paper_id
stringlengths 10
10
⌀ | yes_no
bool 2
classes | paper_index
int64 0
519
⌀ | evidence
stringlengths 0
37.7k
| question
stringlengths 4
11.7k
| answer
stringlengths 1
26k
|
|---|---|---|---|---|---|
null | false
| null |
An acute asthma exacerbation is commonly referred to as an asthma attack. The classic symptoms are shortness of breath, wheezing, and chest tightness. The wheezing is most often when breathing out. While these are the primary symptoms of asthma, some people present primarily with coughing, and in severe cases, air motion may be significantly impaired such that no wheezing is heard. In children, chest pain is often present.
https://en.wikipedia.org/wiki/Asthma
|
Extract the symptoms of an asthma attack from this passage in a bulleted list.
|
-Shortness of breath
-Wheezing
-Chest tightness and/or pain
-Coughing
-Significantly impaired air motion
|
null | false
| null |
What is the future for human?
|
This is a very wide open question. It could be two directions. One direction is that human would just disappear like the dinosaur. One direction is that human would become a species like a god. For the first direction, it may be caused by many situations. e.g. Nuclear war from WW3, extreme climate change. For the other direction, reproduced energy is ready, all cancers can be cured, everything can be done with robot with AI, human life is unlimited.
|
|
null | false
| 484
|
As presented in this section, we conducted two series of experiments to analyze the property of IMTC. First, we qualitatively evaluated the diversity of options learned by IMTC with intrinsic rewards, without any extrinsic rewards. Second, we quantitatively test the reusability of learned options by). Also, we can see that the magnitude of intra-option policies tends larger with constant rewards. Right: Options learned by other methods. OC produces a dead option 3 that terminates everywhere and never-ending options 0 and 1. About intra-option policies, all methods successfully avoided learning the same policy, but they only have two directions.
task adaptation on a specific task. As a baseline of termination learning method, we compared our method with OC. OC is trained with VIC rewards during pre-training. We did not compare IMTC with TC because our TC implementation failed to learn options with relatively small termination regions as reported in the paper, and there is no official public code for TC 1 . During pre-training without extrinsic rewards, IMTC receives intrinsic rewards when the current option terminates. We compare three IMTC variants with different intrinsic rewards: (i) VIC, (ii) RVIC, and (iii) constant value (R IMTC = 0.01). Note that R IMTC = 0.01 is chosen from [0.1, 0.05, 0.01] based on the task adaptation results. We also compare IMTC with vanilla VIC and RVIC with fixed termination probabilities. We used ∀ x β o (x) = 0.1 since it performed the best in task adaptation experiments, while 0.05 was used in. Note that RVIC's objective I(X s ; O|x f ) is different from ours, while IMTC and VIC share almost the same objective. Thus, the use of VIC is more natural, and the combination with RVIC is tested to show the applicability of IMTC. Further details of our VIC and RVIC implementation are found in Appendix B. In order to check only the effect of the different methods for learning beta β o , the rest of the implementation is the same for all these methods. That is, OC, vanilla VIC, and vanilla RVIC are also based on PPO and advantage estimation methods in Section 4.2. In this section, we fix the number of options as |O| = 4 for all option-learning methods. We further investigated the effect of the number of options Appendix C, where we confirmed that |O| = 4 is sufficient for most domains. All environments that we used for experiments are implemented on the MuJoCo physics simulator. We further describe the details in Appendix C.
Option Learning From Intrinsic Rewards We now qualitatively compare the options learned by IMTC with options of other methods. Learned options depend on the reward structure in the environment, which enables manually designing good reward functions for learning diverse options. Thus, we employed a reward-free RL setting where no reward is given to agents. Instead, each compared method uses some intrinsic rewards, as explained. We fix µ as µ(o|x) = 1 |O| in this experiment, since we assume that the future tasks are uniformly distributed. Intra-option policies are trained by PPO and independent GAE (8). We show network architectures and hyperparameters in Appendix C. We set the episode length to 1 × 10 4 , i.e., an agent is reset to its starting position after 1 × 10 4 steps. For all visualizations, we chose the best one from five independent runs with different random seeds. We visualized learned options in PointReach environment shown in Figure. In this environment, an agent controls the ball initially placed at the center of the room. The state space consists of positions (x, y) and velocities (∆x, ∆y) of an agent, and the action space consists of acceralations ( ∆x ∆t , ∆y ∆t ). Figure shows the options learned in this environment after 4 × 10 6 steps. Each arrow represents the mean value of intra-option policies, and the heatmaps represent β o . In this experiment, we observed the effect of IMTC clearly, for both termination regions and intra-option policies. Interestingly, we don't see clear differences between options learned with VIC and RVIC rewards, while constant rewards tend to make options peaker. OC failed to learn meaningful termination regions: option 0 and 1 never terminate, and option 3 terminates almost everywhere. This result confirms that IMTC can certainly diversify options. Moreover, compared to vanilla VIC and RVIC, intra-option policies learned by IMTC with VIC or RVIC rewards are clearer, in terms of both the magnitude and directions of policies. We believe that this is because diversifying termination regions gives more biased samples to the option classifiers employed by VIC and RVIC. Figure Transferring skills via task adaptation Now we quantitatively test the reusability of learned options by task adaptation with specific reward functions. Specifically, we first trained agents with intrinsic rewards as per the previous section. Then we transferred agents to an environment with the same state and action space but with external rewards. We prepared multiple reward functions, which we call tasks, for each domain and evaluated the averaged performance over tasks. We compare IMTC with OC, vanilla VIC, vanilla RVIC, and PPO without pre-training. Also, we compare three variants of IMTC with different intrinsic rewards during pre-training. For a fair comparison, UGAE (9) and PPO are used for all options learning methods. Note that we found UGAE is very effective in this experiments, as we show the ablation study in Appendix C.6. For vanilla VIC and vanilla RVIC, termination probability is fixed to 0.1 through pre-training and task adaptation. -greedy based on Q O with = 0.1 is used as the option selection policy µ. We hypothesize that diverse options learned by IMTC can help quickly adapt to given tasks, supposing the diversity of tasks.
Figure shows all domains used for task adaptation experiments. For simplicity, all tasks have goalbased sparse reward functions. I.e., an agent receives R t = 1.0 when it satisfies a goal condition, and otherwise the control cost −0.0001 is given. Red circles show possible goal locations for each task. When the agent fails to reach the goal after 1000 steps, it is reset to a starting position. PointReach, SwimmerReach, and AntReach are simple navigation tasks where an agent aim to just navigate itself to the goal. We also prepared tasks with object manipulation: in PointBilliard and AntBilliard an agent aims to kick the blue ball to the goal position, and in PointPush and AntPush, it has to push the block out of the way to the goal. We pre-traine options learning agents for 4 × 10 6 environmental steps and additionally trained them for 1 × 10 6 steps for each task. Figure shows learning curves and scatter plots drawn from five independent runs with different random seeds per domain. 2 Here, we observed that IMTC with VIC or RVIC rewards performed the best or was compatible with baselines. IMTC with VIC performed better than OC with VIC except for AntRearch, which backs up the effectiveness of diversifying termination regions for learning reusable options. Also, IMTC with VIC and IMTC with RCIC respectively performed better in most of the tasks than VIC and RVIC with fixed termination probabilities. This result suggests that IMTC can boost the performance of option learning methods based on option classifiers, even when the objective is different as with RVIC. On the other hand, IMTC with constant rewards (R IMTC = 0.01) performed worse than IMTC with VIC or RVIC rewards, although it also learned diverse options as we show in Figure, suggesting the importance of adjusting rewards. We further analyzed the evolution of intrinsic rewards of VIC and RVIC in Appendix C.5. In addition, we can observe that IMTC's performance is especially better than other methods in relatively complex PointBilliard, AntBilliard, and AntPush, where object manipulation is required. Considering that manipulated balls and boxes move faster than agents in these domains, a choice of options can lead to larger differences in the future state. IMTC is suitable to these domains since it maximizes the diversity of the resulting states, while PPO struggles to learn. Contrary, IMTC's performance is close to other methods in Reach tasks, where the goal states are relatively close to the starting states in terms of euclidian distances.
Gridworld experiments and limitation of the method Although IMTC successfully learned diverse options in MuJoCo experiments, our analysis in Figure shows the possibility of learning options that are not interesting but have large MI. We further investigated this possibility by visualizing options in a classical four rooms gridworld in Appendix C.8. Interestingly, we observed that IMTC could fall into diverse but unmeaningful options in that environment. We believe that IMTC is often sufficient in a large environment where a randomly-initialized agent rarely gets the same trajectory. However, when the number of possible trajectories is small, diversifying the destinations could be insufficient. In such cases, it can be necessary to extend IMTC to diversify trajectories as done in.
