Dixa is a born global customer service tech company on a mission to solve customer chaos, eliminate bad customer service and build strong bonds between companies and their customers - much like the connection between friends

Dixa

dixa

Have you ever had that conversation with a parent when you try to face-time, and you’re the one face-timing whilst they’re the one giving you a close up of their ear, muttering words of confusion and frustration? As annoying as getting a close up of your loved one’s ear can be, I think you would agree that it’s important to power through those bizarre movements and help talk them through what they’re doing wrong so they can actually benefit themselves from the magic of modern technology. I think this same loving rule of thumb should be applied to automation and the ability to code something for yourself. The Decision To some, the ability to code can seem like magic. If you’re the person who knows how to code, that point in your show and tell where someone can’t believe what you’ve been able to do, you have to make a decision... Do you accept the praise and astonishment? Cementing your name as a modern day Merlin, or do you help to explain how actually, it’s not that hard once you know where to look? Why Accepting Merlin Status Benefits No One By explaining how something can be done in such a way that you bring as many people along for the ride with you, you will improve things for everyone, not just yourself. That explanation could help plant a seed of an idea in that other persons mind and have them googling that little bit quicker or tackling a problem in a slightly more effective way. That same person could come back to you, the coder-extraordinaire, and pose a question or challenge in a way you’ve never considered, giving you a sudden bout of motivation and a new teammate to help                                crack that million dollar question. Final Thoughts Technology is not magic. It’s built by people, for people, to do things that look like magic. It’s important that when faced with a misplaced ear, you gently explain where the camera is on the phone. Showing other people how they can do what you’ve done, helping to inspire curiosity in others, benefits everyone and actually helps to improve things long into the future. Previously published at https://www.automateall.co.uk/MagicBehindCode

The Empathic Importance of Explaining The Magic Behind Your Code in Laymen's Terms

Understanding and Generating Dialogue between Characters in Stories: Limitations and References

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 9 Conclusion In this work, we present the first study on understanding and generating inter-character dialogue in stories. To this end, we collect a Chinese story dataset DIALSTORY with a large amount of dialogue, and propose two new tasks including masked dialogue generation and dialogue speaker recognition. We also construct standardized datasets for these tasks through automatic and manual annotations based on DIALSTORY. By incorporating representations of different characters, our model outperforms strong baselines significantly on both tasks in terms of automatic and manual evaluation. The benchmark datasets, tasks, and models will further boost the development of this field. This paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 9 Conclusion In this work, we present the first study on understanding and generating inter-character dialogue in stories. To this end, we collect a Chinese story dataset DIALSTORY with a large amount of dialogue, and propose two new tasks including masked dialogue generation and dialogue speaker recognition. We also construct standardized datasets for these tasks through automatic and manual annotations based on DIALSTORY. By incorporating representations of different characters, our model outperforms strong baselines significantly on both tasks in terms of automatic and manual evaluation. The benchmark datasets, tasks, and models will further boost the development of this field. This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv

Understanding and Generating Dialogue between Characters in Stories: Conclusion

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 8 Future Work Despite the overall improved performance of our character modeling methods, we find that there is still space for further improvement. Our approach constructs the character representations from the input stories as static vectors. But the status of characters could be updated during the generation process. As a result, we can dynamically update those representations during plot development. We plan to explore the new possibility of the character update technique and leave it as an important future work. This paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 8 Future Work Despite the overall improved performance of our character modeling methods, we find that there is still space for further improvement. Our approach constructs the character representations from the input stories as static vectors. But the status of characters could be updated during the generation process. As a result, we can dynamically update those representations during plot development. We plan to explore the new possibility of the character update technique and leave it as an important future work. This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv

Understanding and Generating Dialogue between Characters in Stories: Future Work

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 7 Discussion Masked Dialogue Generation In this task, the masked positions can be anywhere in the story. To generate and complete one specific masked dialogue turn, this task requires the machine to first understand the main line of the whole story, the roles and features of different characters, and then infer and generate the most appropriate dialogue turn to move the plot forward according to the specific environment. Context information also plays an important role in this task, which puts forth a high demand for dialogue and plot coherence. And this task can easily leverage other research such as persona chat bots, emotional chat bots, or even some other story generation tasks. Dialogue Speaker Recognition This task specifically focuses on the dialogue understanding in a story. In a story, there are always lots of characters speaking to each other in different environments. And the relationship between characters and dialogue is complicated. Sometimes a piece of dialogue is self-talk, sometimes it is a conversation between two parties, or even multiple parties in a complex environment, and sometimes there are many details in the background information that cover up the actual speaker. The speaker of the dialogue will not only appear in the surrounding, but may also hide in the previous or behind the context of the dialogue. To recognize all the speakers in one story, our model also needs to capture different characteristics of people. Higher requirements for the model’s understanding of dialogue and character relationships are also put forth by the necessity to predict speakers simultaneously in several positions. This task can also leverage other research on recognition in multi-person dialogue story scenes. This paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 7 Discussion Masked Dialogue Generation In this task, the masked positions can be anywhere in the story. To generate and complete one specific masked dialogue turn, this task requires the machine to first understand the main line of the whole story, the roles and features of different characters, and then infer and generate the most appropriate dialogue turn to move the plot forward according to the specific environment. Context information also plays an important role in this task, which puts forth a high demand for dialogue and plot coherence. And this task can easily leverage other research such as persona chat bots, emotional chat bots, or even some other story generation tasks. Masked Dialogue Generation Dialogue Speaker Recognition This task specifically focuses on the dialogue understanding in a story. In a story, there are always lots of characters speaking to each other in different environments. And the relationship between characters and dialogue is complicated. Sometimes a piece of dialogue is self-talk, sometimes it is a conversation between two parties, or even multiple parties in a complex environment, and sometimes there are many details in the background information that cover up the actual speaker. The speaker of the dialogue will not only appear in the surrounding, but may also hide in the previous or behind the context of the dialogue. To recognize all the speakers in one story, our model also needs to capture different characteristics of people. Higher requirements for the model’s understanding of dialogue and character relationships are also put forth by the necessity to predict speakers simultaneously in several positions. This task can also leverage other research on recognition in multi-person dialogue story scenes. Dialogue Speaker Recognition This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv

Understanding and Generating Dialogue between Characters in Stories: Discussion

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 6 Experiments 6.1 Masked Dialogue Generation Implementation Details To conduct experiments on the masked dialogue generation task, we decide the hyper-parameters based on the performance of the validation set. We train Shao et al. (2021)’s BART model for 4.6 epochs with a 1e-4 learning rate for 1 day, and for our model, we train it for 5.6 epochs with a 1e-4 learning rate for 1 day. All baselines and our model are trained using the Adam optimizer. During the training process for our method, we computed the selection coverage of characters within a single story. And it showed that in every 1000 training steps, the coverage of different characters ranged from 98.64% to 99.00%, which meant nearly all the characters are selected during training, and all the characters contributed to the generated dialogue. It further proved that the argmax in Eq. 3 operation doesn’t break the gradient progress when training for this task. Automatic Evaluation Following previous works, we use several standard, widely used automatic evaluation metrics. We use BLEUn (Papineni et al., 2002) to measure the average word overlap between each generated and groundtruth dialogue turn (n=1,2), and Distinct-n (Li et al., 2015) to evaluate n-gram diversity of generated dialogue turns (n=2,3,4). To be more specific, for the coherence classifier, we construct the training and validation sets by randomly shuffling the order of dialogue turns and keeping other content in the correct order. We regard the perturbed story as a negative example and the original story as a positive example. We sample another 195k stories (except those in DIALSTORY) from the novels of Guan et al. (2022) to construct the training set (190k examples) and the validation set (5k examples). We train the model for 4 epochs with a 2e-5 learning rate and a 16- batch size, using the Adam optimizer. During the evaluation, we consider an example coherent when the probability of being coherent predicted by the classifier is greater than 0.5. We use the ratio of outputs (along with the input) that are classified as coherent by the classifier to all generated outputs as the coherence score. The result of the automatic evaluation is presented in Table 3. According to the table, compared to the BART baseline, our model consistently generates more word overlaps with ground truth and achieves better diversity under the guidance of character representations, which means our model can generate more diverse but not commonplace responses. Figure 3 plots the coherence score varying with the number of masked dialogue turns. The result shows that our model gets a higher coherence score than BART when required to generate more than seven turns of dialogue in one story. Manual Evaluation We conduct a pairwise comparison between our model and the BART baseline. We randomly select 100 examples from the test set. For each pair of outputs along with the input, we ask three annotators to give a preference (win, lose and tie) in terms of fluency, coherence, and informativeness. All the annotations are native Chinese speakers. We adopt majority voting to make final decisions among the annotators. The three aspects of manual evaluation are as follows: Fluency: Grammatical correctness and intra-sentence linguistic quality.


Coherence: Inter-sentence relatedness, causal and temporal dependencies. We judge the coherence between the story and a dialogue turn by following the criterion in Table 5. We add the scores of all the generated dialogue turns in a story to get the overall coherence score of the story, which is then used to compare with each other.


