dataset tends to predict true spam and improve the percentage of detected spam tweets. In addition, we have noticed that the data augmentation technique is effective, bringing a consistent improvement to the baseline classifiers.” Page 12, lines 3-6) by Alkadri et al.(“ Enhancing Detection of Arabic Social Spam Using Data Augmentation and Machine Learning”)
Regarding Claim 2, Ridha Mezzi further teaches:
2. The method of training the text classification model for intent detection in the language dialect according to claim 1, further comprising: Ridha Mezzi teaches (“It treats the specific use case of Arabic-speaking patients using a combination of MINI and an adapted BERT model for mental health intent recognition by taking the written or spoken input of a patient and classifying it according to a defined diagnosis, which leads to accurate results with the use of a very unique and first-of-its-kind dataset, built from scratch using the Tunisian dialect phrases with Arabic letters.” Page 2, 2nd bullets) (“4.4. Intent Recognition and Text Classification with BERT Our main goal from using the BERT model is to classify the speech of the patient among three classes which are the diagnosis class (depressed, suicidal, etc.), the other class, and the nothing class, as depicted in Figure 8.” Page 12, section 4.4) by Ridha Mezzi et al. Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model
TARAKJI AHMAD BISHER further teaches:
fine tuning a pre-trained language model using domain specific sentences adapted to the official language and the dialect language using the dialect transformation table. using the dialect transformation table TARAKJI AHMAD BISHER teaches (“… . The device can obtain a common set by combining a set of standard language words and a set of dialect words based on common words used in the standard language and the dialect.” Page 10, lines 20-21) (“Meanwhile, the device may generate the translation mapping table by matching the second vector values and the third vector values. The device can continue to learn and update the generated translation mapping table. For example, the device may acquire a set of new nomenclature words and a set of tongue words, and repeat the process described above. The translation mapping table can learn new information based on previously stored information. Since the common words do not need to be learned separately in the translation mapping table, the unnecessary learning process can be reduced.” Page 6, last 2 lines) (“The processor 710 may generate the translation mapping table by matching the second vector values and the third vector values. Processor 710 may continue to learn and update the generated translation mapping table. For example, the processor 710 may obtain a set of new nomenclature words and a set of tongue words, and repeat the process described above.” Page 13, 2nd para, from bottom page) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because it relates to a method and apparatus for converting a dialect to a standard word.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER, and Alkadri to incorporate the teachings of TARAKJI AHMAD BISHER in order to include dialect conversion table to official language.
One could have been motivated to do so because the device efficient translation can be done. (“…efficient translation We need a way to improve the translation process by using the similarity between the dialects and the standardized words.” Page 2, line (15-17)”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
Regarding Claim 11, the combination teaches the device claim 2 as identified above
Wadhawan teaches further teaches:
11. The method of training the text classification model for intent detection in the language dialect according to claim 2, wherein the text classification model is the fine- tuned pre- trained language model. Wadhawan teaches (“The task is aimed at developing a system that identifies the geographical location( country/province) from where an Arabic tweet in the form of modern standard Arabic or dialect comes from. … … This is followed by carrying out experiments with different versions of two Transformer based models, AraBERT and AraELECTRA. …”) (“The WANLP-2021 Shared Task 1 (Abdul-Mageed et al., 2021) is based on a multi-class classification problem where the aim is to recognize which country or province an Arabic tweet in the form of modern standard Arabic or dialect belongs to.” Page 2, section 2: Task Definition, column 1, first para) (“The NADI 2021 task promotes efforts made towards distinguishing both modern standard Arabic (MSA) and dialects (DA) according to their geographical origin, focusing on fine-grained dialects with new datasets. … … Subtask 1.1: Country-level MSA identification and … … Subtask 2.1: Province-level MSA identification”) page 2, column 1, para1,)
(“Thus, the given dataset is cleaned in the following ways, so that the data used for fine tuning has a similar distribution to that used for the pre-training process: …” page 2, second column first para) (“AraBERT is an Arabic pretrained language model based on Google’s BERT architecture (Antoun et al.). There are six versions of the model: …” page 2, column 2, section 3.2.1) (“We make use of pre-trained AraBERT … … for fine-tuning the transformer based models. We use hugging-face1 API to fetch the pre-trained transformer based models, and then fine tuned the same on our dataset. The hyper parameters used for fine tuning these models have been specified in Table 3.” Section 4.2) by Wadhawan (“Dialect Identification in Nuanced Arabic Tweets Using Farasa
Segmentation and AraBERT”)
Wadhawan is considered to be analogous to the claimed invention because paper presents our approach to address the EACL WANLP-2021 Shared Task 1: Nuanced Arabic Dialect Identification (NADI).