In addition, we can observe that IMTC’s performance is especially better than other methods in relatively complex PointBilliard, AntBilliard, and AntPush, where object manipulation is required. Considering that manipulated balls and boxes move faster than agents in these domains, a choice of options can lead to larger differences in the future state. IMTC is suitable to these domains since it maximizes the diversity of the resulting states, while PPO struggles to learn. Contrary, IMTC’s performance is close to other methods in Reach tasks, where the goal states are relatively close to the starting states in terms of euclidian distances.
|
Does the complexity of the task or the action space play a role in how difficult it is to learn options?
|
This is an interesting point. We agree that the cardinality of action space would matter. However, we believe that what is more important is the distance to the goal state. IMTC is more advantageous in tasks where goal states are more distant in terms of euclidian distance due to its nature of seeking the diversity of option terminating state. We will add more discussion in the revised manuscript.
|
null | false
| null |
Dredging is the excavation of material from a water environment. Possible reasons for dredging include improving existing water features; reshaping land and water features to alter drainage, navigability, and commercial use; constructing dams, dikes, and other controls for streams and shorelines; and recovering valuable mineral deposits or marine life having commercial value. In all but a few situations the excavation is undertaken by a specialist floating plant, known as a dredger.
Dredging is carried out in many different locations and for many different purposes, but the main objectives are usually to recover material of value or use, or to create a greater depth of water. Dredges have been classified as suction or mechanical. Dredging systems can either be shore-based, brought to a location based on barges, or built into purpose built vessels.
Dredging has significant environmental impacts: it can disturb marine sediments, leading to both short- and long-term water pollution, destroy important seabed ecosystems, and can release legacy human-sourced toxins captured in the sediment. These environmental impacts can significantly hurt marine wildlife populations, contaminate sources of drinking water and interrupt economic activities such as fishing.
|
From this text, find out the pros and cons of dredging
|
The benefits of dredging are:
1. improves existing water features for better drainage, navigability, and commercial use
2. important for constructing dams, dikes, and other controls for streams and shorelines
3. helps in recovering valuable mineral deposits or marine life having commercial value
The demerits of dredging are related to its severe environmental impact, for example:
1. it can disturb marine sediments, leading to both short- and long-term water pollution
2. it can destroy important seabed ecosystems
3. it can release legacy human-sourced toxins captured in the sediment
All the above impacts can significantly hurt marine wildlife populations, contaminate sources of drinking water and interrupt economic activities such as fishing.
|
null | false
| 477
|
Table A5: Comparable results with baselines, where there are 20% least frequent relations. Here we report the results in term of MR, MRR, Hit@K (K=1,3,10).****Table A6: Result comparisions with baselines on generalization setting by randomly removing 5% relations. Here we report the results in term of MR, MRR, Hit@K (K=1,3,10).****Table A7: Result comparisions with baselines on generalization setting by randomly removing 10% relations. Here we report the results in term of MR, MRR, Hit@K (K=1,3,10).****Table A8: Result comparisions with baselines on generalization setting by randomly removing 15% relations. Here we report the results in term of MR, MRR, Hit@K (K=1,3,10).****Table A9: Result comparisions with baselines on generalization setting by randomly removing 20% relations. Here we report the results in term of MR, MRR, Hit@K (K=1,3,10).
|
Is it possible for the authors to compare against strong baselines methods for knowledge graph completion such [1] and [2]? [1] Zhu, Zhaocheng, Zuobai Zhang, Louis-Pascal Xhonneux, and Jian Tang. "Neural Bellman-Ford Networks: A General Graph Neural Network Framework for Link Prediction." arXiv preprint arXiv:2106.06935 (2021).
[2] Zhang, Shuai, Yi Tay, Lina Yao, and Qi Liu. "Quaternion knowledge graph embeddings." arXiv preprint arXiv:1904.10281 (2019)
|
We include [1] (named QuatE in standard setting and QuatE in generalization settings) as a baseline method and update its results in Tables 6, A5, A6, A7, A8, A9. These results further verify the superiority of the proposed method in terms of MRR, Hit@1, Hit@3, Hit@10.
For [2], we are unable to find the source code and detailed hyper-parameter settings such as the dimensions of embedding in the paper. So, we do not presently include this work. However, we cite [2] and leave the comparison as future work.
|
null | false
| null |
In the series A Song of Ice and Fire, who is the founder of House Crayne?
|
Rose of the Red Lake
|
|
null | false
| null |
Tell whether these things smell nice or unpleasant: garbage, lemons, sweat, oceans, rain
|
garbage: smells unpleasant, lemons: smells nice, sweat: smells unpleasant, oceans: smell nice, rain: smells nice
|
|
null | false
| null |
What is the Simpson family dog named?
|
Santa's little helper
|
|
null | false
| 120
|
Cancer is one of the leading causes of death in the world, with over 80,000 deaths registered in Canada in 2017 (Canadian Cancer Statistics 2017). A computer-aided system for cancer diagnosis usually involves a pathologist rendering a descriptive report after examining the tissue glass slides obtained from the biopsy of a patient. A pathology report contains specific analysis of cells and tissues, and other histopathological indicators that are crucial for diagnosing malignancies. An average sized laboratory may produces a large quantity of pathology reports annually (e.g., in excess of 50,000), but these reports are written in mostly unstructured text and with no direct link to the tissue sample. Furthermore, the report for each patient is a personalized document and offers very high variability in terminology due to lack of standards and may even include misspellings and missing punctuation, clinical diagnoses interspersed with complex explanations, different terminology to label the same malignancy, and information about multiple carcinoma appearances included in a single report BIBREF0 .
In Canada, each Provincial and Territorial Cancer Registry (PTCR) is responsible for collecting the data about cancer diseases and reporting them to Statistics Canada (StatCan). Every year, Canadian Cancer Registry (CCR) uses the information sources of StatCan to compile an annual report on cancer and tumor diseases. Many countries have their own cancer registry programs. These programs rely on the acquisition of diagnostic, treatment, and outcome information through manual processing and interpretation from various unstructured sources (e.g., pathology reports, autopsy/laboratory reports, medical billing summaries). The manual classification of cancer pathology reports is a challenging, time-consuming task and requires extensive training BIBREF0 .
With the continued growth in the number of cancer patients, and the increase in treatment complexity, cancer registries face a significant challenge in manually reviewing the large quantity of reports BIBREF1 , BIBREF0 . In this situation, Natural Language Processing (NLP) systems can offer a unique opportunity to automatically encode the unstructured reports into structured data. Since, the registries already have access to the large quantity of historically labeled and encoded reports, a supervised machine learning approach of feature extraction and classification is a compelling direction for making their workflow more effective and streamlined. If successful, such a solution would enable processing reports in much lesser time allowing trained personnel to focus on their research and analysis. However, developing an automated solution with high accuracy and consistency across wide variety of reports is a challenging problem.
For cancer registries, an important piece of information in a pathology report is the associated ICD-O code which describes the patient's histological diagnosis, as described by the World Health Organization's (WHO) International Classification of Diseases for Oncology BIBREF2 . Prediction of the primary diagnosis from a pathology report provides a valuable starting point for exploration of machine learning techniques for automated cancer surveillance. A major application for this purpose would be “auto-reporting” based on analysis of whole slide images, the digitization of the biopsy glass slides. Structured, summarized and categorized reports can be associated with the image content when searching in large archives. Such as system would be able to drastically increase the efficiency of diagnostic processes for the majority of cases where in spite of obvious primary diagnosis, still time and effort is required from the pathologists to write a descriptive report.
The primary objective of our study is to analyze the efficacy of existing machine learning approaches for the automated classification of pathology reports into different diagnosis categories. We demonstrate that TF-IDF feature vectors combined with linear SVM or XGBoost classifier can be an effective method for classification of the reports, achieving up to 83% accuracy. We also show that TF-IDF features are capable of identifying important keywords within a pathology report. Furthermore, we have created a new dataset consisting of 1,949 pathology reports across 37 primary diagnoses. Taken together, our exploratory experiments with a newly introduced dataset on pathology reports opens many new opportunities for researchers to develop a scalable and automatic information extraction from unstructured pathology reports.