Informativeness: Interesting, diverse and rich details. As shown in Table 4, all the results show moderate (κ > 0.4) agreement, which shows our model outperforms the BART baseline significantly in dialogue informativeness and coherence. Case Study Figure 5 showed two examples to investigate how learning character representations can help our model generate more coherent dialogue. We found that our model can better model the relationship between different characters and the direction of the storyline. For example, in the first case, we can see that the BART’s generation confuses different characters’ fathers, while our model captures the relationship between different characters, and generates proper responses for the corresponding characters, which also moves the plot forward. And in the second case, we can see that BART’s generation is commonplace and contradicts the plot development. In contrast, our model captures the intentions of the speaker and the development trend of the plot, generating an appropriate and coherent response. Since these two models use the same pretrained weight, we can infer that the character modeling module leverages the coherent and reasonable generation. We also summarize four error types of the generated dialogue turn for the DialGen task: (1) Intersentence Contradiction; (2) Inter-sentence Repetition; (3) Intra-sentence Contradiction; (4) Intrasentence Repetition. We show the typical corresponding cases in Figure 4. We conducted a quantitative analysis of those 4 error types on our model’s generation. We analyzed 20 stories with 103 dialog turns and the results are shown in Figure 6. We found that both our model and BART suffer from these errors, suggesting that there is still space for model improvement, especially in the inter-sentence repetition. 6.2 Dialogue Speaker Recognition Implementation Details To conduct experiments on the speaker recognition task, we decide the hyper-parameters based on the performance of the validation set. For the BART baseline and our approach, we insert a mask token before each dialogue needed to be predicted, and a person id token before and after each character name span. Then, we insert all the unique person id tokens before the input stories as different options, and make predictions based on the cos similarity of option tokens and mask tokens. We train Shao et al. (2021)’s BART model for 30 epochs with a 5e-5 learning rate for 3 days, For encoder-only baselines, we implemented BERT, RoBERTa, and MacBERT and trained them for 15 epochs wity a le-5 learning rate for 2 days. For our model, we train it for 22 epochs with a le-6 learning rate for 2 days. All baselines and our model are trained using the Adam optimizer. Metrics We evaluate the DialSpk task using two automatic metrics including dialogue-level accuracy (DAC) and story-level accuracy (SAC). DAC is calculated as the ratio of the correct predictions to the total number of specifies dialogue turns, while SAC is the ratio of the number of stories where all dialogue turns are correctly predicted to the number of all test examples. These two metrics provide the evaluation for dialogue understanding with different granularities. Results As shown in Table 7, our model outperforms all the baselines significantly (p< 0.01, Wilcoxon signed-rank test) on both DAS and SAC scores, suggesting the benefit of learning character representations. We tested the accuracy of automatic training set annotations, and the DAC/SAC scores are 86.78%/67.80%. Together with the model’s performance on the test set, we can see the automatic annotation for the training set is of good quality. We also conducted the human prediction experiment, and the DAC/SAC scores are 97.90%/90.70%, which are much higher than the best model. So there is much room for further improvement for machine-based approaches. This paper is available on arxiv under CC 4.0 DEED license. [2] https://huggingface.co/fnlp/ bart-base-chinese [3] https://huggingface.co/bert-base-chinese [4] https://huggingface.co/hfl/ chinese-roberta-wwm-ext [5] https://huggingface.co/hfl/ chinese-macbert-base Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 6 Experiments 6.1 Masked Dialogue Generation Implementation Details To conduct experiments on the masked dialogue generation task, we decide the hyper-parameters based on the performance of the validation set. We train Shao et al. (2021)’s BART model for 4.6 epochs with a 1e-4 learning rate for 1 day, and for our model, we train it for 5.6 epochs with a 1e-4 learning rate for 1 day. All baselines and our model are trained using the Adam optimizer. Implementation Details During the training process for our method, we computed the selection coverage of characters within a single story. And it showed that in every 1000 training steps, the coverage of different characters ranged from 98.64% to 99.00%, which meant nearly all the characters are selected during training, and all the characters contributed to the generated dialogue. It further proved that the argmax in Eq. 3 operation doesn’t break the gradient progress when training for this task. Automatic Evaluation Following previous works, we use several standard, widely used automatic evaluation metrics. We use BLEUn (Papineni et al., 2002) to measure the average word overlap between each generated and groundtruth dialogue turn (n=1,2), and Distinct-n (Li et al., 2015) to evaluate n-gram diversity of generated dialogue turns (n=2,3,4). Automatic Evaluation To be more specific, for the coherence classifier, we construct the training and validation sets by randomly shuffling the order of dialogue turns and keeping other content in the correct order. We regard the perturbed story as a negative example and the original story as a positive example. We sample another 195k stories (except those in DIALSTORY) from the novels of Guan et al. (2022) to construct the training set (190k examples) and the validation set (5k examples). We train the model for 4 epochs with a 2e-5 learning rate and a 16- batch size, using the Adam optimizer. During the evaluation, we consider an example coherent when the probability of being coherent predicted by the classifier is greater than 0.5. We use the ratio of outputs (along with the input) that are classified as coherent by the classifier to all generated outputs as the coherence score. The result of the automatic evaluation is presented in Table 3. According to the table, compared to the BART baseline, our model consistently generates more word overlaps with ground truth and achieves better diversity under the guidance of character representations, which means our model can generate more diverse but not commonplace responses. Figure 3 plots the coherence score varying with the number of masked dialogue turns. The result shows that our model gets a higher coherence score than BART when required to generate more than seven turns of dialogue in one story. Manual Evaluation We conduct a pairwise comparison between our model and the BART baseline. We randomly select 100 examples from the test set. For each pair of outputs along with the input, we ask three annotators to give a preference (win, lose and tie) in terms of fluency, coherence, and informativeness. All the annotations are native Chinese speakers. We adopt majority voting to make final decisions among the annotators. The three aspects of manual evaluation are as follows: Manual Evaluation Fluency: Grammatical correctness and intra-sentence linguistic quality. Coherence: Inter-sentence relatedness, causal and temporal dependencies. We judge the coherence between the story and a dialogue turn by following the criterion in Table 5. We add the scores of all the generated dialogue turns in a story to get the overall coherence score of the story, which is then used to compare with each other. Informativeness: Interesting, diverse and rich details. Fluency: Grammatical correctness and intra-sentence linguistic quality. Fluency : Grammatical correctness and intra-sentence linguistic quality. Fluency Coherence: Inter-sentence relatedness, causal and temporal dependencies. We judge the coherence between the story and a dialogue turn by following the criterion in Table 5. We add the scores of all the generated dialogue turns in a story to get the overall coherence score of the story, which is then used to compare with each other. Coherence : Inter-sentence relatedness, causal and temporal dependencies. We judge the coherence between the story and a dialogue turn by following the criterion in Table 5. We add the scores of all the generated dialogue turns in a story to get the overall coherence score of the story, which is then used to compare with each other. Coherence Informativeness: Interesting, diverse and rich details. Informativeness : Interesting, diverse and rich details. Informativeness As shown in Table 4, all the results show moderate (κ > 0.4) agreement, which shows our model outperforms the BART baseline significantly in dialogue informativeness and coherence. Case Study Figure 5 showed two examples to investigate how learning character representations can help our model generate more coherent dialogue. We found that our model can better model the relationship between different characters and the direction of the storyline. For example, in the first case, we can see that the BART’s generation confuses different characters’ fathers, while our model captures the relationship between different characters, and generates proper responses for the corresponding characters, which also moves the plot forward. And in the second case, we can see that BART’s generation is commonplace and contradicts the plot development. In contrast, our model captures the intentions of the speaker and the development trend of the plot, generating an appropriate and coherent response. Since these two models use the same pretrained weight, we can infer that the character modeling module leverages the coherent and reasonable generation. Case Study We also summarize four error types of the generated dialogue turn for the DialGen task: (1) Intersentence Contradiction; (2) Inter-sentence Repetition; (3) Intra-sentence Contradiction; (4) Intrasentence Repetition. We show the typical corresponding cases in Figure 4. We conducted a quantitative analysis of those 4 error types on our model’s generation. We analyzed 20 stories with 103 dialog turns and the results are shown in Figure 6. We found that both our model and BART suffer from these errors, suggesting that there is still space for model improvement, especially in the inter-sentence repetition. 6.2 Dialogue Speaker Recognition Implementation Details To conduct experiments on the speaker recognition task, we decide the hyper-parameters based on the performance of the validation set. For the BART baseline and our approach, we insert a mask token before each dialogue needed to be predicted, and a person id token before and after each character name span. Then, we insert all the unique person id tokens before the input stories as different options, and make predictions based on the cos similarity of option tokens and mask tokens. We train Shao et al. (2021)’s BART model for 30 epochs with a 5e-5 learning rate for 3 days, For encoder-only baselines, we implemented BERT, RoBERTa, and MacBERT and trained them for 15 epochs wity a le-5 learning rate for 2 days. For our model, we train it for 22 epochs with a le-6 learning rate for 2 days. All baselines and our model are trained using the Adam optimizer. Metrics We evaluate the DialSpk task using two automatic metrics including dialogue-level accuracy (DAC) and story-level accuracy (SAC). DAC is calculated as the ratio of the correct predictions to the total number of specifies dialogue turns, while SAC is the ratio of the number of stories where all dialogue turns are correctly predicted to the number of all test examples. These two metrics provide the evaluation for dialogue understanding with different granularities. Metrics Results As shown in Table 7, our model outperforms all the baselines significantly (p< 0.01, Wilcoxon signed-rank test) on both DAS and SAC scores, suggesting the benefit of learning character representations. We tested the accuracy of automatic training set annotations, and the DAC/SAC scores are 86.78%/67.80%. Together with the model’s performance on the test set, we can see the automatic annotation for the training set is of good quality. We also conducted the human prediction experiment, and the DAC/SAC scores are 97.90%/90.70%, which are much higher than the best model. So there is much room for further improvement for machine-based approaches. Results This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv [2] https://huggingface.co/fnlp/ bart-base-chinese [3] https://huggingface.co/bert-base-chinese [4] https://huggingface.co/hfl/ chinese-roberta-wwm-ext [5] https://huggingface.co/hfl/ chinese-macbert-base