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi to incorporate the teachings of Wadhawan in order to invoking text classification model is the fine- tuned pre- trained language model.
One could have been motivated to do so because the device performs all other models for subtasks and produces the best results on the validation set. (“[0182] … AraBERTv2-large out performs all other models for subtasks 1.1 and 1.2. For subtasks 2.1 and 2.2, AraBERTv0.2-base produces the best results on the validation set.” Page 4, column 2, pont 3. …”) by Wadhawan (“Dialect Identification in Nuanced Arabic Tweets Using Farasa Segmentation and AraBERT”)
Regarding Claim 4, the combination teaches the method claim 1 as identified above
Ridha Mezzi further teaches:
4. The method of training the text classification model for intent detection in the language dialect according to claim 1, wherein the official language is Modern Standard Arabic and the dialect language is a dialect of Arabic. Ridha Mezzi teaches (“We had to deal with some limits, such as the complexity of the Tunisian “Darija” (different accents, different languages included in it other than standard Arabic, such as French, Amazigh, Turkish, Maltese, Italian, etc.), which made it very difficult to convert the audio data into a text in a specific language. Our closest option was to convert the speech to text with Arabic letters, although the previously mentioned issues led to some errors such as missing some letters or failing to separate between words due to the tricky pronunciation of the dialect, which we had to take into account while building our dataset. …” page 10, lines 7-14) (“They can even interact with many different languages, such as English, Spanish, French, and even Arabic; however, to our knowledge, there is no system or application in mental health in the world that can interact with a sixty-year-old Tunisian man with no academic level, little understanding of standard Arabic, and no knowledge of any other language. For that reason, our application makes a difference because it is in the Tunisian dialect, which makes every Tunisian capable of interacting with it. …” page 846, section 6. Discussion, second paragraph) by Ridha Mezzi et al. Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model
Regarding Claim 7, the combination teaches the device claim 1 as identified above
TARAKJI AHMAD BISHER further teaches:
7. The method of training the text classification model for intent detection in the language dialect according to claim 1, wherein each n-gram has a 2-, 3-, or 4-word combinations occurring side-by-side in a text of the domain dataset. TARAKJI AHMAD BISHER teaches (“For example, a device may obtain a vector value corresponding to a word using an n-gram scheme, which is one of representative stochastic language models. An n-gram may refer to a contiguous sequence of n items in a given sentence. Items can be phonemes, syllables, letters, words, and so on. The n-gram scheme may be a probabilistic language model for predicting the n-th item based on the arrangement of n-1 items. The device can predict the vector value corresponding to the word based on the vector values corresponding to the words having the arrangement of the characters and the arrangement of the characters constituting the words not included in the translation mapping table.” page 7, 4th paragraph) (“Also, the device can assign corresponding vector values to words of a set of standard language words and a set of tongue words using an n-gram scheme. For example, the device may stochastically represent a chain of n words, thereby assigning corresponding vector values to words contained in a set of standard language words and a set of tongue words.” Page 10, 3rd paragraph from bottom) (“The device can assign corresponding vector values to words outside the lexical word using an n-gram scheme. The device can predict the vector value corresponding to the word based on the vector values corresponding to the words having the arrangement of the characters and the arrangement of the characters constituting the words not included in the translation mapping table. For example, the device may assign corresponding vector values for words outside the lexical scope by stochastic representation of a chain of n words.” Page 11, 6th paragraph from bottom page.) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because it relates to a method and apparatus for converting a dialect to a standard word.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, and Gamal to incorporate the teachings of TARAKJI AHMAD BISHER and Alkadri in order to include dialect conversion table to official language.