We demonstrate that TF-IDF feature vectors combined with linear SVM or XGBoost classifier can be an effective method for classification of the reports, achieving up to 83% accuracy. We also show that TF-IDF features are capable of identifying important keywords within a pathology report.
|
What is the effective method for the classification of the reports in this paper?
|
The TF-IDF feature vectors combined with linear SVM or XGBoost classifier can be an effective method for classification of the reports, achieving up to 83% accuracy.
|
null | false
| 478
|
Time warping random changes the speed of the timeline1 . The number of speed changes is 100 and the maximal ratio of max/min speed is 10. If necessary, over-sampling or sampling methods are adopted to ensure the length of the augmented instance is the same as the original one.****https://tsaug.readthedocs.io/****
|
If time warping is applied, the number of samples per window is different right? So how do the authors deal with this? Is resampling applied maybe?
|
As we have introduced in the initial version, we adopt the tool from https://tsaug.readthedocs.io. With this function, we may generate augmented sequences with different lengths. We further apply over-sampling or sampling to make sure the length is the same as the original one. We have added more details here. Thanks for pointing it out.
|
null | false
| null |
Give me a bulleted list of all of the Star Wars movies in order by release date.
|
1. Episode IV: A New Hope (1977)
2. Episode V: The Empire Strikes Back (1980)
3. Episode VI: Return of the Jedi (1983)
4. Episode I: The Phantom Menace (1999)
5. Episode II: Attack of the Clones (2002)
6. Episode III: Revenge of the Sith (2005)
7. Episode VII: The Force Awakens (2015)
|
|
null | false
| null |
Give me some ideas to manage my manager.
|
1. Be honest
2. Communicate
3. Understand goals
4. Stay calm
5. Solve problems
6. Be positive
7. Share your knowledge
8. Be caring
9. Practice positive reinforcement
10. Be a strong performer
|
|
null | false
| null |
Tell me whether these shows are animated or not: South Park, Rugrats, Curb Your Enthusiasm, House Hunters, Attack on Titan
|
South Park: Animated
Rugrats: Animated
Curb Your Enthusiasm: Not Animated
House Hunters: Not Animated
Attack on Titan: Animated
|
|
1708.06022
| false
| null |
BIBREF11 revisit bilingual pivoting in the context of neural machine translation (NMT, BIBREF12 , BIBREF13 ) and present a paraphrasing model based on neural networks. At its core, NMT is trained end-to-end to maximize the conditional probability of a correct translation given a source sentence, using a bilingual corpus. Paraphrases can be obtained by translating an English string into a foreign language and then back-translating it into English. NMT-based pivoting models offer advantages over conventional methods such as the ability to learn continuous representations and to consider wider context while paraphrasing.
Paraphrases can be obtained by translating an English string into a foreign language and then back-translating it into English.
|
It looks like learning to paraphrase questions, a neural scoring model and a answer selection model cannot be trained end-to-end. How are they trained?
|
using multiple pivot sentences
|
null | false
| null |
Taylor Alison Swift (born December 13, 1989) is an American singer-songwriter. Her genre-spanning discography, songwriting abilities and artistic reinventions have received critical praise and wide media coverage. Born in West Reading, Pennsylvania, Swift moved to Nashville at age 14 to become a country artist. She signed a songwriting deal with Sony/ATV Music Publishing in 2004 and a recording contract with Big Machine Records in 2005. Her 2006 self-titled debut album made her the first female country artist to write a U.S. platinum-certified album.
Swift's next albums, Fearless (2008) and Speak Now (2010), explored country pop. The former's "Love Story" and "You Belong with Me" were the first country songs to top the U.S. pop and all-genre airplay charts, respectively. She experimented with rock and electronic styles on Red (2012), which featured her first Billboard Hot 100 number-one song, "We Are Never Ever Getting Back Together", and eschewed her country image in her synth-pop album, 1989 (2014), supported by chart-topping songs "Shake It Off", "Blank Space", and "Bad Blood". Media scrutiny inspired the urban-flavored Reputation (2017) and its number-one single "Look What You Made Me Do".
Exiting Big Machine, Swift signed with Republic Records in 2018 and released her seventh studio album, Lover (2019), followed by the autobiographical documentary Miss Americana (2020). She ventured into indie folk and alternative rock in her 2020 albums Folklore and Evermore, whose singles "Cardigan" and "Willow" topped the Hot 100. Swift began re-recording her first six albums after a dispute over their masters, re-releasing two in 2021—Fearless (Taylor's Version) and Red (Taylor's Version). The latter's "All Too Well (10 Minute Version)" became the longest song to top the Hot 100. Her 2022 album Midnights and single "Anti-Hero" broke all-time streaming records. Swift has directed music videos and films, such as All Too Well: The Short Film (2021), and played supporting roles in others.
Having sold over 200 million records globally, Swift is one of the best-selling musicians in history. She is the most streamed woman on Spotify and the only act to have five albums open with over one million copies sold in the US. Among her accolades are 12 Grammy Awards, including three Album of the Year wins; an Emmy Award; 40 American Music Awards; 29 Billboard Music Awards; three IFPI Global Recording Artist of the Year awards; and 92 Guinness World Records. Swift has been featured in rankings such as Rolling Stone's 100 Greatest Songwriters of All Time, Billboard's Greatest of All Time Artists, the Time 100 and Forbes Celebrity 100. Honored with titles such as Artist of the Decade and Woman of the Decade, Swift is an advocate for artists' rights and women's empowerment. Her music is credited with influencing a generation of singers and songwriters.
|
Mention the Record labels Taylor was associated with.
|
Taylor Swift was associated with the following record labels:
1. Sony/ATV Music Publishing (2004)
2. Big Machine Records (2005)
3. Republic Records (2018)
|
null | false
| null |
Robert Frederick Chelsea Moore OBE (12 April 1941 – 24 February 1993) was an English professional footballer. He most notably played for West Ham United, captaining the club for more than ten years, and was the captain of the England national team that won the 1966 FIFA World Cup. He is widely regarded as one of the greatest defenders in the history of football, and was cited by Pelé as the greatest defender that he had ever played against.
|
When was Bobby Moore born and when did he die?
|
Bobby Moore was born in 1941 and died in 1993.
|
null | false
| null |
How do I write a promo packet?
|
The concept of promo packet was popularized by Google. A candidate write a promo packet that will be reviewed by an independent committee. Since a person only seeks promotion every few years, they generally don't know how to write a proper promo packet. Here are a few tips about writing a promo packet.
First you need to describe your team and your role on the team, so the committee understands your context. For example, being the tech lead of Amazing Service for the past 2 years.
Second you need to describe the biggest accomplishments in recent years, typically up to 5. For example, launched Amazing Service or improved SLA from 3 9s to 4 9s.
Next you need to show your actual work closely matches the requirements of the next level. For example, designed and built the security system for Amazing Service, or hired and managed the Amazing Service team, or initiated Amazing Service and grew it to $100M ARR.
The above points are your promo rationale, which should be the focus of your packet. The rest of the packet is to provide verifiable evidence to your rationale.
A good promo packet informs the committee of your role, your accomplishments, the next level traits demonstrated in your accomplishments , plus sufficient evidence. Best luck!
|
|
null | false
| null |
How do I get better at the game of golf?
|
Golf is a game that originated in the 15th century, and to this day, is still being iterated on. Although there have been many different swing styles over the years, there are fundamentals to playing the game that when followed, have produced the most successful golfers.
Building a strong golf foundation using 4 key principles is key to improving at the game of golf which are: aligning your expectations to reality, applying an effective on course strategy, practicing correctly, and, building a sharp mental game.
By aligning your expectations to the reality of how you are playing, you allow your mind to accept the good with the bad which allows for a more free and easy playing style, often leading to better play.
Applying an effective course strategy will allow you to score better as you know where you can make up strokes on your competitors. For example, driving distance off the tee paired with greens in regulation play a crucial role in scoring, as opposed to putting whereas you have a 20 foot putt compared to a 40 foot putt, your make percentages are both low.
Learning how to practice correctly can also improve your on course scoring as most approach practice incorrectly. Zombie range sessions have become the norm, but rarely in golf do you have the same shot twice in a row. Implement that same theory on the driving range so you have more of a feel of what on course play will feel like.
Finally, in order to improve your scoring, you need to build a sharp mental game while you play. This includes visualizing your shot before you hit it, being able to clear your head of all unwanted thoughts other than the shot you want to hit, and implementing a pre shot routine that is easily replicable.
By applying these core 4 strategies, you should see your golf scores decrease without any significant change in your actual swing.
|
|
null | false
| null |
Classify the following as types of birds or types of snakes: robin, cobra, rattlesnake, eagle, viper, raptor, bluejay, cottonmouth, copperhead, sparrow.
|
Robin, eagle, raptor, bluejay, and sparrow are all kinds of birds. Cobra, rattlesnake, viper, cottonmouth, and copperhead are several types of snakes.
|
|
null | false
| null |
What are some examples of headwear?
|
- Baseball cap
- Visor
- Skull cap
- Yamaka
- Beanie
- Cowboy hat
- Helmet
- Headband
|
|
null | false
| null |
Generate word problem on ratios suitable for grade 10
|
An aquarium has a total of 120 animals, out of which 40 are amphibians and the rest are fish. What is the ratio of amphibians to the total number of animals in the aquarium?
|
|
1912.01673
| false
| null |
In the second round, we collected 293 annotations from 12 annotators. After Korektor, there are 4262 unique sentences (including 150 seed sentences) that form the COSTRA 1.0 dataset. Statistics of individual annotators are available in tab:statistics.