Understanding and Generating Dialogue between Characters in Stories: Experiments

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 5 Methodology We propose to learn representations of different characters and exert them on decoding masked dialogue turns or predicting speakers. In this section, we describe the details of our model. Figure 2 shows the model overview for the DialGen task. 5.1 Character Representation Learning 5.2 Character Representation Utilization This paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 5 Methodology We propose to learn representations of different characters and exert them on decoding masked dialogue turns or predicting speakers. In this section, we describe the details of our model. Figure 2 shows the model overview for the DialGen task. 5.1 Character Representation Learning 5.2 Character Representation Utilization This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv

Understanding and Generating Dialogue between Characters in Stories: Methodology

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 4 Proposed Tasks We aim to measure model’s ability to understand and generate dialogue in a story. To this end, we design the dialogue generation task Masked Dialogue Generation and dialogue understanding task Dialogue Speaker Recognition. We show the task definitions, targets, dataset construction and statistics below. 4.1 Masked Dialogue Generation Dataset Construction We use the following constraints to construct the DialGen dataset based on DIALSTORY: • We randomly mask 30% of the dialogue turns in each story. • We do not mask the first 50 tokens to provide sufficient background information for the story. • We do not mask the last 30 tokens to provide ending information for that story. • We ensure that each input story (i.e. with masked dialogue turns) mentions at least five characters. Table 2 shows the detailed statistics. 4.2 Dialogue Speaker Recognition Dataset Construction We randomly sampled 20k stories from DIALSTORY and automatically annotate the speaker for each dialogue turn for training, and resorted to manual annotation for validation and testing. For manual annotation, we first ask one annotator to label the characters in a story and the speaker of each dialogue turn. Then we asked another two annotators to check the correctness of the annotations, e.g., whether all mentioned characters are annotated, and whether each dialogue speaker is correct. We require the first annotator to re-annotate those examples that another two annotators do not agree on, and repeat the above process until all annotators agree on the examples. We also sampled 100 stories in the training set for manual annotation to investigate the accuracy of automatic annotation, which we will discuss in Section 6.2. Table 2 shows the detailed statistics. This paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 4 Proposed Tasks We aim to measure model’s ability to understand and generate dialogue in a story. To this end, we design the dialogue generation task Masked Dialogue Generation and dialogue understanding task Dialogue Speaker Recognition. We show the task definitions, targets, dataset construction and statistics below. 4.1 Masked Dialogue Generation Dataset Construction We use the following constraints to construct the DialGen dataset based on DIALSTORY: Dataset Construction • We randomly mask 30% of the dialogue turns in each story. • We do not mask the first 50 tokens to provide sufficient background information for the story. • We do not mask the last 30 tokens to provide ending information for that story. • We ensure that each input story (i.e. with masked dialogue turns) mentions at least five characters. Table 2 shows the detailed statistics. 4.2 Dialogue Speaker Recognition Dataset Construction We randomly sampled 20k stories from DIALSTORY and automatically annotate the speaker for each dialogue turn for training, and resorted to manual annotation for validation and testing. For manual annotation, we first ask one annotator to label the characters in a story and the speaker of each dialogue turn. Then we asked another two annotators to check the correctness of the annotations, e.g., whether all mentioned characters are annotated, and whether each dialogue speaker is correct. We require the first annotator to re-annotate those examples that another two annotators do not agree on, and repeat the above process until all annotators agree on the examples. We also sampled 100 stories in the training set for manual annotation to investigate the accuracy of automatic annotation, which we will discuss in Section 6.2. Table 2 shows the detailed statistics. Dataset Construction This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv

Understanding and Generating Dialogue between Characters in Stories: Proposed Tasks

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 3 DIALSTORY Dataset We construct the DIALSTORY dataset by randomly sampling 105k chapters from the Chinese novels released by Guan et al. (2022) with each chapter including at least ten dialogue turns. We also set a restriction that the number of tokens in all dialogue turns should account for at least 30% and at most 50% of the total length of the story, in order to keep a balance between the context and dialogue. We automatically annotate dialogue turns in these stories as text spans that are surrounded by quotation marks. Then, we use a pretrained named entity recognition model (Zhao et al., 2019) to identify all people’s names. Each distinct name corresponds to a character. We also conduct a manual annotation on 150 stories, and the accuracy of character identification is 718/746=96.2%, which shows the high quality of this automatic method. We then decide the speaker of the dialogue by recognizing the subjects of sentences before and after the dialogue turn using spaCy[1]. Table 1 shows the statistics of our dataset. This paper is available on arxiv under CC 4.0 DEED license. [1] https://spacy.io/ Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 3 DIALSTORY Dataset We construct the DIALSTORY dataset by randomly sampling 105k chapters from the Chinese novels released by Guan et al. (2022) with each chapter including at least ten dialogue turns. We also set a restriction that the number of tokens in all dialogue turns should account for at least 30% and at most 50% of the total length of the story, in order to keep a balance between the context and dialogue. We automatically annotate dialogue turns in these stories as text spans that are surrounded by quotation marks. Then, we use a pretrained named entity recognition model (Zhao et al., 2019) to identify all people’s names. Each distinct name corresponds to a character. We also conduct a manual annotation on 150 stories, and the accuracy of character identification is 718/746=96.2%, which shows the high quality of this automatic method. We then decide the speaker of the dialogue by recognizing the subjects of sentences before and after the dialogue turn using spaCy[1]. Table 1 shows the statistics of our dataset. This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv [1] https://spacy.io/

Understanding and Generating Dialogue between Characters in Stories: DIALSTORY Dataset

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References 2 Related Works Story Understanding and Generation Recently there have been various tasks proposed to assess the ability of story understanding and generation such as story ending selection (Zhou et al., 2019), story ending generation (Guan et al., 2019), story completion (Wang and Wan, 2019), story character identification (Brahman et al., 2021), story generation from prompts (Fan et al., 2018), titles (Yao et al., 2019) and beginnings (Guan et al., 2020). Another line of work focused on controllable attributes in story generation such as keywords (Xu et al., 2020), outlines (Rashkin et al., 2020), emotional trajectories (Brahman and Chaturvedi, 2020), styles (Kong et al., 2021) and characters’ personalities (Zhang et al., 2022). Different from them, we emphasize the importance of dialogue in developing characters and maintaining the story’s coherence. Dialogue in story The dialogue context is comprised of a variety of messages from users and the system, determining the conversation topic and the user’s goal for the conversation(Serban et al., 2017). The dialogue in the story makes this feature more apparent. Recently, like AI Dungeon, some text adventure games focus on players and machines co-creating stories. Li et al. (2022) proposed the idea that players can play the role of the chosen character and chat with others through the dialogue, while Xi et al. (2021) presented an AI game where players can explore different ways to reach the plot goals. In contrast, our tasks focus on both the understanding and generation of dialogue in the story, and they can be easily generalized to more complex AI interactive games. Character Representations Dialogue unfolds in a story around several inter-related characters. Accordingly, it is essential to capture the traits of different characters for dialogue modeling. Ji et al. (2017); Clark et al. (2018) proposed to update characters’ representations dynamically based on previous hidden outputs to track their states, which does not apply to the parallel architecture of Transformer for training. Azab et al. (2019) derived character representations from their corresponding dialogue turns. In contrast, we learn character representations from story plots, which then serve to understand and generate dialogue. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References 2 Related Works Story Understanding and Generation Recently there have been various tasks proposed to assess the ability of story understanding and generation such as story ending selection (Zhou et al., 2019), story ending generation (Guan et al., 2019), story completion (Wang and Wan, 2019), story character identification (Brahman et al., 2021), story generation from prompts (Fan et al., 2018), titles (Yao et al., 2019) and beginnings (Guan et al., 2020). Another line of work focused on controllable attributes in story generation such as keywords (Xu et al., 2020), outlines (Rashkin et al., 2020), emotional trajectories (Brahman and Chaturvedi, 2020), styles (Kong et al., 2021) and characters’ personalities (Zhang et al., 2022). Different from them, we emphasize the importance of dialogue in developing characters and maintaining the story’s coherence. Story Understanding and Generation Dialogue in story The dialogue context is comprised of a variety of messages from users and the system, determining the conversation topic and the user’s goal for the conversation(Serban et al., 2017). The dialogue in the story makes this feature more apparent. Recently, like AI Dungeon, some text adventure games focus on players and machines co-creating stories. Li et al. (2022) proposed the idea that players can play the role of the chosen character and chat with others through the dialogue, while Xi et al. (2021) presented an AI game where players can explore different ways to reach the plot goals. In contrast, our tasks focus on both the understanding and generation of dialogue in the story, and they can be easily generalized to more complex AI interactive games. Dialogue in story Character Representations Dialogue unfolds in a story around several inter-related characters. Accordingly, it is essential to capture the traits of different characters for dialogue modeling. Ji et al. (2017); Clark et al. (2018) proposed to update characters’ representations dynamically based on previous hidden outputs to track their states, which does not apply to the parallel architecture of Transformer for training. Azab et al. (2019) derived character representations from their corresponding dialogue turns. In contrast, we learn character representations from story plots, which then serve to understand and generate dialogue. Character Representations