One could have been motivated to do so because the device efficient translation can be done. (“…efficient translation We need a way to improve the translation process by using the similarity between the dialects and the standardized words.” Page 2, line (15-17)”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
Regarding Claim 8, the combination teaches the device claim 1 as identified above
TARAKJI AHMAD BISHER further teaches:
8. The method of training the text classification model for intent detection in the language dialect according to claim 1, wherein when the domain dataset comprises n-grams in the language dialect, extracting a plurality of n-grams in the language dialect from the domain dataset providing an extracted plurality of n-grams in the language dialect. TARAKJI AHMAD BISHER teaches (“On the other hand, the translation mapping table can be generated by the device. In some embodiments, the device may obtain a set of standard language words and a set of dialects. The device can extract common words used in the standard word and the dialect from a set of standard word words and a set of dialect words and assign a first vector value to the extracted common words.” Page 6, 4th paragraph from bottom page.) (“For example, a device may obtain a vector value corresponding to a word using an n-gram scheme, which is one of representative stochastic language models. An n-gram may refer to a contiguous sequence of n items in a given sentence. Items can be phonemes, syllables, letters, words, and so on. The n-gram scheme may be a probabilistic language model for predicting the n-th item based on the arrangement of n-1 items. The device can predict the vector value corresponding to the word based on the vector values corresponding to the words having the arrangement of the characters and the arrangement of the characters constituting the words not included in the translation mapping table.” page 7, 4th paragraph) (“For example, vector values corresponding to the words included in the standard sentence "Do you want go to PER?" Are already stored in the translation mapping table, and the device is included in the "Wanna go PER?" Let's assume that we want to assign the vector values of the words that are to be found. vector values corresponding to "Wanna" can be derived based on the vector values assigned to the common words "go" and "PER" The keywords corresponding to the common words and the common words can be utilized as a baseline for learning the dialect vocabulary. Standard language words and dialect words can be projected in the same hyper-nodal space based on vector values corresponding to common words.” 2nd paragraph from bottom page) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
Regarding Claim 9, the combination teaches the method claim 8 as identified above
The combination does not explicitly teach
TARAKJI AHMAD BISHER further teaches:
9. The method of training the text classification model for intent detection in the language dialect according to claim 8, wherein creating the dialect transformation table further comprises providing equivalent official language words for words in each n-gram of the extracted plurality of n-grams in the language dialect. TARAKJI AHMAD BISHER teaches (“On the other hand, the translation mapping table can be generated by the device. In some embodiments, the device may obtain a set of standard language words and a set of dialects. The device can extract common words used in the standard word and the dialect from a set of standard word words and a set of dialect words and assign a first vector value to the extracted common words.” Page 6, 6th paragraph) (“For example, a device may obtain a vector value corresponding to a word using an n-gram scheme, which is one of representative stochastic language models. An n-gram may refer to a contiguous sequence of n items in a given sentence. Items can be phonemes, syllables, letters, words, and so on. The n-gram scheme may be a probabilistic language model for predicting the n-th item based on the arrangement of n-1 items. The device can predict the vector value corresponding to the word based on the vector values corresponding to the words having the arrangement of the characters and the arrangement of the characters constituting the words not included in the translation mapping table.” page 7, 4th paragraph) (“Also, the device can assign corresponding vector values to words of a set of standard language words and a set of tongue words using an n-gram scheme. For example, the device may stochastically represent a chain of n words, thereby assigning corresponding vector values to words contained in a set of standard language words and a set of tongue words.” Page 10, 3rd para from bottom page) (“The device can assign corresponding vector values to words outside the lexical word using an n-gram scheme. …” page 11, line 23) (“For example, vector values corresponding to the words included in the standard sentence "Do you want go to PER?" Are already stored in the translation mapping table, and the device is included in the "Wanna go PER?" Let's assume that we want to assign the vector values of the words that are to be found. vector values corresponding to "Wanna" can be derived based on the vector values assigned to the common words "go" and "PER" The keywords corresponding to the common words and the common words can be utilized as a baseline for learning the dialect vocabulary. Standard language words and dialect words can be projected in the same hyper-nodal space based on vector values corresponding to common words.” Page 6, 2nd paragraph from bottom page) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because art presents converting a dialect into a standard language.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, and Gamal, TARAKJI AHMAD BISHER and Alkadri to incorporate the teachings of TARAKJI AHMAD BISHER in order to create conversion table.