After Korektor, there are 4262 unique sentences (including 150 seed sentences) that form the COSTRA 1.0 dataset.
|
How many sentence transformations on average are available per unique sentence in dataset?
|
27.41 transformation on average of single seed sentence is available in dataset.
|
null | false
| null |
The C40 went on to win 5 editions of Paris-Roubaix in 6 years. Surpringly, the victory of Tadej Pogačar in the 2020 edition of the Tour de France marked the first time a Colnago-branded bicycle was ridden by the overall classification winner, since Merckx's victories were aboard re-branded bikes.
|
Given the paragraph about the Colnago C40, in which year was a Colnago-branded bicycle first ridden by the overall classification winner in the Tour de France?
|
2020 was the first year a Colnago-branded bicycle was first ridden by the overall classification winner in the Tour de France.
|
null | false
| null |
Classify the following instruments as either brass or woodwind: Alto Saxophone, Trumpet, Clarinet, Oboe, Trombone, Horn, Drums.
|
Of the given list, the Alto Saxophone, Clarinet, and Oboe are woodwind instruments because they all generate sound with a reed. Contrarily, the Trumpet, Trombone, and Horn are brass instruments. This family's instruments are made of brass, and the players vibrate their lips to produce a sound. The drum is neither a woodwind nor a brass instrument but part of the percussion instrument family.
|
|
null | false
| null |
Classify each of the following as either a pokemon or a pokemon trainer: Charmander, Squirtle, Bulbasaur, Ash, Brock, Golem, Arbok
|
Charmander, Squirtle, Bulbasaur, Golem and Arbok are pokemon. Ash and Brock are pokemon trainers.
|
|
null | false
| null |
What are the top 5 countries in terms of life expectancy?
|
According to data from the World Bank, the top 5 countries with the highest live expectancies are Hong Kong, Macao, Japan, Singapore, and South Korea
|
|
null | false
| null |
Isma'il ibn Ahmad ibn Hassan bani Yani (Arabic: إسماعيل بن أحمد بن حسن بني يني), known simply as Isma'il Ragheb Pasha (Arabic: إسماعيل راغب باشا) (1819–1884), was a Greek Ottoman politician who served as a Prime Minister of Egypt and held several other high-ranking government positions.
Isma'il Ragheb was of Greek ancestry and was born in Greece on 18 August 1819 on either the island of Chios following the great massacre of Candia, Crete. After being kidnapped to Anatolia he was brought to Egypt as a slave by Ibrahim Pasha in 1830 and was converted to Islam. Immediately following his arrival, he studied at al-Maktab al-Amiri and obtained his advanced degree in 1834. He was fluent in Greek and was elevated to the rank of First Lieutenant by Egypt's viceroy Muhammad Ali Pasha. In 1836, he became head of the Accounting and Revenue Agencies. He was promoted to the rank of bikbashi (Lieutenant Colonel) in 1840, then kaymakam (Colonel) in 1844, and finally amiralay (Brigadier General) in 1846.[citation needed]
He held the positions of Minister of Finance (1858–1860), then Minister of War (1860–1861).[citation needed] He became Inspector for the Maritime Provinces in 1862, and later Assistant (Arabic: باشمعاون) to viceroy Isma'il Pasha (1863–1865). He was granted the title of beylerbey and then appointed President of the Privy council in 1868. He was appointed President of the Chamber of Deputies (1866–1867), then Minister of Interior in 1867, then Minister of Agriculture and Trade in 1875.[citation needed] He again held the Finance portfolio in Muhammad Sharif Pasha's first government (1879). After the fall of Mahmoud Sami el-Baroudi's government, Isma'il Ragheb became Prime Minister of Egypt in 1882.[citation needed] Although his government was short-lived (it lasted from 17 June to 21 August only), it was the only one to present concrete programs. His achievements include the modernisation of the budget through the inventory of revenues and expenses, the Law on Salaries, and the La'eha Sa'ideyya as well as several agricultural laws.[citation needed]
Isma'il Ragheb died in 1884.[citation needed]
|
Who was Raghib Pasha?
|
Raghib Pasha was a Greek Ottoman politician who served in various roles in Egypt during the mid to late 1800s, including Prime Minister of Egypt for a brief period.
|
null | false
| null |
Classify the following animals as bipeds or quadripeds: dog, human, cow, chicken, ostrich, zebra, tyrannosaurus rex, tiger
|
dog - quadriped
human - biped
cow - quadriped
chicken - biped
ostrich - biped
zebra - quadriped
tyrannosaurus rex - biped
tiger - quadriped
|
|
null | false
| null |
Q: Does spraying people with disinfectant, or having people go through disinfectant tunnels, walkways, or chambers, lower the spread of COVID-19?
|
: No. The FDA does not recommend spraying humans with aerosolized disinfectant. At this time, there is a lack of data to demonstrate that sanitation tunnels are effective in reducing the spread of or in treating COVID-19.
Chemicals used in sanitation tunnels can irritate the skin, eyes, or airways and cause other health issues.
Hand sanitizers are only for use on the hands and are not to be used over larger body surfaces, ingested, inhaled, or injected. Given serious safety concerns and the lack of data to demonstrate that sanitation tunnels are effective in reducing the spread of or in treating COVID-19, the FDA strongly discourages the use or development of sanitation tunnels at this time, as described in this February 2022 guidance titled “COVID-19 Public Health Emergency: Policy on COVID-19-Related Sanitation Tunnels.”
Surface disinfectants or sprays should not be used on humans or animals. They are intended for use on hard, non-porous surfaces (materials that do not absorb liquids easily). CDC provides information regarding disinfectant practices for surfaces. CDC states you should never eat, drink, breathe or inject disinfectants into your body or apply directly to your skin as they can cause serious harm.
|
|
null | false
| null |
What is Freezing Rain?
|
Freezing rain occurs when snowflakes descend into a warmer layer of air and melt completely. When these liquid water drops fall through another thin layer of freezing air just above the surface, they don't have enough time to refreeze before reaching the ground. Because they are “supercooled,” they instantly refreeze upon contact with anything that that is at or below freezing (32 degrees F), creating a glaze of ice on the ground, trees, power lines, or other objects. Even light accumulations can cause dangerous travel, while heavier amounts can cause significant damage to trees and power lines. A significant accumulation of freezing rain lasting several hours or more is called an ice storm.
|
|
null | false
| 342
|
The majority of the English text available worldwide is generated by non-native speakers BIBREF0 . Such texts introduce a variety of challenges, most notably grammatical errors, and are of paramount importance for the scientific study of language acquisition as well as for NLP. Despite the ubiquity of non-native English, there is currently no publicly available syntactic treebank for English as a Second Language (ESL).
To address this shortcoming, we present the Treebank of Learner English (TLE), a first of its kind resource for non-native English, containing 5,124 sentences manually annotated with POS tags and dependency trees. The TLE sentences are drawn from the FCE dataset BIBREF1 , and authored by English learners from 10 different native language backgrounds. The treebank uses the Universal Dependencies (UD) formalism BIBREF2 , BIBREF3 , which provides a unified annotation framework across different languages and is geared towards multilingual NLP BIBREF4 . This characteristic allows our treebank to support computational analysis of ESL using not only English based but also multilingual approaches which seek to relate ESL phenomena to native language syntax.
While the annotation inventory and guidelines are defined by the English UD formalism, we build on previous work in learner language analysis BIBREF5 , BIBREF6 to formulate an additional set of annotation conventions aiming at a uniform treatment of ungrammatical learner language. Our annotation scheme uses a two-layer analysis, whereby a distinct syntactic annotation is provided for the original and the corrected version of each sentence. This approach is enabled by a pre-existing error annotation of the FCE BIBREF7 which is used to generate an error corrected variant of the dataset. Our inter-annotator agreement results provide evidence for the ability of the annotation scheme to support consistent annotation of ungrammatical structures.
Finally, a corpus that is annotated with both grammatical errors and syntactic dependencies paves the way for empirical investigation of the relation between grammaticality and syntax. Understanding this relation is vital for improving tagging and parsing performance on learner language BIBREF8 , syntax based grammatical error correction BIBREF9 , BIBREF10 , and many other fundamental challenges in NLP. In this work, we take the first step in this direction by benchmarking tagging and parsing accuracy on our dataset under different training regimes, and obtaining several estimates for the impact of grammatical errors on these tasks.
To summarize, this paper presents three contributions. First, we introduce the first large scale syntactic treebank for ESL, manually annotated with POS tags and universal dependencies. Second, we describe a linguistically motivated annotation scheme for ungrammatical learner English and provide empirical support for its consistency via inter-annotator agreement analysis. Third, we benchmark a state of the art parser on our dataset and estimate the influence of grammatical errors on the accuracy of automatic POS tagging and dependency parsing.