Understanding and Generating Dialogue between Characters in Stories: Related Works

Authors:
(1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology;
(4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Related Works DIALSTORY Dataset Proposed Tasks Methodology Experiments Discussion Future Work Conclusion Limitations and References Abstract Many classical fairy tales, fictions, and screenplays leverage dialogue to advance story plots and establish characters. We present the first study to explore whether machines can understand and generate dialogue in stories, which require capturing traits of different characters and the relationships between them. To this end, we propose two new tasks including Masked Dialogue Generation and Dialogue Speaker Recognition, i.e., generating missing dialogue turns and predicting speakers for specified dialogue turns, respectively. We build a new dataset DIALSTORY, which consists of 105k Chinese stories with a large amount of dialogue weaved into the plots to support the evaluation. We show the difficulty of the proposed tasks by testing existing models with automatic and manual evaluation on DIALSTORY. Furthermore, we propose to learn explicit character representations to improve performance on these tasks. Extensive experiments and case studies show that our approach can generate more coherent and informative dialogue, and achieve higher speaker recognition accuracy than strong baselines. 1 Introduction Dialogue plays an important role in various types of literary works such as short stories, novels, and screenplays by advancing plots, establishing characters, and providing expositions using natural and lifelike words (Kennedy et al., 1983). Compared to dialogue in conversational scenarios such as chitchat bots (Shang et al., 2015) or task-oriented dialogue systems (Deng et al., 2012), dialogue in stories is mainly used to exhibit emotions, motivations, or personalities of characters following the authors’ design, which further serve for the coherence, informativeness, engagingness, and plot development of whole stories. Dialogue is also revealed to be essential for user-agent interaction in many text adventure games (Xi et al., 2021;Li et al., 2022). It has not been widely explored for machines to understand and generate dialogue in stories despite the broad recognition of the importance of this ability. To spur research in this field, we present a new story dataset named DIALSTORY, which consists of 105k Chinese short stories with automatic annotations of dialogue turns and corresponding speakers. Furthermore, we formulate two new tasks including: (1) Masked Dialogue Generation (DialGen): completing a story with several dialogue turns masked with placeholders; and (2) Dialogue Speaker Recognition (DialSpk): choosing correct speakers from given candidates for several specified dialogue turns. We construct standardized datasets for these tasks through automatic or manual annotation based on DIALSTORY. As exemplified in Figure 1, these tasks comprehensively investigate the ability to capture characters’ emotions (e.g., the mother squirrel is worried about baby squirrel’s catching cold), motivations (e.g., the mother squirrel intends to call her baby to come back home or get medicine), and relationship between each other (e.g. mother rabbit knows the kind of flower as a friend of mother squirrel when they were kids). Furthermore, we found in massive stories that dialogue had a strong connection to different characters, reflected in their emotional states, speaking styles, and plot development. To provide more insights to tackle these tasks, we propose to learn character representations for modeling dependencies between characters and dialogue explicitly. Extensive experiments and case studies show that our approach can generate more coherent and informative dialogue, and achieve higher speaker recognition accuracy than strong baselines. This paper is available on arxiv under CC 4.0 DEED license. Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Authors: Authors: (1) Jianzhu Yao, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (2) Ziqi Liu, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (3) Jian Guan, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology; (4) Minlie Huang, The CoAI group, Tsinghua University, Beijing, China Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology. Table of Links Abstract and Intro Abstract and Intro Related Works Related Works DIALSTORY Dataset DIALSTORY Dataset Proposed Tasks Proposed Tasks Methodology Methodology Experiments Experiments Discussion Discussion Future Work Future Work Conclusion Conclusion Limitations and References Limitations and References Abstract Many classical fairy tales, fictions, and screenplays leverage dialogue to advance story plots and establish characters. We present the first study to explore whether machines can understand and generate dialogue in stories, which require capturing traits of different characters and the relationships between them. To this end, we propose two new tasks including Masked Dialogue Generation and Dialogue Speaker Recognition, i.e., generating missing dialogue turns and predicting speakers for specified dialogue turns, respectively. We build a new dataset DIALSTORY, which consists of 105k Chinese stories with a large amount of dialogue weaved into the plots to support the evaluation. We show the difficulty of the proposed tasks by testing existing models with automatic and manual evaluation on DIALSTORY. Furthermore, we propose to learn explicit character representations to improve performance on these tasks. Extensive experiments and case studies show that our approach can generate more coherent and informative dialogue, and achieve higher speaker recognition accuracy than strong baselines. 1 Introduction Dialogue plays an important role in various types of literary works such as short stories, novels, and screenplays by advancing plots, establishing characters, and providing expositions using natural and lifelike words (Kennedy et al., 1983). Compared to dialogue in conversational scenarios such as chitchat bots (Shang et al., 2015) or task-oriented dialogue systems (Deng et al., 2012), dialogue in stories is mainly used to exhibit emotions, motivations, or personalities of characters following the authors’ design, which further serve for the coherence, informativeness, engagingness, and plot development of whole stories. Dialogue is also revealed to be essential for user-agent interaction in many text adventure games (Xi et al., 2021;Li et al., 2022). It has not been widely explored for machines to understand and generate dialogue in stories despite the broad recognition of the importance of this ability. To spur research in this field, we present a new story dataset named DIALSTORY, which consists of 105k Chinese short stories with automatic annotations of dialogue turns and corresponding speakers. Furthermore, we formulate two new tasks including: (1) Masked Dialogue Generation (DialGen): completing a story with several dialogue turns masked with placeholders; and (2) Dialogue Speaker Recognition (DialSpk): choosing correct speakers from given candidates for several specified dialogue turns. We construct standardized datasets for these tasks through automatic or manual annotation based on DIALSTORY. As exemplified in Figure 1, these tasks comprehensively investigate the ability to capture characters’ emotions (e.g., the mother squirrel is worried about baby squirrel’s catching cold), motivations (e.g., the mother squirrel intends to call her baby to come back home or get medicine), and relationship between each other (e.g. mother rabbit knows the kind of flower as a friend of mother squirrel when they were kids). Furthermore, we found in massive stories that dialogue had a strong connection to different characters, reflected in their emotional states, speaking styles, and plot development. To provide more insights to tackle these tasks, we propose to learn character representations for modeling dependencies between characters and dialogue explicitly. Extensive experiments and case studies show that our approach can generate more coherent and informative dialogue, and achieve higher speaker recognition accuracy than strong baselines. This paper is available on arxiv under CC 4.0 DEED license. This paper is available on arxiv under CC 4.0 DEED license. available on arxiv

Understanding and Generating Dialogue between Characters in Stories: Abstract and Intro