One could have been motivated to do so because the device may improve the translation process by using the similarity of the dialects and the standard language in translating the dialects of the predetermined language and the standard language. (“The system 100 according to some embodiments may improve the translation process by using the similarity of the dialects and the standard language in translating the dialects of the predetermined language and the standard language,. …”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
Regarding Claim 10, the combination teaches the method claim 1 as identified above
Ridha Mezzi further teaches:
10. The method of training the text classification model for intent detection in the language dialect according to claim 1, wherein the text classification model is a Bidirectional Encoder Representations from Transformers (BERT) model. Ridha Mezzi teaches pretrained BERT model (“4.4. Intent Recognition and Text Classification with BERT Our main goal from using the BERT model is to classify the speech of the patient among three classes which are the diagnosis class (depressed, suicidal, etc.), the other class, and the nothing class, as depicted in Figure 8.” Page 12, section 4.4) by Ridha Mezzi et al. Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model
Regarding Claim 12, the combination teaches the device claim 1 as identified above
Ridha Mezzi further teaches:
12. A text classification model for intent detection in a language dialect, wherein the text classification model for intent detection is a text classification model trained according to the method of claim 1. Ridha Mezzi teaches (“It treats the specific use case of Arabic-speaking patients using a combination of MINI and an adapted BERT model for mental health intent recognition by taking the written or spoken input of a patient and classifying it according to a defined diagnosis, which leads to accurate results with the use of a very unique and first-of-its-kind dataset, built from scratch using the Tunisian dialect phrases with Arabic letters.” Page 2nd bullet) (“4.4. Intent Recognition and Text Classification with BERT Our main goal from using the BERT model is to classify the speech of the patient among three classes which are the diagnosis class (depressed, suicidal, etc.), the other class, and the nothing class, as depicted in Figure 8.” Page 12, section 4.4) by Ridha Mezzi et al. Mental Health Intent Recognition for Arabic-Speaking Patients Using the Mini International Neuropsychiatric Interview (MINI) and BERT Model
Regarding Claim 13, the combination teaches the device claim 1 as identified above
TARAKJI AHMAD BISHER further teaches:
13. A non-transitory computer readable medium, comprising execution instruction, wherein when a processor of an electronic device executes the instructions, the electronic device performing the method according to claim 1. TARAKJI AHMAD BISHER teaches (“On the other hand, the method of operation of the device 700 may be recorded in a computer-readable recording medium having recorded thereon one or more programs including instructions for executing the method. Examples of the computer-readable recording medium include magnetic media such as a hard disk, a floppy disk and a magnetic tape, optical media such as CD-ROM and DVD, optical disks such as a floppy disk, And hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like.” Page 14, para. 4) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because it relates to a method and apparatus for converting a dialect to a standard word.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, and Gamal TARAKJI AHMAD BISHER , and Alkadri to incorporate the teachings of TARAKJI AHMAD BISHER in order to include dialect conversion table to official language.
One could have been motivated to do so because the device can have computer-readable recording medium include magnetic media such as a hard disk, a floppy disk and a magnetic tape, optical media such as CD-ROM and DVD, optical disks such as a floppy disk. (“…(“On the other hand, the method of operation of the device 700 may be recorded in a computer-readable recording medium having recorded thereon one or more programs including instructions for executing the method. Examples of the computer-readable recording medium include magnetic media such as a hard disk, a floppy disk and a magnetic tape, optical media such as CD-ROM and DVD, optical disks such as a floppy disk, …” Page 2, line (15-17)”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
Regarding Claim 14, the combination teaches the device claim 1 as identified above
TARAKJI AHMAD BISHER further teaches:
14. An electronic device, comprising: a processor, a memory, and a bus; wherein the memory is configured to store execution instruction, the processor and the memory are connected through the bus, and when the electronic device runs, the processor executes instructions stored in the memory to cause the processor to perform the method according to claim 1. TARAKJI AHMAD BISHER teaches (“The processor 710 may perform an overall role for controlling the device 700. For example, the processor 710 may control the device 700 as a whole by executing programs stored in the memory720 in the device 700. For example, The processor 710 may also perform the functions of the device700 described in Figures 2-6 by executing programs stored in the memory 720.” Page 12, para. 5) (“The programs stored in the memory 720 can be classified into a plurality of modules according to their functions, for example, a dialect detector, an entity name recognizer, and a sequence-to-sequence converter.”) (“On the other hand, the method of operation of the device 700 may be recorded in a computer-readable recording medium having recorded thereon one or more programs including instructions for executing the method. Examples of the computer-readable recording medium include magnetic media such as a hard disk, a floppy disk and a magnetic tape, optical media such as CD-ROM and DVD, optical disks such as a floppy disk, And hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like.” Page 14, para. 4) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because it relates to a method and apparatus for converting a dialect to a standard word.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, and Gamal TARAKJI AHMAD BISHER , and Alkadri to incorporate the teachings of TARAKJI AHMAD BISHER in order to include dialect conversion table to official language.