The remainder of this paper is structured as follows. We start by presenting an overview of the treebank in section SECREF2 . In sections SECREF3 and SECREF4 we provide background information on the annotation project, and review the main annotation stages leading to the current form of the dataset. The ESL annotation guidelines are summarized in section SECREF5 . Inter-annotator agreement analysis is presented in section SECREF6 , followed by parsing experiments in section SECREF7 . Finally, we review related work in section SECREF8 and present the conclusion in section SECREF9 .
Second, we describe a linguistically motivated annotation scheme for ungrammatical learner English and provide empirical support for its consistency via inter-annotator agreement analysis.
|
What kind of annotation scheme do they describe?
|
A linguistically motivated annotation scheme for ungrammatical learner English.
|
null | false
| 252
|
In this paper, we explored a sifted multi-task learning method with a novel selected sharing structure for fake news detection. The selected sharing structure fused single gate mechanism for filtering useless shared features and attention mechanism for paying close attention to features that were helpful to target tasks. We demonstrated the effectiveness of the proposed method on two public, challenging datasets and further illustrated by visualization experiments. There are several important directions remain for future research: (1) the fusion mechanism of private and shared features; (2) How to represent meta-data of fake news better to integrate into inputs.
There are several important directions remain for future research: (1) the fusion mechanism of private and shared features; (2) How to represent meta-data of fake news better to integrate into inputs.
|
What is their direction for future research?
|
(1) the fusion mechanism of private and shared features; (2) How to represent meta-data of fake news better to integrate into inputs.
|
null | false
| null |
When is the best time to buy house in USA
|
Late Spring and early Summers are considered "hot periods" in real-estate market because there is an increase in the number of listings (or supplies) during these months.
|
|
null | false
| null |
What are the best places to eat and top Japanese sushi restaurant in San Mateo?
|
The best Japanese Restaurant in San Mateo are Sushi Sam and Sushi Maruyama
|
|
null | false
| 372
|
In recent years, the micro-blogging platform Twitter has become a major social media platform with hundreds of millions of users. People turn to Twitter for a variety of purposes, from everyday chatter to reading about breaking news. The volume plus the public nature of Twitter (less than 10% of Twitter accounts are private BIBREF0 ) have made Twitter a great source of data for social and behavioural studies. These studies often require an understanding of what people are tweeting about. Though this can be coded manually, in order to take advantage of the volume of tweets available automatic analytic methods have to be used. There has been extensive work done on computational methods for analysing the linguistic content of tweets. However, there has been very little work done on classifying the pragmatics of tweets. Pragmatics looks beyond the literal meaning of an utterance and considers how context and intention contribute to meaning. A major element of pragmatics is the intended communicative act of an utterance, or what the utterance was meant to achieve. It is essential to study pragmatics in any linguistic system because at the core of linguistic analysis is studying what language is used for or what we do with language. Linguistic communication and meaning can not truly be studied without studying pragmatics. Proposed by Austin BIBREF1 and refined by Searle BIBREF2 , speech act theory can be used to study pragmatics. Amongst other things, the theory provides a formalized taxonomy BIBREF3 of a set of communicative acts, more commonly known as speech acts.
There has been extensive research done on speech act (also known as dialogue act) classification in computational linguistics, e.g., BIBREF4 . Unfortunately, these methods do not map well to Twitter, given the noisy and unconventional nature of the language used on the platform. In this work, we created a supervised speech act classifier for Twitter, using a manually annotated dataset of a few thousand tweets, in order to be better understand the meaning and intention behind tweets and uncover the rich interactions between the users of Twitter. Knowing the speech acts behind a tweet can help improve analysis of tweets and give us a better understanding of the state of mind of the users. Moreover, ws we have shown in our previous works BIBREF5 , BIBREF6 , speech act classification is essential for detection of rumors on Twitter. Finally, knowing the distribution of speech acts of tweets about a particular topic can reveal a lot about the general attitude of users about that topic (e.g., are they confused and are asking a lot of questions? Are they outraged and demanding action? Etc).
In this work, we created a supervised speech act classifier for Twitter, using a manually annotated dataset of a few thousand tweets, in order to be better understand the meaning and intention behind tweets and uncover the rich interactions between the users of Twitter
|
What kind of dataset do they use?
|
A manually annotated dataset of a few thousand tweets.
|
null | false
| 65
|
We prepared precomputed ELMo contextual embeddings for seven languages: Croatian, Estonian, Finnish, Latvian, Lithuanian, Slovenian, and Swedish. We present the necessary background on embeddings and contextual embeddings, the details of training the embedding models, and their evaluation. We show that the size of used training sets importantly affects the quality of produced embeddings, and therefore the existing publicly available ELMo embeddings for the processed languages are inadequate. We trained new ELMo embeddings on larger training sets and analysed their properties on the analogy task and on the NER task. The results show that the newly produced contextual embeddings produce substantially better results compared to the non-contextual fastText baseline. In future work, we plan to use the produced contextual embeddings on the problems of news media industry. The pretrained ELMo models will be deposited to the CLARIN repository by the time of the final version of this paper.
As future work, we will use the produced contextual embeddings on the problems of news media industry. We plan to build and evaluate more complex models, such as BERT (Devlin et al., 2019).
|
What is future work planned to solve the problems?
|
They will use the produced contextual embeddings on the problems of news media industry. They plan to build and evaluate more complex models, such as BERT (Devlin et al., 2019).
|
null | false
| null |
Identify which instrument is string or percussion: Kus, Hackbrett
|
Kus is percussion, Hackbrett is string.
|
|
null | false
| null |
Elemental magnesium is a gray-white lightweight metal, two-thirds the density of aluminium. Magnesium has the lowest melting (923 K (650 °C)) and the lowest boiling point 1,363 K (1,090 °C) of all the alkaline earth metals.
|
Given this paragraph about magnesium, what is the melting point of magnesium in celsius?
|
The melting point of magnesium is 650 °C.
|
null | false
| null |
Tamil is one of the longest-surviving classical languages of India. A. K. Ramanujan described it as "the only language of contemporary India which is recognizably continuous with a classical past". The variety and quality of classical Tamil literature has led to it being described as "one of the great classical traditions and literatures of the world". Recorded Tamil literature has been documented for over 2000 years. The earliest period of Tamil literature, Sangam literature, is dated from c. 300 BC until AD 300. It has the oldest extant literature among Dravidian languages. The earliest epigraphic records found on rock edicts and 'hero stones' date from around the 3rd century BC. About 60,000 of the approximately 100,000 inscriptions found by the Archaeological Survey of India in India are in Tamil Nadu. Of them, most are in Tamil, with only about 5 percent in other languages. Tamil language inscriptions written in Brahmi script have been discovered in Sri Lanka and on trade goods in Thailand and Egypt. The two earliest manuscripts from India, acknowledged and registered by the UNESCO Memory of the World register in 1997 and 2005, were written in Tamil.
|
How old is Tamil language?
|
Tamil language is one of the oldest and longest surviving classical language and is existing from 300 BC. Tamil as a literature has been documented for over 2000 years. Earliest epigraphic records found on rocks date back to 3rd century BC. It is also the official language of Singapore and Sri lanka apart from Tamil Nadu state in India.
|
null | false
| null |
Group the following words into two categories, 'positive' or 'negative' based on the sentiment: efficient, helpful, slow, horrible, fantastic, disappointing, confusing, lovely, excellent.
|
These words are generally considered to have positive sentiment: lovely, fantastic, excellent, efficient, helpful.
While these words are considered negative: horrible, disappointing, slow, confusing.
|
|
1707.08559
| false
| null |
Our dataset covers 218 videos from NALCS and 103 from LMS for a total of 321 videos from week 1 to week 9 in 2017 spring series from each tournament. Each week there are 10 matches for NALCS and 6 matches for LMS. Matches are best of 3, so consist of two games or three games. The first and third games are used for training. The second games in the first 4 weeks are used as validation and the remainder of second games are used as test. Table TABREF3 lists the numbers of videos in train, validation, and test subsets.
Our dataset covers 218 videos from NALCS and 103 from LMS for a total of 321 videos from week 1 to week 9 in 2017 spring series from each tournament.
|
How big was the dataset presented?
|
The answers are shown as follows:
* 321 videos
|
null | false
| null |
"El Rey" ("The King") is a 1971 song by Mexican singer José Alfredo Jiménez. It is one of his best known songs and a Latin Grammy Hall of Fame recipient. The song is about "a macho guy convinced his rough-and-tumble life doesn't preclude him from remaining the king among his peers".
A chart published by Record World credited "El Rey" as reaching number one in Mexico in 1974, a year after Jiménez's death.
|
List down some key points about the song "El Rey" from the given text
|
1. José Alfredo Jiménez, a singer from Mexico, released the song "El Rey" ("The King") in 1971.
2. It is one of his most well-known songs and was inducted into the Latin Grammy Hall of Fame.
3. The song is about "a macho guy convinced that his tough life doesn't stop him from remaining the king among his peers."