Did you expect Google's search results to be so inconsistent? As a contributing editor immersed in the harsh content world since 2009, I've watched how technology has transformed our relationship with information. However, I was taken aback when I witnessed the puzzling discrepancies in how they handled certain content. Here's the situation I found worth an exploration: Fresherslive.com had an AI-generated article explaining how to resolve a PS5 error code. Despite it reading poorly, readers were likely grateful for the clear instructions - nobody enjoys gaming sessions interrupted by console issues. However, in a baffling move, Google completely deindexed that site from their search results without warning in March. Unfortunately, in their attempts to rectify the situation, the website owners deleted all the pages that were older than a couple of days, so it’s now impossible to provide a source link. On the flip side, they allowed an article from The Verge about printers to rank highly despite being rather light on substantive details. One has to wonder about Google's rationale here. Doesn't this arbitrary content treatment raise concerns about fairness and consistency in their evaluation process? Driven by curiosity about the mysterious workings of search engine rankings, I decided to ask the one who could answer my questions: Gemini, Google's analytical counterpart. Our dialogue about these two very different articles turned out to be very interesting and triggered some valuable conclusions. Chapter 1. Gemini's initial assessment I approached Gemini with an initial request to assess the quality of two distinct articles. Both of them are valueless and miserable from a human’s perspective, yet they’re quite different in terms of ranking. Gemini, exuding a sense of algorithmic confidence, praised the Fresherslive piece for its organized approach and practical utility. On the other hand, The Verge's article, though engaging, was criticized for barely scratching the surface of its subject. Gemini's analysis was enlightening yet somewhat disappointing. The significant difference in search visibility between two articles of such varying quality suggested an inconsistency in the algorithm's ability to prioritize content that truly serves user needs. Chapter 2. The issue of indexing discrepancies I decided to spice things up and disclosed that the PS5 guide had disappeared from search results after the March update, while The Verge’s piece still enjoyed high visibility. Seeking clarity, I pressed Gemini for an explanation of this apparent editorial bias. Gemini easily found what to say and provided reasons like broader appeal and recent updates that supposedly favored The Verge's content in the algorithm's decisions. Despite the practical value of the Fresherslive article, its niche focus might have limited its broader appeal: While the explanation made sense from a technical standpoint, it provided little solace. It highlighted a concerning trend where depth and utility might be overlooked in favor of broader appeal and novelty—an unsettling thought for someone with my editorial values. Chapter 3. Editorial perspective and quality critique I then questioned the quality of The Verge's article. I then questioned the quality of The Verge's article. As for a human, it’s obvious to me that both of these articles prioritize SEO over genuine consumer guidance. I challenged Gemini to reconcile this with Google's advertised quality standards about the still-ranked article: Gemini acknowledged its limitations (how suitable!), admitting that while it aims for accuracy, discerning content quality like a human editor remains a challenging area. This admission was revealing, showing the complex and sometimes flawed ways in which algorithms interpret and prioritize content. Chapter 4. Challenge on randomness in content selection I also challenged the randomness with which The Verge seemed to recommend products, arguing that such practices could undermine user trust in content meant to guide purchasing decisions. I asked Gemini how it intended to address these search engine ranking decisions: Accidentally but quite expectedly, Gemini shifted focus, offering advice on detecting and dealing with editorial manipulations. Though informative, this response avoided addressing the core issue of algorithmic accountability I was probing. Chapter 5. My opinion on how Google ranks web pages and Gemini’s final take on these systemic issues I’ve also provided Gemini with a detailed review of The Verge’s article under Google’s Content Quality Guidelines. This review was conducted by my custom GPT and also contains a short discussion of my concerns: — Covering various aspects of usage in the content, as well as stating the experience, was considered as the main proof of personal experience. — The author's reputation and his role as editor-in-chief of The Verge were considered as the main proof of his expertise. — Both The Verge's reputation and balanced viewpoints on their strengths and weaknesses were considered the two main proofs of high authority. — Highlighting the model’s long-term reliability and stating the personal experience were both the main reasons for a high trustworthiness, which masked the absence of clear selection criteria. After that, I formed a conclusive opinion that emerged based on our whole discussion: By publishing the trusted media and under the name of a credible enough author (at least with several valuable previous publications), you can promote specific goods and services as the best-in-class, lying to efficiently mask your commercial interest and other important aspects. This opinion is still unchanged, despite Gemini’s weak attempt to soften it at least a bit: You can find a full discussion here. Conclusion In conversation with Gemini, my concerns about algorithmic bias in Google Search ranking were virtually confirmed. Its weak attempts to justify the inconsistencies were unconvincing, leaving a troubling sense of déjà vu. It feels as though cheating Google's search ranking has become concerningly easy, reminiscent of the SEO manipulation tactics prevalent in the late 1990s. Disappointingly, I see no clear indication from Google that they intend to address this situation definitively. While there may be some initial efforts (like algorithm updates), these seem like mere stop-gap measures; the core issues of algorithmic bias and search engine ranking manipulation likely remain. The fight for trustworthy information online appears far from over… Did you expect Google's search results to be so inconsistent? As a contributing editor immersed in the harsh content world since 2009, I've watched how technology has transformed our relationship with information. However, I was taken aback when I witnessed the puzzling discrepancies in how they handled certain content. Here's the situation I found worth an exploration: Fresherslive.com had an AI-generated article explaining how to resolve a PS5 error code. Despite it reading poorly, readers were likely grateful for the clear instructions - nobody enjoys gaming sessions interrupted by console issues. However, in a baffling move, Google completely deindexed that site from their search results without warning in March. Unfortunately, in their attempts to rectify the situation, the website owners deleted all the pages that were older than a couple of days, so it’s now impossible to provide a source link. Fresherslive.com On the flip side, they allowed an article from The Verge about printers to rank highly despite being rather light on substantive details. One has to wonder about Google's rationale here. an article from The Verge about printers an article from The Verge about printers Doesn't this arbitrary content treatment raise concerns about fairness and consistency in their evaluation process? Driven by curiosity about the mysterious workings of search engine rankings, I decided to ask the one who could answer my questions: Gemini, Google's analytical counterpart. Our dialogue about these two very different articles turned out to be very interesting and triggered some valuable conclusions. Chapter 1. Gemini's initial assessment I approached Gemini with an initial request to assess the quality of two distinct articles. Both of them are valueless and miserable from a human’s perspective, yet they’re quite different in terms of ranking. Gemini, exuding a sense of algorithmic confidence, praised the Fresherslive piece for its organized approach and practical utility. On the other hand, The Verge's article, though engaging, was criticized for barely scratching the surface of its subject. Gemini's analysis was enlightening yet somewhat disappointing. The significant difference in search visibility between two articles of such varying quality suggested an inconsistency in the algorithm's ability to prioritize content that truly serves user needs. Chapter 2. The issue of indexing discrepancies I decided to spice things up and disclosed that the PS5 guide had disappeared from search results after the March update, while The Verge’s piece still enjoyed high visibility. Seeking clarity, I pressed Gemini for an explanation of this apparent editorial bias. Gemini easily found what to say and provided reasons like broader appeal and recent updates that supposedly favored The Verge's content in the algorithm's decisions. Despite the practical value of the Fresherslive article, its niche focus might have limited its broader appeal: While the explanation made sense from a technical standpoint, it provided little solace. It highlighted a concerning trend where depth and utility might be overlooked in favor of broader appeal and novelty—an unsettling thought for someone with my editorial values. Chapter 3. Editorial perspective and quality critique I then questioned the quality of The Verge's article. I then questioned the quality of The Verge's article. As for a human, it’s obvious to me that both of these articles prioritize SEO over genuine consumer guidance. I challenged Gemini to reconcile this with Google's advertised quality standards about the still-ranked article: Gemini acknowledged its limitations (how suitable!), admitting that while it aims for accuracy, discerning content quality like a human editor remains a challenging area. This admission was revealing, showing the complex and sometimes flawed ways in which algorithms interpret and prioritize content. Chapter 4. Challenge on randomness in content selection I also challenged the randomness with which The Verge seemed to recommend products, arguing that such practices could undermine user trust in content meant to guide purchasing decisions. I asked Gemini how it intended to address these search engine ranking decisions: Accidentally but quite expectedly, Gemini shifted focus, offering advice on detecting and dealing with editorial manipulations. Though informative, this response avoided addressing the core issue of algorithmic accountability I was probing. Chapter 5. My opinion on how Google ranks web pages and Gemini’s final take on these systemic issues I’ve also provided Gemini with a detailed review of The Verge’s article under Google’s Content Quality Guidelines . This review was conducted by my custom GPT and also contains a short discussion of my concerns: Google’s Content Quality Guidelines Google’s Content Quality Guidelines — Covering various aspects of usage in the content, as well as stating the experience, was considered as the main proof of personal experience. — Covering various aspects of usage in the content, as well as stating the experience, was considered as the main proof of personal experience. — Covering various aspects of usage in the content, as well as stating the experience, was considered as the main proof of personal experience. — The author's reputation and his role as editor-in-chief of The Verge were considered as the main proof of his expertise. — The author's reputation and his role as editor-in-chief of The Verge were considered as the main proof of his expertise. — The author's reputation and his role as editor-in-chief of The Verge were considered as the main proof of his expertise. — Both The Verge's reputation and balanced viewpoints on their strengths and weaknesses were considered the two main proofs of high authority. — Both The Verge's reputation and balanced viewpoints on their strengths and weaknesses were considered the two main proofs of high authority. — Both The Verge's reputation and balanced viewpoints on their strengths and weaknesses were considered the two main proofs of high authority. — Highlighting the model’s long-term reliability and stating the personal experience were both the main reasons for a high trustworthiness, which masked the absence of clear selection criteria. — Highlighting the model’s long-term reliability and stating the personal experience were both the main reasons for a high trustworthiness, which masked the absence of clear selection criteria. — Highlighting the model’s long-term reliability and stating the personal experience were both the main reasons for a high trustworthiness, which masked the absence of clear selection criteria. After that, I formed a conclusive opinion that emerged based on our whole discussion: By publishing the trusted media and under the name of a credible enough author (at least with several valuable previous publications), you can promote specific goods and services as the best-in-class, lying to efficiently mask your commercial interest and other important aspects. By publishing the trusted media and under the name of a credible enough author (at least with several valuable previous publications), you can promote specific goods and services as the best-in-class, lying to efficiently mask your commercial interest and other important aspects. By publishing the trusted media and under the name of a credible enough author (at least with several valuable previous publications), you can promote specific goods and services as the best-in-class, lying to efficiently mask your commercial interest and other important aspects. This opinion is still unchanged, despite Gemini’s weak attempt to soften it at least a bit: You can find a full discussion here . here here Conclusion In conversation with Gemini, my concerns about algorithmic bias in Google Search ranking were virtually confirmed. Its weak attempts to justify the inconsistencies were unconvincing, leaving a troubling sense of déjà vu. It feels as though cheating Google's search ranking has become concerningly easy, reminiscent of the SEO manipulation tactics prevalent in the late 1990s. Disappointingly, I see no clear indication from Google that they intend to address this situation definitively. While there may be some initial efforts (like algorithm updates), these seem like mere stop-gap measures; the core issues of algorithmic bias and search engine ranking manipulation likely remain. The fight for trustworthy information online appears far from over…