One could have been motivated to do so because the device efficient translation can be done. (“…The translation mapping table may store information for mapping a dialect word and a standard word corresponding to the corresponding dialect word based on various information about the dialect word and the standard word.” Page 6, para 5, line (4-7)”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
Claim 3, is/are rejected under 35 U.S.C. 103 as being unpatentable over Ridha Mezzi, Wadhawan, Gamal TARAKJI AHMAD BISHER , and Alkadri and further in view of MAHFUZ et al. US 20240134908 A1 Wadhawan
Regarding Claim 3, the combination teaches the device claim 2 as identified above
TARAKJI AHMAD BISHER further teaches:
3. The method of training the text classification model for intent detection in the language dialect according to claim 2, wherein the step of fine tuning the pre-trained language model using domain specific sentences adapted to the official language and the dialect using the dialect transformation table comprises: TARAKJI AHMAD BISHER teaches (“… . The device can obtain a common set by combining a set of standard language words and a set of dialect words based on common words used in the standard language and the dialect.”) (“Meanwhile, the device may generate the translation mapping table by matching the second vector values and the third vector values. The device can continue to learn and update the generated translation mapping table. For example, the device may acquire a set of new nomenclature words and a set of tongue words, and repeat the process described above. The translation mapping table can learn new information based on previously stored information. Since the common words do not need to be learned separately in the translation mapping table, the unnecessary learning process can be reduced.”) (“The processor 710 may generate the translation mapping table by matching the second vector values and the third vector values. Processor 710 may continue to learn and update the generated translation mapping table. For example, the processor 710 may obtain a set of new nomenclature words and a set of tongue words, and repeat the process described above.” by TARAKJI AHMAD BISHER et al. KR 20180114781 A
collecting domain specific sentences for training; using the dialect transformation table by creating combinations of domain specific sentences by replacing matching words and n-grams in the sentences with corresponding ones from the dialect transformation table, TARAKJI AHMAD BISHER teaches (“In step 640, the device may learn the sequence-to-sequence converter using the LSTM algorithm. The device can learn the relation between the array of vector values corresponding to the standard word reference sentence and the array of vector values corresponding to the tongue reference sentence through the artificial neural network. A device can input a number of input and output sentence pairs into an artificial neural network to derive an optimal weight for outputting a target sentence when a source sentence is input. The learned sequence sequence can be used in the translation process described with reference to FIGS. 2 and 3.”) (“The processor 710 may generate the translation mapping table by matching the second vector values and the third vector values. Processor 710 may continue to learn and update the generated translation mapping table. For example, the processor 710 may obtain a set of new nomenclature words and a set of tongue words, and repeat the process described above.” by TARAKJI AHMAD BISHER et al. KR 20180114781 A
wherein each pair of sentences is labeled as entailed sentences; TARAKJI AHMAD BISHER teaches (“In step 640, the device may learn the sequence-to-sequence converter using the LSTM algorithm. The device can learn the relation between the array of vector values corresponding to the standard word reference sentence and the array of vector values corresponding to the tongue reference sentence through the artificial neural network. A device can input a number of input and output sentence pairs into an artificial neural network to derive an optimal weight for outputting a target sentence when a source sentence is input. The learned sequence sequence can be used in the translation process described with reference to FIGS. 2 and 3.”) .” by TARAKJI AHMAD BISHER et al. KR 20180114781 A
TARAKJI AHMAD BISHER is considered to be analogous to the claimed invention because it relates to a method and apparatus for converting a dialect to a standard word.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, , Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri to incorporate the teachings of TARAKJI AHMAD BISHER in order to include dialect conversion table to official language.