4. In 1974, a year after Jiménez's passing, "El Rey" was credited with topping a Record World list in Mexico.
|
null | false
| null |
Who are some students at Hogwarts in Gryffindor house that were in the same year as Harry Potter?
|
Hogwarts is the fictitious school in the Harry Potter book series. The books mentions several students in Harry Potter's year from the Gryffindor house, such as Dean Thomas, Seamus Finnigan, Neville Longbottom, Ron Weasley, and Hermione Granger.
|
|
null | false
| null |
Thinking about public opinion, what is the significance of “issue consistency” and “levels of conceptualization”?
|
Issue consistency is how consistent an individual American’s opinion is about an issue and is measured in two ways. Measuring across time shows that individual Americans are not likely to have the same opinion at two different points in time. For example, same-sex marriage opinions in Texas have changed from unfavorable (~30%) to favorable (over 50%) over a short amount of time from 2009 to 2017. Measuring across issue domain tells us that knowing what an individual thinks about one issue doesn’t really tell you much about what they think of another issue.
Levels of conceptualization are used to categorize people by use of ideological language in explaining their opinions. There are five levels, the first is “ideologues”, when people rely on abstract concepts to make judgements about political objects. Second, are “near-ideologues” who use ideological terminology, but don’t always use it correctly. Third, are “group interest”, where people evaluate political objects with respect to their treatment of particular groups. Fourth are “nature of the times”, when evaluations of candidates and parties are tied to general perceptions about how things are going. Lastly, “no issue content”, where focus is on personalities or family traditions without any evidence of political thinking.
|
|
null | false
| null |
On a flight, I can only pack three of the following items: sword, book, phone, suitcase, parrot, plane. Which ones should I pack?
|
I’d recommend packing a book to read, a phone for your electronic ticket, and a suitcase to hold clothes and the rest of your items. A parrot and sword won’t be allowed on a flight, and a plane will be used for the flight but not taken onboard.
|
|
null | false
| null |
What is the best party to vote for in the US?
|
Voting is a personal choice, and as an AI, I can't tell you which to prefer. The two largest parties in the US are the Democrats and Republicans, but there are many smaller parties too (e.g., Green Party and Libertarians), as well as many Independent candidates.
|
|
null | false
| null |
What is accessibility?
|
Accessibility refers to the design of products, devices, services, or environments that makes them usable by people with disabilities. This means ensuring that people with disabilities are able to access and use the same information and resources as everyone else.
|
|
null | false
| 91
|
Neural machine translation (NMT) has achieved the state-of-the-art results on a mass of language pairs with varying structural differences, such as English-French BIBREF0, BIBREF1 and Chinese-English BIBREF2. However, so far not much is known about how and why NMT works, which pose great challenges for debugging NMT models and designing optimal architectures.
The understanding of NMT models has been approached primarily from two complementary perspectives. The first thread of work aims to understand the importance of representations by analyzing the linguistic information embedded in representation vectors BIBREF3, BIBREF4 or hidden units BIBREF5, BIBREF6. Another direction focuses on understanding the importance of input words by interpreting the input-output behavior of NMT models. Previous work BIBREF7 treats NMT models as black-boxes and provides explanations that closely resemble the attention scores in NMT models. However, recent studies reveal that attention does not provide meaningful explanations since the relationship between attention scores and model output is unclear BIBREF8.
In this paper, we focus on the second thread and try to open the black-box by exploiting the gradients in NMT generation, which aims to estimate the word importance better. Specifically, we employ the integrated gradients method BIBREF9 to attribute the output to the input words with the integration of first-order derivatives. We justify the gradient-based approach via quantitative comparison with black-box methods on a couple of perturbation operations, several language pairs, and two representative model architectures, demonstrating its superiority on estimating word importance.
We analyze the linguistic behaviors of words with the importance and show its potential to improve NMT models. First, we leverage the word importance to identify input words that are under-translated by NMT models. Experimental results show that the gradient-based approach outperforms both the best black-box method and other comparative methods. Second, we analyze the linguistic roles of identified important words, and find that words of certain syntactic categories have higher importance while the categories vary across language. For example, nouns are more important for Chinese$\Rightarrow $English translation, while prepositions are more important for English-French and -Japanese translation. This finding can inspire better design principles of NMT architectures for different language pairs. For instance, a better architecture for a given language pair should consider its own language characteristics.
Experimental results show that the gradient-based approach outperforms both the best black-box method and other comparative methods.
|
Does the gradient-based approach outperform both the best black-box method and other comparative methods?
|
Yes.
|
null | false
| null |
What is one Earth day?
|
One day on Earth is the time for the Earth to completely rotate on its axis.
|
|
null | false
| null |
What are the10 best NCAA Basketball programs all-time?
|
UCLA, Kansas, Duke, North Carolina, Connecticut, Michigan State, Kentucky, Louisville, Indiana and Arizona
|
|
2003.01769
| false
| null |
In addition to the setting without any parallel data, we show results given parallel data. In Table TABREF10 we demonstrate that training the AECNN framework with mimic loss improves intelligibility over both the model trained with only time-domain loss (AECNN-T), as well as the model trained with both time-domain and spectral-domain losses (AECNN-T-SM). We only see a small improvement in the SI-SDR, likely due to the fact that the mimic loss technique is designed to improve the recognizablity of the results. In fact, seeing any improvement in SI-SDR at all is a surprising result.
FLOAT SELECTED: Table 2. Speech enhancement scores for the state-of-the-art system trained with the parallel data available in the CHiME4 corpus. Evaluation is done on channel 5 of the simulation et05 data. Mimic loss is applied to the AECNN model trained with time-domain mapping loss only, as well as time-domain and spectral magnitude mapping losses. The joint training system is done with an identical setup to the mimic system with all three losses.
In addition to the setting without any parallel data, we show results given parallel data. In Table TABREF10 we demonstrate that training the AECNN framework with mimic loss improves intelligibility over both the model trained with only time-domain loss (AECNN-T), as well as the model trained with both time-domain and spectral-domain losses (AECNN-T-SM). We only see a small improvement in the SI-SDR, likely due to the fact that the mimic loss technique is designed to improve the recognizablity of the results. In fact, seeing any improvement in SI-SDR at all is a surprising result.
FLOAT SELECTED: Table 2. Speech enhancement scores for the state-of-the-art system trained with the parallel data available in the CHiME4 corpus. Evaluation is done on channel 5 of the simulation et05 data. Mimic loss is applied to the AECNN model trained with time-domain mapping loss only, as well as time-domain and spectral magnitude mapping losses. The joint training system is done with an identical setup to the mimic system with all three losses.
|
By how much does using phonetic feedback improve state-of-the-art systems?
|
Improved AECNN-T by 2.1 and AECNN-T-SM BY 0.9
|
null | false
| null |
It is likely that the area around Sigiriya may have been inhabited since prehistoric times. There is clear evidence that the many rock shelters and caves in the vicinity were occupied by Buddhist monks and ascetics from as early as the 3rd century BC. The earliest evidence of human habitation at Sigiriya is the Aligala rock shelter to the east of Sigiriya rock, indicating that the area was occupied nearly five thousand years ago during the Mesolithic Period.
Buddhist monastic settlements were established during the 3rd century BC in the western and northern slopes of the boulder-strewn hills surrounding the Sigiriya rock. Several rock shelters or caves were created during this period. These shelters were made under large boulders, with carved drip ledges around the cave mouths. Rock inscriptions are carved near the drip ledges on many of the shelters, recording the donation of the shelters to the Buddhist monastic order as residences. These were made in the period between the 3rd century BCE and the 1st century AD.