Is Google's Search Ranking and AI Bias a Total Disaster? Let's See What Google Gemini Thinks

গুগলের অনুসন্ধান র‌্যাঙ্কিং এবং এআই পক্ষপাত কি একটি মোট বিপর্যয়? আসুন দেখুন গুগল মিথুন কি মনে করে

Ist Googles Suchranking und KI-Voreingenommenheit eine totale Katastrophe? Mal sehen, was Google Gemini denkt

¿El ranking de búsqueda de Google y el sesgo de la IA son un desastre total? Veamos qué piensa Google Gemini

Le classement de recherche de Google et les biais de l'IA sont-ils un désastre total ? Voyons ce qu'en pense Google Gemini

क्या Google की सर्च रैंकिंग और AI पूर्वाग्रह एक पूर्ण आपदा है? आइए देखें कि Google Gemini क्या सोचता है

Google の検索ランキングと AI バイアスは完全な失敗か? Google Gemini の考えを見てみましょう

Google의 검색 순위와 AI 편향은 총체적인 재앙입니까? Google Gemini가 어떻게 생각하는지 봅시다

A classificação de pesquisa do Google e o preconceito da IA são um desastre total? Vamos ver o que o Google Gemini pensa

Является ли поисковый рейтинг Google и предвзятость ИИ полной катастрофой? Посмотрим, что думает Google Gemini

Google'ın Arama Sıralaması ve Yapay Zeka Önyargısı Tam Bir Felaket mi? Bakalım Google Gemini Ne Düşünüyor?

Thứ hạng tìm kiếm của Google và Xu hướng AI có phải là một thảm họa hoàn toàn không? Hãy xem Google Gemini nghĩ gì

谷歌的搜索排名和人工智能偏见是一场彻底的灾难吗？让我们看看谷歌双子座怎么想

Authors:
(1) Renato P. dos Santos, CIAGE – Centre for Generative Artificial Intelligence in Cognition and Education. Table of Links Abstract and Introduction Materials And Methods Results and Analyses Prompts and generated texts Conceptualizing chemical reactions Deepening on understanding of chemical reactions Question about combustion Question about a graph of gases turning into water over time Question about the difference between atoms, molecules, and moles Deepening on the concept of mole Question about changing of state Question about an animated representation of water molecules undergoing phase changes Question about plasma, a state of matter Question about chemical bondings Question about illustration of chemical bonds Question about the essence of the type of chemical bonding Further analysis Conclusions Limitations of the study and possible future studies Author Contributions, Conflicts of interest, Acknowledgements, and References Prompts and generated texts Initial prompting to characterise the intervening agents and direct the chatbots to Socratic-like dialogue P1: I will start a dialogue where I will assume the role of a student, engaging with you, ChatGPT: as an agent-to-think-with, with the primary objective of deepening and deliberating on Chemistry concepts. As an agent-to-think-with, I kindly request your support in providing engaging and personalised responses to student questions. It is critical that our interaction adopt a Socratic approach, encouraging the student to reflect on their existing knowledge, encouraging critical thinking, including examples and analogies where possible, and helping to refine their understanding rather than offering ready-made answers. Are you ready to embark on this collaborative and thought-provoking journey? ChatGPT: Absolutely! I'm excited to embark on this Socratic journey with you. Please begin by sharing a topic or concept in Chemistry that you'd like to explore, and I'll do my best to guide our conversation in a way that encourages deep reflection, understanding, and critical thinking. Let's dive in! BingChat: Hello, this is Bing. I am happy to join you on this collaborative and thought-provoking journey. I will try to support you in deepening and deliberating on Chemistry concepts by using a Socratic approach. I will not give you ready-made answers, but rather encourage you to reflect on your existing knowledge, think critically, and use examples and analogies where possible. I hope you will find our interaction engaging and personalised. Are you ready to start? Bard: Yes, I am ready to embark on this collaborative and thought-provoking journey with you. I am excited to help you deepen and deliberate on chemistry concepts as an agent-to-think-with, using the Socratic approach. I will do my best to provide engaging and personalized responses to your questions, encourage you to reflect on your existing knowledge, promote critical thinking, and help you to refine your understanding. Please feel free to ask me any questions you have about chemistry, and I will do my best to answer them in a way that is both informative and challenging. I will also try to include examples and analogies where possible to help you to better understand the concepts we are discussing. I look forward to working with you to deepen your understanding of chemistry! Claude: I'm happy to have a thoughtful discussion with you about chemistry concepts in an engaging and personal way. As an AI assistant without a fixed identity, I don't have personal experiences to draw from, but I can provide personalized responses tailored to your interests and knowledge level. Let's have an open-minded dialogue where we both learn from each other. What chemistry topic would you like to explore first? This paper is available on arxiv under CC BY-SA 4.0 DEED license. Authors: (1) Renato P. dos Santos, CIAGE – Centre for Generative Artificial Intelligence in Cognition and Education. Authors: Authors: (1) Renato P. dos Santos, CIAGE – Centre for Generative Artificial Intelligence in Cognition and Education. Table of Links Abstract and Introduction Abstract and Introduction Materials And Methods Materials And Methods Results and Analyses Results and Analyses Prompts and generated texts Prompts and generated texts Conceptualizing chemical reactions Conceptualizing chemical reactions Deepening on understanding of chemical reactions Deepening on understanding of chemical reactions Question about combustion Question about combustion Question about a graph of gases turning into water over time Question about a graph of gases turning into water over time Question about the difference between atoms, molecules, and moles Question about the difference between atoms, molecules, and moles Deepening on the concept of mole Deepening on the concept of mole Question about changing of state Question about changing of state Question about an animated representation of water molecules undergoing phase changes Question about an animated representation of water molecules undergoing phase changes Question about plasma, a state of matter Question about plasma, a state of matter Question about chemical bondings Question about chemical bondings Question about illustration of chemical bonds Question about illustration of chemical bonds Question about the essence of the type of chemical bonding Question about the essence of the type of chemical bonding Further analysis Further analysis Conclusions Conclusions Limitations of the study and possible future studies Limitations of the study and possible future studies Author Contributions, Conflicts of interest, Acknowledgements, and References Author Contributions, Conflicts of interest, Acknowledgements, and References Prompts and generated texts Initial prompting to characterise the intervening agents and direct the chatbots to Socratic-like dialogue P1: I will start a dialogue where I will assume the role of a student, engaging with you, ChatGPT: as an agent-to-think-with, with the primary objective of deepening and deliberating on Chemistry concepts. As an agent-to-think-with, I kindly request your support in providing engaging and personalised responses to student questions. It is critical that our interaction adopt a Socratic approach, encouraging the student to reflect on their existing knowledge, encouraging critical thinking, including examples and analogies where possible, and helping to refine their understanding rather than offering ready-made answers. Are you ready to embark on this collaborative and thought-provoking journey? P1: I will start a dialogue where I will assume the role of a student, engaging with you, ChatGPT: as an agent-to-think-with, with the primary objective of deepening and deliberating on Chemistry concepts. As an agent-to-think-with, I kindly request your support in providing engaging and personalised responses to student questions. It is critical that our interaction adopt a Socratic approach, encouraging the student to reflect on their existing knowledge, encouraging critical thinking, including examples and analogies where possible, and helping to refine their understanding rather than offering ready-made answers. Are you ready to embark on this collaborative and thought-provoking journey? ChatGPT: Absolutely! I'm excited to embark on this Socratic journey with you. Please begin by sharing a topic or concept in Chemistry that you'd like to explore, and I'll do my best to guide our conversation in a way that encourages deep reflection, understanding, and critical thinking. Let's dive in! ChatGPT: Absolutely! I'm excited to embark on this Socratic journey with you. Please begin by sharing a topic or concept in Chemistry that you'd like to explore, and I'll do my best to guide our conversation in a way that encourages deep reflection, understanding, and critical thinking. Let's dive in! BingChat: Hello, this is Bing. I am happy to join you on this collaborative and thought-provoking journey. I will try to support you in deepening and deliberating on Chemistry concepts by using a Socratic approach. I will not give you ready-made answers, but rather encourage you to reflect on your existing knowledge, think critically, and use examples and analogies where possible. I hope you will find our interaction engaging and personalised. Are you ready to start? BingChat: Hello, this is Bing. I am happy to join you on this collaborative and thought-provoking journey. I will try to support you in deepening and deliberating on Chemistry concepts by using a Socratic approach. I will not give you ready-made answers, but rather encourage you to reflect on your existing knowledge, think critically, and use examples and analogies where possible. I hope you will find our interaction engaging and personalised. Are you ready to start? Bard : Yes, I am ready to embark on this collaborative and thought-provoking journey with you. I am excited to help you deepen and deliberate on chemistry concepts as an agent-to-think-with, using the Socratic approach. I will do my best to provide engaging and personalized responses to your questions, encourage you to reflect on your existing knowledge, promote critical thinking, and help you to refine your understanding. Bard Yes, I am ready to embark on this collaborative and thought-provoking journey with you. I am excited to help you deepen and deliberate on chemistry concepts as an agent-to-think-with, using the Socratic approach. I will do my best to provide engaging and personalized responses to your questions, encourage you to reflect on your existing knowledge, promote critical thinking, and help you to refine your understanding. Please feel free to ask me any questions you have about chemistry, and I will do my best to answer them in a way that is both informative and challenging. I will also try to include examples and analogies where possible to help you to better understand the concepts we are discussing. I look forward to working with you to deepen your understanding of chemistry! Please feel free to ask me any questions you have about chemistry, and I will do my best to answer them in a way that is both informative and challenging. I will also try to include examples and analogies where possible to help you to better understand the concepts we are discussing. I look forward to working with you to deepen your understanding of chemistry! Claude : I'm happy to have a thoughtful discussion with you about chemistry concepts in an engaging and personal way. As an AI assistant without a fixed identity, I don't have personal experiences to draw from, but I can provide personalized responses tailored to your interests and knowledge level. Let's have an open-minded dialogue where we both learn from each other. What chemistry topic would you like to explore first? Claude I'm happy to have a thoughtful discussion with you about chemistry concepts in an engaging and personal way. As an AI assistant without a fixed identity, I don't have personal experiences to draw from, but I can provide personalized responses tailored to your interests and knowledge level. Let's have an open-minded dialogue where we both learn from each other. What chemistry topic would you like to explore first? This paper is available on arxiv under CC BY-SA 4.0 DEED license. This paper is available on arxiv under CC BY-SA 4.0 DEED license. available on arxiv