One could have been motivated to do so because the device efficient translation can be done. (“…efficient translation We need a way to improve the translation process by using the similarity between the dialects and the standardized words.” Page 2, line (15-17)”) by TARAKJI AHMAD BISHER et al. KR 20180114781 A
The combination does not teach sentence entailment tasks.
MAHFUZ teaches:
fine tuning the pre-trained language model for sentence entailment tasks by updating and adapting all parameters of the language model for the specific domain. MAHFUZ teaches Sentence-BERT is used as the caption embedding generator 688 and is configured or trained to distinguish whether two sentences are in entailment of. (“[0091] During the training process illustrated in FIG. 6, a loss calculator 640 determines a loss metric based on one or more of a plurality of difference calculations (“Diff. Calc.” in FIG. 6). The training process is iterative, and during each iteration, changes in the loss metric are used to adjust machine-learning parameters of one or more of the machine-learning models of the sound captioning engine 670. For example, a machine-learning optimizer 650 may use one or more backpropagation operations (or another machine-learning optimization process) to adjust the machine-learning parameters of the machine-learning model(s) to reduce the loss metric.”) (“[0100] The machine-learning optimizer 650 is operable to modify machine-learning parameters (e.g., weights) of the audio embedding generator 680, the tag embedding generator 684, the caption embedding generator 688, or a combination thereof, to reduce the loss metric. In some implementations, the audio embedding generator 680 is pretrained and static, and the machine-learning optimizer 650 is operable to modify machine-learning parameters (e.g., weights) of the tag embedding generator 684, the caption embedding generator 688, or both, to reduce the loss metric. In some implementations, as described with reference to FIG. 7, the audio embedding generator 680, the tag embedding generator 684, and the caption embedding generator 688 are selected to be differentiable to enable the use of backpropagation to modify (e.g., train or fine-tune) machine-learning parameters (e.g., weights) of each based on the loss metric.”) (“[0105] In some implementations, Sentence-BERT is used as the caption embedding generator 688 and is configured or trained to distinguish whether two sentences are in entailment of, in contradiction to, or neutral with respect to each other. In this example, the tag embedding generator 684 is also a BERT network so that the predicted token embeddings 610 generated by the tag embedding generator 684 can be directly input into the caption embedding generator 688 (e.g., Sentence-BERT) to enable end-to-end backpropagation. In other implementations, other machine-learning models are used instead of or in addition to Sentence-BERT. For example, word2vec or FastText can be used as the caption embedding generator 688.”) (“[0108] … For example, referring to FIG. 6, the machine-learning optimizer 650 can use backpropagation to train or fine tune weights of the audio embedding generator 680, the tag embedding generator 684, and/or the caption embedding generator 688 to minimize a loss function that is based on the difference calculation 634. …”) (“[0105] In some implementations, Sentence-BERT is used as the caption embedding generator 688 and is configured or trained to distinguish whether two sentences are in entailment of, in contradiction to, or neutral with respect to each other. In this example, the tag embedding generator 684 is also a BERT network so that the predicted token embeddings 610 generated by the tag embedding generator 684 can be directly input into the caption embedding generator 688 (e.g., Sentence-BERT) to enable end-to-end backpropagation. In other implementations, other machine-learning models are used instead of or in addition to Sentence-BERT. For example, word2vec or FastText can be used as the caption embedding generator 688.”) by MAHFUZ et al. US 20240134908 A1
MAHFUZ is considered to be analogous to the claimed invention because it related to searching media content for particular sounds.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri to incorporate the teachings of MAHFUZ in order to include Sentence-BERT in the device.
One could have been motivated to do so because the device can be improved because more accurate embedding will have. (“[0102] … Training to make the word/token embeddings (e.g., the predicted token embeddings 610) more accurate can be improved by also configuring the loss calculator 640 to determine the loss metric partially based on the cosine distance between word/token embeddings (e.g., the ground truth token embeddings 612 and corresponding predicted token embeddings 610”)....” MAHFUZ et al. US 20240134908 A1
Claim 5, is/are rejected under 35 U.S.C. 103 as being unpatentable over Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri in view of GUNASEKARA et al. US 20220382999 A1
Regarding Claim 5, the combination teaches the method claim 4 as identified above
The combination does not explicitly teach wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Sudanese Arabic, Peninsular Arabic, Maghrebi Arabic, or Levantine Arabic.