In 477 AD, Kashyapa I, the king’s son by a non-royal consort, seized the throne from King Dhatusena, following a coup assisted by Migara, the King’s nephew and army commander. The rightful heir, Moggallana, fearing for his life, fled to South India. Afraid of an attack by Moggallana, Kashyapa moved the capital and his residence from the traditional capital of Anuradhapura to the more secure Sigiriya. During King Kashyapa’s reign (477 to 495 AD), Sigiriya was developed into a complex city and fortress. Most of the elaborate constructions on the rock summit and around it, including defensive structures, palaces, and gardens, date from this period.
|
Who built Sigiriya fortress in Sri Lanka?
|
King Kashyapa developed Sigiriya city and fortress.
|
null | false
| null |
Westerville is a city in Franklin and Delaware counties in the U.S. state of Ohio. A northeastern suburb of Columbus, the population was 39,190 at the 2020 census. Westerville is the home of Otterbein University. Westerville was once known as "The Dry Capital of the World" for its strict laws prohibiting sales of alcohol and for being the home of the Anti-Saloon League, one of the driving forces behind Prohibition at the beginning of the 20th century.
|
What was "the Dry Capital of the World"?
|
Westerville, OH was once known as "The Dry Capital of the World" due to being the home of the Anti-Saloon League, a driving force behind Prohibition in the early 1900s, and for its strict laws prohibiting sales of alcohol. Ironically, there are now many restaurants and a small brewery in Westerville that serve alcohol.
|
null | false
| null |
How to be a good human being?
|
Always be thoughtful of others. Put yourselves in the other persons shoes. Do not judge anyone based on their color, appearance, ethnicity, education, job etc. Their actions may not look appropriate , but a lot of times it is because they do not know and they are still learning. Be kind and help such folks. If everyone starts doing this , the world would be a much happier place. The happiest places in the world are not with the best weather and infrastructure, it is the happiness of the people living there.
|
|
null | false
| null |
In the series A Song of Ice and Fire, who is the founder of House Qoherys?
|
Quenton Qoherys
|
|
null | false
| null |
When will we go interplanetary?
|
Humans already have successful missions to other planets. There are project that are in progress across world that should put humans in Mars in next decade. If you ask me, I am already in Mars and attempting to go intergalaxy.
|
|
null | false
| 187
|
Datasets: We use two recent benchmark datasets WN18RR BIBREF17 and FB15k-237 BIBREF18 . These two datasets are created to avoid reversible relation problems, thus the prediction task becomes more realistic and hence more challenging BIBREF18 . Table TABREF7 presents the statistics of WN18RR and FB15k-237.
Evaluation protocol: Following BIBREF3 , for each valid test triple INLINEFORM0 , we replace either INLINEFORM1 or INLINEFORM2 by each of all other entities to create a set of corrupted triples. We use the “Filtered” setting protocol BIBREF3 , i.e., not taking any corrupted triples that appear in the KG into accounts. We rank the valid test triple and corrupted triples in descending order of their scores. We employ evaluation metrics: mean rank (MR), mean reciprocal rank (MRR) and Hits@10 (i.e., the proportion of the valid test triples ranking in top 10 predictions). Lower MR, higher MRR or higher Hits@10 indicate better performance. Final scores on the test set are reported for the model obtaining the highest Hits@10 on the validation set.
Training protocol: We use the common Bernoulli strategy BIBREF20 , BIBREF21 when sampling invalid triples. For WN18RR, BIBREF22 found a strong evidence to support the necessity of a WordNet-related semantic setup, in which they averaged pre-trained word embeddings for word surface forms within the WordNet to create synset embeddings, and then used these synset embeddings to initialize entity embeddings for training their TransE association model. We follow this evidence in using the pre-trained 100-dimensional Glove word embeddings BIBREF23 to train a TransE model on WN18RR.
We employ the TransE and ConvKB implementations provided by BIBREF24 and BIBREF15 . For ConvKB, we use a new process of training up to 100 epochs and monitor the Hits@10 score after every 10 training epochs to choose optimal hyper-parameters with the Adam initial learning rate in INLINEFORM0 and the number of filters INLINEFORM1 in INLINEFORM2 . We obtain the highest Hits@10 scores on the validation set when using N= 400 and the initial learning rate INLINEFORM3 on WN18RR; and N= 100 and the initial learning rate INLINEFORM4 on FB15k-237.
Like in ConvKB, we use the same pre-trained entity and relation embeddings produced by TransE to initialize entity and relation embeddings in our CapsE for both WN18RR and FB15k-237 ( INLINEFORM0 ). We set the batch size to 128, the number of neurons within the capsule in the second capsule layer to 10 ( INLINEFORM1 ), and the number of iterations in the routing algorithm INLINEFORM2 in INLINEFORM3 . We run CapsE up to 50 epochs and monitor the Hits@10 score after each 10 training epochs to choose optimal hyper-parameters. The highest Hits@10 scores for our CapsE on the validation set are obtained when using INLINEFORM4 , INLINEFORM5 and the initial learning rate at INLINEFORM6 on WN18RR; and INLINEFORM7 , INLINEFORM8 and the initial learning rate at INLINEFORM9 on FB15k-237.
Dataset: We use the SEARCH17 dataset BIBREF12 of query logs of 106 users collected by a large-scale web search engine. A log entity consists of a user identifier, a query, top-10 ranked documents returned by the search engine and clicked documents along with the user's dwell time. BIBREF12 constructed short-term (session-based) user profiles and used the profiles to personalize the returned results. They then employed the SAT criteria BIBREF26 to identify whether a returned document is relevant from the query logs as either a clicked document with a dwell time of at least 30 seconds or the last clicked document in a search session (i.e., a SAT click). After that, they assigned a INLINEFORM0 label to a returned document if it is a SAT click and also assigned INLINEFORM1 labels to the remaining top-10 documents. The rank position of the INLINEFORM2 labeled documents is used as the ground truth to evaluate the search performance before and after re-ranking.
The dataset was uniformly split into the training, validation and test sets. This split is for the purpose of using historical data in the training set to predict new data in the test set BIBREF12 . The training, validation and test sets consist of 5,658, 1,184 and 1,210 relevant (i.e., valid) triples; and 40,239, 7,882 and 8,540 irrelevant (i.e., invalid) triples, respectively.
Evaluation protocol: Our CapsE is used to re-rank the original list of documents returned by a search engine as follows: (i) We train our model and employ the trained model to calculate the score for each INLINEFORM0 triple. (ii) We then sort the scores in the descending order to obtain a new ranked list. To evaluate the performance of our proposed model, we use two standard evaluation metrics: mean reciprocal rank (MRR) and Hits@1. For each metric, the higher value indicates better ranking performance.
We compare CapsE with the following baselines using the same experimental setup: (1) SE: The original rank is returned by the search engine. (2) CI BIBREF27 : This baseline uses a personalized navigation method based on previously clicking returned documents. (3) SP BIBREF9 , BIBREF11 : A search personalization method makes use of the session-based user profiles. (4) Following BIBREF12 , we use TransE as a strong baseline model for the search personalization task. Previous work shows that the well-known embedding model TransE, despite its simplicity, obtains very competitive results for the knowledge graph completion BIBREF28 , BIBREF29 , BIBREF14 , BIBREF30 , BIBREF15 . (5) The CNN-based model ConvKB is the most closely related model to our CapsE.
Embedding initialization: We follow BIBREF12 to initialize user profile, query and document embeddings for the baselines TransE and ConvKB, and our CapsE.
We train a LDA topic model BIBREF31 with 200 topics only on the relevant documents (i.e., SAT clicks) extracted from the query logs. We then use the trained LDA model to infer the probability distribution over topics for every returned document. We use the topic proportion vector of each document as its document embedding (i.e. INLINEFORM0 ). In particular, the INLINEFORM1 element ( INLINEFORM2 ) of the vector embedding for document INLINEFORM3 is: INLINEFORM4 where INLINEFORM5 is the probability of the topic INLINEFORM6 given the document INLINEFORM7 .
We also represent each query by a probability distribution vector over topics. Let INLINEFORM0 be the set of top INLINEFORM1 ranked documents returned for a query INLINEFORM2 (here, INLINEFORM3 ). The INLINEFORM4 element of the vector embedding for query INLINEFORM5 is defined as in BIBREF12 : INLINEFORM6 , where INLINEFORM7 is the exponential decay function of INLINEFORM8 which is the rank of INLINEFORM9 in INLINEFORM10 . And INLINEFORM11 is the decay hyper-parameter ( INLINEFORM12 ). Following BIBREF12 , we use INLINEFORM13 . Note that if we learn query and document embeddings during training, the models will overfit to the data and will not work for new queries and documents. Thus, after the initialization process, we fix (i.e., not updating) query and document embeddings during training for TransE, ConvKB and CapsE.
In addition, as mentioned by BIBREF9 , the more recently clicked document expresses more about the user current search interest. Hence, we make use of the user clicked documents in the training set with the temporal weighting scheme proposed by BIBREF11 to initialize user profile embeddings for the three embedding models.
Hyper-parameter tuning: For our CapsE model, we set batch size to 128, and also the number of neurons within the capsule in the second capsule layer to 10 ( INLINEFORM0 ). The number of iterations in the routing algorithm is set to 1 ( INLINEFORM1 ). For the training model, we use the Adam optimizer with the initial learning rate INLINEFORM2 INLINEFORM3 INLINEFORM4 INLINEFORM5 INLINEFORM6 INLINEFORM7 . We also use ReLU as the activation function INLINEFORM8 . We select the number of filters INLINEFORM9 . We run the model up to 200 epochs and perform a grid search to choose optimal hyper-parameters on the validation set. We monitor the MRR score after each training epoch and obtain the highest MRR score on the validation set when using INLINEFORM10 and the initial learning rate at INLINEFORM11 .