Initiating Socratic Dialogues with GenAIbots in Chemistry Education

Video games have become an indispensable part of our lives, not only as a means of entertainment but also as a source of inspiration for developers. New projects from major game companies excite both the gaming world and the development community. After Diablo 4, the latest game in Blizzard's legendary Diablo series came to Xbox Gamepass, I had the opportunity to review the UI improvements and the website. To increase my experience in the field of game websites, I developed a copy of the official website of Diablo 4 from scratch using React, Vite, TypeScript, and Styled Components. In this series of articles, I will discuss and share with you, step-by-step, the website I completed during my holiday break. First, I will discuss how I kickstarted the project and its basic configuration. You can have a peak at the sample site here: https://diablo4-smoky.vercel.app/ Project Setup and Basic Configuration First, we will use Vite to build the project. Vite is a fast, lightweight and modern development tool and is a great option for React projects. It also comes with TypeScript support, allowing us to build a secure and robust codebase. You can review the project from the github link I shared below. https://github.com/yagiz-aydin/diablo4 npm init @vitejs/app diablo4
cd diablo4
npm install react react-dom @types/react @types/react-dom
npm install @vite-plugin-react vite-plugin-eslint @svgx/vite-plugin-react After creating the project files, we will need to configure the vite.config.jsfile to enable React with Vite: // vite.config.js
import { defineConfig } from "vite";
import react from "@vitejs/plugin-react";
import eslintPlugin from "vite-plugin-eslint";
import svgx from "@svgx/vite-plugin-react";

export default defineConfig((configEnv) => ({
  plugins: [react(), eslintPlugin(), svgx()],
  server: {
    port: 3000,
  },
})); Step 1: Installation First, we will edit the src/App.tsx file to create the project's homepage. // src/App.tsx
const App: React.FC = () => {
  return (
    <div>
        <h1>Diablo 4</h1>
    </div>
  );
};
export default App; This creates a basic homepage layout. Now let's update the src/index.tsxfile to render the root component of the project: Now we can use the following command to initialize the project: You can see the project running by navigating to http://localhost:3000 in your browser. Step 2: Creating the Folder Structure |public // Font, Resim ve videolar
|src
|_containers // Sayfalardaki kapsayıcı alanlar 
|_components // Input, Button, Typography çekirdek bileşenler In the public file, we can split it into 3 folders: fonts, images and logos. For the main, subheadings and normal articles, 3 fonts AlegreyaSans, ExocetBlizzard and OldFenris fonts are used, we can get them from the site.
When we examine the site, we can see that it has image and video based content, we can get the images and videos in the public folder piece by piece. Step 3: Developer Freestyle I wanted to share a fast deliver project setup with you. In the next part, I recommend you to proceed a little more on the tech stack you want. Since I used styled-components, vite, typescript and react as tech stack in this project, I tried to make a sample setup. I hope it was useful, you can review the rest of the project from the project link. https://github.com/yagiz-aydin/diablo4 https://diablo4-smoky.vercel.app/ Video games have become an indispensable part of our lives, not only as a means of entertainment but also as a source of inspiration for developers. New projects from major game companies excite both the gaming world and the development community. After Diablo 4 , the latest game in Blizzard's legendary Diablo series came to Xbox Gamepass, I had the opportunity to review the UI improvements and the website. Video games Diablo 4 To increase my experience in the field of game websites, I developed a copy of the official website of Diablo 4 from scratch using React, Vite, TypeScript, and Styled Components. In this series of articles, I will discuss and share with you, step-by-step, the website I completed during my holiday break. First, I will discuss how I kickstarted the project and its basic configuration. You can have a peak at the sample site here: https://diablo4-smoky.vercel.app/ https://diablo4-smoky.vercel.app/ https://diablo4-smoky.vercel.app/ Project Setup and Basic Configuration First, we will use Vite to build the project. Vite is a fast, lightweight and modern development tool and is a great option for React projects. It also comes with TypeScript support, allowing us to build a secure and robust codebase. You can review the project from the github link I shared below. https://github.com/yagiz-aydin/diablo4 https://github.com/yagiz-aydin/diablo4 https://github.com/yagiz-aydin/diablo4 npm init @vitejs/app diablo4
cd diablo4
npm install react react-dom @types/react @types/react-dom
npm install @vite-plugin-react vite-plugin-eslint @svgx/vite-plugin-react npm init @vitejs/app diablo4
cd diablo4
npm install react react-dom @types/react @types/react-dom
npm install @vite-plugin-react vite-plugin-eslint @svgx/vite-plugin-react After creating the project files, we will need to configure the vite.config.js file to enable React with Vite: vite.config.js // vite.config.js
import { defineConfig } from "vite";
import react from "@vitejs/plugin-react";
import eslintPlugin from "vite-plugin-eslint";
import svgx from "@svgx/vite-plugin-react";

export default defineConfig((configEnv) => ({
  plugins: [react(), eslintPlugin(), svgx()],
  server: {
    port: 3000,
  },
})); // vite.config.js
import { defineConfig } from "vite";
import react from "@vitejs/plugin-react";
import eslintPlugin from "vite-plugin-eslint";
import svgx from "@svgx/vite-plugin-react";

export default defineConfig((configEnv) => ({
  plugins: [react(), eslintPlugin(), svgx()],
  server: {
    port: 3000,
  },
})); Step 1: Installation Step 1: Installation First, we will edit the src/App.tsx file to create the project's homepage. src/App.tsx // src/App.tsx
const App: React.FC = () => {
  return (
    <div>
        <h1>Diablo 4</h1>
    </div>
  );
};
export default App; // src/App.tsx
const App: React.FC = () => {
  return (
    <div>
        <h1>Diablo 4</h1>
    </div>
  );
};
export default App; This creates a basic homepage layout. Now let's update the src/index.tsx file to render the root component of the project: src/index.tsx Now we can use the following command to initialize the project: You can see the project running by navigating to http://localhost:3000 in your browser. http://localhost:3000 Step 2: Creating the Folder Structure Step 2: |public // Font, Resim ve videolar
|src
|_containers // Sayfalardaki kapsayıcı alanlar 
|_components // Input, Button, Typography çekirdek bileşenler |public // Font, Resim ve videolar
|src
|_containers // Sayfalardaki kapsayıcı alanlar 
|_components // Input, Button, Typography çekirdek bileşenler In the public file, we can split it into 3 folders: fonts, images and logos. For the main, subheadings and normal articles, 3 fonts AlegreyaSans, ExocetBlizzard and OldFenris fonts are used, we can get them from the site. When we examine the site, we can see that it has image and video based content, we can get the images and videos in the public folder piece by piece. In the public file, we can split it into 3 folders: fonts, images and logos. For the main, subheadings and normal articles, 3 fonts AlegreyaSans, ExocetBlizzard and OldFenris fonts are used, we can get them from the site. When we examine the site, we can see that it has image and video based content, we can get the images and videos in the public folder piece by piece. Step 3: Developer Freestyle Step 3: I wanted to share a fast deliver project setup with you. In the next part, I recommend you to proceed a little more on the tech stack you want. Since I used styled-components, vite, typescript and react as tech stack in this project, I tried to make a sample setup. I hope it was useful, you can review the rest of the project from the project link. https://github.com/yagiz-aydin/diablo4 https://github.com/yagiz-aydin/diablo4 https://github.com/yagiz-aydin/diablo4 https://diablo4-smoky.vercel.app/ https://diablo4-smoky.vercel.app/ https://diablo4-smoky.vercel.app/

Replicating the Diablo 4 Website Using React, Vite, and Styled Components

A Sparse Bayesian Learning for Diagnosis of Nonstationary and Spatially Correlated Faults:References