GUNASEKARA teaches:
5. The method of training the text classification model for intent detection in the language dialect according to claim 4, wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Sudanese Arabic, Peninsular Arabic, Maghrebi Arabic, or Levantine Arabic. GUNASEKARA teaches (“[0074] In some examples, the machine learning models of the STT engine may also be trained for other languages such as Farsi and Levantine Arabic. Machine learning models of the STT engine may also be fine-tuned and trained specifically for different languages or dialects. For example, in the case of Levantine Arabic, it is possible to use a large Modern Standard Arabic (MSA) model with 1000 hour MSA data, then transfer the learning onto Levantine Arabic with 350 hours of Levantine Arabic data. …”) (“[0121] In addition, in relation to Levantine Arabic, an Adam optimizer may be used for scheduling the learning rate. Moreover, fine tuning may be performed on a subset of Levantine Arabic training datasets. Given the scarcity of the available Levantine corpora, initially the model may be trained on a Modern Standard Arabic (MSA) dataset and the resulting model may then be finetuned on a subset of pure Levantine-English parallel text. Moreover, in order to avoid overshooting and overfitting the model on the subset of the Levantine Arabic parallel text, finetuning may be realized with a relatively smaller learning rate.”) by GUNASEKARA et al. US 20220382999 A1
GUNASEKARA is considered to be analogous to the claimed invention because it relates to methods and systems for translation, and in particular to methods and systems for speech-to-speech translation.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri to incorporate the teachings of GUNASEKARA in order to training and fine-tuning in Levantine Arabic.
One could have been motivated to do so because the device may be performed on a subset of Levantine Arabic training datasets so that different conversant people can communicate with each other. (“[0003] People in different regions of the globe speak and understand different languages. Due to this difference in languages, a first group of people conversant in a first language may not be able to communicate with a second group of people conversant in a second language. This language barrier to communication may pose challenges to understanding and cooperation between the first and second groups of people”).GUNASEKARA et al. US 20220382999 A1
Claim 6 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri and GUNASEKARA in view of Manar (“ Neural Machine Translation for Arabic Language”) and further in view of Mukhtar et al. (“SudaBERT: A Pre-trained Encoder Representation For Sudanese Arabic Dialect”)
.
The combination does not explicitly teach wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Sudanese Arabic, Peninsular Arabic, Maghrebi Arabic, and Levantine Arabic
Regarding Claim 6, the combination teaches the device claim 5 as identified above
GUNASEKARA further teaches:
6. The method of training the text classification model for intent detection in the language dialect according to claim 5, wherein the language dialect is . GUNASEKARA teaches (“[0121] In addition, in relation to Levantine Arabic, an Adam optimizer may be used for scheduling the learning rate. Moreover, fine tuning may be performed on a subset of Levantine Arabic training datasets. Given the scarcity of the available Levantine corpora, initially the model may be trained on a Modern Standard Arabic (MSA) dataset and the resulting model may then be finetuned on a subset of pure Levantine-English parallel text. Moreover, in order to avoid overshooting and overfitting the model on the subset of the Levantine Arabic parallel text, finetuning may be realized with a relatively smaller learning rate.”) by GUNASEKARA et al. US 20220382999 A1
GUNASEKARA is considered to be analogous to the claimed invention because it relates to methods and systems for translation, and in particular to methods and systems for speech-to-speech translation.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri to incorporate the teachings of GUNASEKARA in order to training and fine-tuning in Levantine Arabic.
One could have been motivated to do so because the device may be performed on a subset of Levantine Arabic training datasets so that different conversant people can communicate with each other. (“[0003] People in different regions of the globe speak and understand different languages. Due to this difference in languages, a first group of people conversant in a first language may not be able to communicate with a second group of people conversant in a second language. This language barrier to communication may pose challenges to understanding and cooperation between the first and second groups of people”).GUNASEKARA et al. US 20220382999 A1
The combination does not explicitly teach wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Sudanese Arabic, Peninsular Arabic, and Maghrebi Arabic.
Manar teaches:
wherein the language dialect is Palestinian Arabic, Egyptian Arabic, Mesopotamian Arabic, Peninsular Arabic, Maghrebi Arabic Manar teaches (“Modern Standard Arabic (MSA) is used for TV, newspapers, poetry and in books. Arabic Courses at the Arab Academy are also taught in the Modern Standard form. The MSA can be transformed to adapt to new words that need to be created because of science or technology. However, the written Arabic script has seen no change in the alphabet, spelling or vocabulary in at least four millenniums. Hardly any living language can claim such a distinction. Dialect Arabic or “colloquial Arabic” is casually utilized daily by Arabs. It is found in various nations and districts of a nation (Shaalan 2014). It is grouped into Mesopotamian Arabic, Arabian Peninsula Arabic, Syro Palestanian Arabic, Egyptian and Maghrebi Arabic. Arabic dialect generally used, mostly written, by Internet clients (Al-Kabi et al. 2014) and social media (Shaalan 2014); varies from locale to area is Dialect Arabic. In vernacular Arabic, portions of the words are acquired from MSA (Abo Bakr et al. 2008). Farghaly and Shaalan (2009) showed the significance of building local devices to chip away both Modern Standard and Dialect Arabic. Abo Bakr et al. (2008) presented a hybrid pre- processing approach that has the ability to convert paraphrases of Egyptian dialectal input into MSA such that the available NLP tools can be applied to the converted text. Siddiqui et al. (2016) worked on Sentiment Analysis on the data containing different Arabic Dialects.”) by Manar,(“ Neural Machine Translation for Arabic Language”)
Manar is considered to be analogous to the claimed invention because it relates to a Neural Machine Translation for Arabic Language.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER , and Alkadri and GUNASEKARA to incorporate the teachings of Manar in order to include dialect conversion table to official language.
One could have been motivated to do so because Arabic dialect can be converted to standard/official Arabic. (“… presented a hybrid pre- processing approach that has the ability to convert paraphrases of Egyptian dialectal input into MSA such that the available NLP tools can be applied to the converted text. Siddiqui et al. (2016) worked on Sentiment Analysis on the data containing different Arabic Dialects.” Page 10, section 2.1, last 4 lines”) by Manar,(“ Neural Machine Translation for Arabic Language”)
The combination does not explicitly teach Sudanese arabic dialect classification.
Mukhtar teaches :
Sudanese Arabic dialect, Mukhtar teaches (“In this paper, we describe the process of collecting Sudanese Dialect Data, pre-training BERT transformer model for Sudanese Arabic Data. We evaluate our model on two Arabic Natural Language Understanding downstream tasks that are different tasks I) Sentiment Analysis II) Named Entity Recognition.” Page 1, column 2, 2nd para.) (“1) Sentence classification: Before feeding the text into BERT, the [CLS] token is prepended to each sentence to work as a sentence representation. In order to fine-tune BERT for sentence classification, we inserted a classifier layer on top of the final hidden state corresponding to the [CLS] token. So, the model should learn to encode all information it needs in that hidden state. Figure 1 illustrates the steps we followed to fine-tune BERT for this task.” Page 2, second column. Section D) by Mukhtar et al. (“SudaBERT: A Pre-trained Encoder Representation For Sudanese Arabic Dialect”)
Mukhtar is considered to be analogous to the claimed invention because it relates to a Pre-trained Encoder Representation For Sudanese Arabic Dialect.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Ridha Mezzi, Wadhawan, Gamal, TARAKJI AHMAD BISHER, GUNASEKARA and Manar to incorporate the teachings of Mukhtar in order to include dialect conversion table to official language.
One could have been motivated to do so because Arabic dialect can be Sudanese arabic. (“Conclusion… The experimental results show higher performance of SudaBERT as compared to Arabic-BERT when dealing with Sudanese dialect, while Arabic-BERT was better in understanding MSA and other Arabic dialects..” Page 4, last 4 lines”) Mukhtar et al. (“SudaBERT: A Pre-trained Encoder Representation For Sudanese Arabic Dialect”)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to FOUZIA HYE SOLAIMAN whose telephone number is (571)270-5656. The examiner can normally be reached M-F (8-5)AM.
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, Paras D. Shah can be reached at (571) 270-1650. 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.
/F.H.S./Examiner, Art Unit 2653
/Paras D Shah/Supervisory Patent Examiner, Art Unit 2653
10/29/2025
Prosecution Insights