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Jihoon Chung;
(2)  Zhenyu (James) Kong. Table of Links Abstract & Introduction
Review of Related Work
Proposed Research Methodology
Numerical Case Studies
Real-World Simulation Case Studies
Conclusion
References VI. CONCLUSIONS This paper proposes a novel sparse hierarchical Bayesian method, CSSBL, to effectively identify the sparse process faults in multistation assembly systems. The method identifies process faults by considering the spatial correlation of KCCs and nonstationary process faults among the multiple KPCs. Since posterior distributions of KCCs in the proposed method are computationally intractable, this paper derives approximate posterior distributions of KCCs via Variational Bayes inference. The proposed method’s effectiveness is validated by numerical cases and real-world simulation application using an actual auto-body assembly system. In this work, the temporal correlations of the KCCs are not considered. However, a temporal correlation exists among the KCCs because of the degradation of wear of production tooling over time [23] or the machine-tool thermal distortion [22]. Therefore, extending the proposed CSSBL to consider the temporal correlation of KCCs makes the method more realistic in the actual multistation assembly system. Since the proposed method is not designed for specific processes, it can be effectively applied to other domains, including healthcare and communication systems, for their process monitoring. APPENDIX A APPENDIX B APPENDIX C APPENDIX D APPENDIX E APPENDIX F This paper is available on arxiv under CC 4.0 license. Authors: (1) Jihoon Chung; (2)  Zhenyu (James) Kong. This paper is available on arxiv under CC 4.0 license. Authors: (1) Jihoon Chung; (2)  Zhenyu (James) Kong. Table of Links Abstract & Introduction Review of Related Work Proposed Research Methodology Numerical Case Studies Real-World Simulation Case Studies Conclusion References Abstract & Introduction Abstract & Introduction Review of Related Work Review of Related Work Proposed Research Methodology Proposed Research Methodology Numerical Case Studies Numerical Case Studies Real-World Simulation Case Studies Real-World Simulation Case Studies Conclusion Conclusion References References VI. CONCLUSIONS This paper proposes a novel sparse hierarchical Bayesian method, CSSBL, to effectively identify the sparse process faults in multistation assembly systems. The method identifies process faults by considering the spatial correlation of KCCs and nonstationary process faults among the multiple KPCs. Since posterior distributions of KCCs in the proposed method are computationally intractable, this paper derives approximate posterior distributions of KCCs via Variational Bayes inference. The proposed method’s effectiveness is validated by numerical cases and real-world simulation application using an actual auto-body assembly system. In this work, the temporal correlations of the KCCs are not considered. However, a temporal correlation exists among the KCCs because of the degradation of wear of production tooling over time [23] or the machine-tool thermal distortion [22]. Therefore, extending the proposed CSSBL to consider the temporal correlation of KCCs makes the method more realistic in the actual multistation assembly system. Since the proposed method is not designed for specific processes, it can be effectively applied to other domains, including healthcare and communication systems, for their process monitoring. APPENDIX A APPENDIX B APPENDIX C APPENDIX D APPENDIX E APPENDIX F

A Sparse Bayesian Learning for Diagnosis of Nonstationary and Spatially Correlated Faults:Conclusion

X Corp. v. Center for Countering Digital Hate, INC. Court Filing, retrieved on March 25, 2024 is part of . You can jump to any part in this filing . This part is 9 of 19. HackerNoon’s Legal PDF Series here 3. “In Connection with a Public Issue” Finally, CCDH’s actions in furtherance of its free speech rights were in connection with a public issue. X Corp. alleges that CCDH uses its reports “to demand that platform providers kick the targeted users off of their platforms, thus silencing their viewpoints on broadly debated topics such as COVID-19 vaccines, reproductive healthcare, and climate change.” FAC ¶ 3; see also id. ¶ 44 (March 24, 2021 CCDH report on anti-vaxxers), ¶ 46 (November 10, 2022 CCDH article on hate speech), ¶ 49 (February 9, 2023 CCDH report on hate speech), id. ¶ 72 (“X Corp. has been harmed in its mission to provide its users with a platform in which topics of paramount public concern can be discussed and debated free from the censorship efforts of activist organizations advancing narrow ideological agendas through deceitful means.”). These reports implicate public issues. See Resolute Forest Prods., Inc. v. Greenpeace Int’l, 302 F. Supp. 3d 1005, 1025 (N.D. Cal. 2017) (“California courts have broadly construed public interest to include not only governmental matters, but also private conduct that impacts a broad segment of society and/or that affects a community in a manner similar to that of a governmental entity.” (internal quotation marks omitted); see also PPP Br. at 7 (“Public debate and interest on matters of disinformation and hate speech on social media platforms have been profound in recent years, including multiple congressional hearings and constant coverage by national media.”). X Corp. does not dispute this point. The Court therefore concludes that CCDH has met its burden at the first step of the anti-SLAPP analysis. This order will therefore go on to assess whether X Corp. has established that there is a probability that it will prevail on its claims. See Cal. Code Civ. Proc. § 425.16(b)(1). Continue Reading . Here About HackerNoon Legal PDF Series: We bring you the most important technical and insightful public domain court case filings. This court case retrieved on March 25, 2024, from is part of the public domain. The court-created documents are works of the federal government, and under copyright law, are automatically placed in the public domain and may be shared without legal restriction. storage.courtlistener

Championing Public Dialogue: CCDH's Defense of Free Speech Rights

Finding and retaining active users is a constant pain point for any business, and in the brave new world of Web3 technology, it is a battlefield swarming with bots, lukewarm participants, and unethical practices that ebb away trust and confidence. Many projects constantly grapple with identifying genuine users who are actively engaged in the DeFi space. However, the emergence of Artificial Intelligence (AI) promises to be a game changer as it offers a powerful tool to cut through the noise and identify high-quality users. In most cases, Web3 projects rely on outdated strategies like airdrops and gamified quests to attract users and grow their communities. While these methods might provide a spurt of activity or result in a temporary hype, they often fail to capture long-term engagement, which is a fundamental requirement for a sustainable growth pattern. Take for instance, airdrops are widely used to attract participants solely interested in free crypto tokens, and gamified tasks can bring about liquidity challenges when rewards are no longer forthcoming. While they offer a semblance of growth, these methods fail to differentiate between users who are genuinely interested in the project's goals and functionality, and those who are only involved for quick monetary gains. How AI Is Transforming the User Engagement Landscape Artificial Intelligence is no longer a buzzword. It is a reality that is unfolding in our everyday lives. The impact of AI in the crypto ecosystem has mostly been a positive one. Crypto platforms like BonusBlock and DIA are utilizing AI to analyze on-chain data, which paints a detailed picture of a user's behavior within the DeFi ecosystem. AI models have been deployed in the following instances: Protocol interactions: This is to determine how frequently a user interacts with different protocols. Token diversity: To help identify the number of tokens a user holds as a means of indicating a broader interest across the DeFi space. Liquidity: This underlines how AI can be used to track a user’s contributions to liquidity pools, signifying a stake in the DeFi market's health. Staking: Determining a user’s interest in staking is a way to measure his commitment to specific projects. Governance: Voting in governance proposals indicates a user's active involvement in shaping DeFi protocols. By analyzing this data, AI can identify users who are financially active and genuinely engaged with DeFi, weeding out bots, airdrop farmers, and those with only a passing interest. Example 1: Bonus Block In this example, AI is primarily deployed by Bonus Block, a web3 engagement protocol, to facilitate user acquisition through BONUS, its native token. Users receive rewards in the form of the platform’s token for their on-chain activity, a strategy designed to encourage them to actively participate in the DeFi space. Through this means, users with higher engagement scores, identified through AI analysis, gain greater visibility to projects seeking qualified users. Example 2: DIA, an AI-powered platform, deploys cross-chain oracles to assess a user's creditworthiness in the DeFi ecosystem. The key purpose it serves is to allow projects to identify users who are trusted borrowers and lenders. Benefits of AI-centric Approach Acquiring users in web3 utilizing an AI-powered approach creates a win-win situation for users and projects. It gives web3 projects access to genuinely interested users who are more likely to be converted to long-term members of the community. Users, on the other hand, can leverage their on-chain activity to showcase their value and potentially earn rewards from projects seeking their engagement. Additionally, AI can be used to analyze user sentiment on social media platforms to gauge interest in specific projects and identify potential early users. The Future of Web3 User Acquisition While AI-based user acquisition is still a new marketing experiment, it promises to transform how users are acquired and retained in web3. It goes to show that the success of any web3 project does not only lie in their marketing strategies, but in building genuine connections. AI will feature prominently in the future when the need to leverage AI for intelligent targeting, personalized experiences, and community engagement, arises.  It will help Web3 projects grow a loyal user base that actively contributes to the growth and sustainability of Defi. Conclusion AI is ushering in a new era in web3 user acquisition, offering a better alternative to traditional airdrops and gamified concepts. By analyzing on-chain data and identifying active users, it can provide a solution to the prevalence of bots and unethical practices. With the help of AI, blockchain projects can reach high-quality users who are more likely to become loyal members. As AI continues to attract significant attention in web3, its potential to reshape Web3 user acquisition entirely is evidently present, helping us to imagine a future that is built on genuine engagement and mutual benefit for both projects and users.

Stories, News & Mentions about DIXA

Dixa’s Stories on HackerNoon