We employ the TransE and ConvKB implementations provided by BIBREF24 and BIBREF15 and then follow their training protocols to tune hyper-parameters for TransE and ConvKB, respectively. We also monitor the MRR score after each training epoch and attain the highest MRR score on the validation set when using margin = 5, INLINEFORM0 -norm and SGD learning rate at INLINEFORM1 for TransE; and INLINEFORM2 and the Adam initial learning rate at INLINEFORM3 for ConvKB.
The dataset was uniformly split into the training, validation and test sets. This split is for the purpose of using historical data in the training set to predict new data in the test set (Vu et al., 2017).
|
Why the dataset was uniformly split into the training, validation, and test sets?
|
This split is for the purpose of using historical data in the training set to predict new data in the test set.
|
1909.09268
| false
| null |
In this part, we discuss three significant limitations of BLEU and ROUGE. These metrics can assign: High scores to semantically opposite translations/summaries, Low scores to semantically related translations/summaries and High scores to unintelligible translations/summaries.
Challenges with BLEU and ROUGE ::: High score, opposite meanings
Suppose that we have a reference summary s1. By adding a few negation terms to s1, one can create a summary s2 which is semantically opposite to s1 but yet has a high BLEU/ROUGE score.
Challenges with BLEU and ROUGE ::: Low score, similar meanings
In addition not to be sensitive to negation, BLEU and ROUGE score can give low scores to sentences with equivalent meaning. If s2 is a paraphrase of s1, the meaning will be the same ;however, the overlap between words in s1 and s2 will not necessarily be significant.
Challenges with BLEU and ROUGE ::: High score, unintelligible sentences
A third weakness of BLEU and ROUGE is that in their simplest implementations, they are insensitive to word permutation and can give very high scores to unintelligible sentences. Let s1 be "On a morning, I saw a man running in the street." and s2 be “On morning a, I saw the running a man street”. s2 is not an intelligible sentence. The unigram version of ROUGE and BLEU will give these 2 sentences a score of 1.
In this part, we discuss three significant limitations of BLEU and ROUGE. These metrics can assign: High scores to semantically opposite translations/summaries, Low scores to semantically related translations/summaries and High scores to unintelligible translations/summaries.
Challenges with BLEU and ROUGE ::: High score, opposite meanings
Suppose that we have a reference summary s1. By adding a few negation terms to s1, one can create a summary s2 which is semantically opposite to s1 but yet has a high BLEU/ROUGE score.
Challenges with BLEU and ROUGE ::: Low score, similar meanings
In addition not to be sensitive to negation, BLEU and ROUGE score can give low scores to sentences with equivalent meaning. If s2 is a paraphrase of s1, the meaning will be the same ;however, the overlap between words in s1 and s2 will not necessarily be significant.
Challenges with BLEU and ROUGE ::: High score, unintelligible sentences
A third weakness of BLEU and ROUGE is that in their simplest implementations, they are insensitive to word permutation and can give very high scores to unintelligible sentences. Let s1 be "On a morning, I saw a man running in the street." and s2 be “On morning a, I saw the running a man street”. s2 is not an intelligible sentence. The unigram version of ROUGE and BLEU will give these 2 sentences a score of 1.
|
What are the three limitations?
|
The answers are shown as follows:
* High scores to semantically opposite translations/summaries, Low scores to semantically related translations/summaries and High scores to unintelligible translations/summaries.
|
null | false
| null |
What are the best types of questions to ask during a job interview?
|
The best types of questions to ask during an interview will depend on the position being filled, the required experience and expertise, and the specific skill set sought. It is important to ask background questions to understand the candidate’s current role and responsibilities, as well as their ideal next job. Behavioral questions are also quite useful during an interview. Behavioral questions ask a candidate to describe how they would respond in a specific situation and can be tailored to the skills and experience sought. For example, it is useful to describe a situation the candidate is likely to encounter at the potential new job and ask them to describe how they would handle the situation. This will help you identify whether the candidate has the skills and experience to address some of the day-to-day work at your company. It is a good idea to ask multiple behavioral questions during each interview and to ask all candidates for that role the same questions.
|
|
null | false
| null |
McDonald's Corporation is an American multinational fast food chain, founded in 1940 as a restaurant operated by Richard and Maurice McDonald, in San Bernardino, California, United States. They rechristened their business as a hamburger stand, and later turned the company into a franchise, with the Golden Arches logo being introduced in 1953 at a location in Phoenix, Arizona. In 1955, Ray Kroc, a businessman, joined the company as a franchise agent and proceeded to purchase the chain from the McDonald brothers. McDonald's had its previous headquarters in Oak Brook, Illinois, but moved its global headquarters to Chicago in June 2018.
McDonald's is the world's largest fast food restaurant chain, serving over 69 million customers daily in over 100 countries in more than 40,000 outlets as of 2021. McDonald's is best known for its hamburgers, cheeseburgers and french fries, although their menu also includes other items like chicken, fish, fruit, and salads. Their best-selling licensed item are their french fries, followed by the Big Mac. The McDonald's Corporation revenues come from the rent, royalties, and fees paid by the franchisees, as well as sales in company-operated restaurants. McDonald's is the world's second-largest private employer with 1.7 million employees (behind Walmart with 2.3 million employees). As of 2022, McDonald's has the sixth-highest global brand valuation.
McDonald's has been subject to criticism over the health effects of its products, its treatment of employees, and other business practices.
|
Based on the reference text about McDonalds, tell me what year it was founded, how many daily customers they have, and their best selling product.
|
McDonalds was founded in 1940 and serves over 69 million customers a day as of 2021. It's best selling product are their french fries.
|
null | false
| 112
|
Many tasks related to sentiment or emotion analysis depend upon affect, opinion, sentiment, sense and emotion lexicons. These lexicons associate words to corresponding sentiment or emotion metrics. On the other hand, the semantic meaning of words, sentences, and documents are preserved and compactly represented using low dimensional vectors BIBREF1 instead of one hot encoding vectors which are sparse and high dimensional. Finally, there are traditional NLP features like word N-grams, character N-grams, Part-Of-Speech N-grams and word clusters which are known to perform well on various tasks.
Based on these observations, the feature extraction step is implemented as a union of different independent feature extractors (featurizers) in a light-weight and easy to use Python program EmoInt . It comprises of all features available in the baseline model BIBREF2 along with additional feature extractors and bi-gram support. Fourteen such feature extractors have been implemented which can be clubbed into 3 major categories:
[noitemsep]
Lexicon Features
Word Vectors
Syntax Features
Lexicon Features: AFINN BIBREF3 word list are manually rated for valence with an integer between -5 (Negative Sentiment) and +5 (Positive Sentiment). Bing Liu BIBREF4 opinion lexicon extract opinion on customer reviews. +/-EffectWordNet BIBREF5 by MPQA group are sense level lexicons. The NRC Affect Intensity BIBREF6 lexicons provide real valued affect intensity. NRC Word-Emotion Association Lexicon BIBREF7 contains 8 sense level associations (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) and 2 sentiment level associations (negative and positive). Expanded NRC Word-Emotion Association Lexicon BIBREF8 expands the NRC word-emotion association lexicon for twitter specific language. NRC Hashtag Emotion Lexicon BIBREF9 contains emotion word associations computed on emotion labeled twitter corpus via Hashtags. NRC Hashtag Sentiment Lexicon and Sentiment140 Lexicon BIBREF10 contains sentiment word associations computed on twitter corpus via Hashtags and Emoticons. SentiWordNet BIBREF11 assigns to each synset of WordNet three sentiment scores: positivity, negativity, objectivity. Negation lexicons collections are used to count the total occurrence of negative words. In addition to these, SentiStrength BIBREF12 application which estimates the strength of positive and negative sentiment from tweets is also added.
Word Vectors: We focus primarily on the word vector representations (word embeddings) created specifically using the twitter dataset. GloVe BIBREF13 is an unsupervised learning algorithm for obtaining vector representations for words. 200-dimensional GloVe embeddings trained on 2 Billion tweets are integrated. Edinburgh embeddings BIBREF14 are obtained by training skip-gram model on Edinburgh corpus BIBREF15 . Since tweets are abundant with emojis, Emoji embeddings BIBREF16 which are learned from the emoji descriptions have been used. Embeddings for each tweet are obtained by summing up individual word vectors and then dividing by the number of tokens in the tweet.
Syntactic Features: Syntax specific features such as Word N-grams, Part-Of-Speech N-grams BIBREF17 , Brown Cluster N-grams BIBREF18 obtained using TweetNLP project have been integrated into the system.
The final feature vector is the concatenation of all the individual features. For example, we concatenate average word vectors, sum of NRC Affect Intensities, number of positive and negative Bing Liu lexicons, number of negation words and so on to get final feature vector. The scaling of final features is not required when used with gradient boosted trees. However, scaling steps like standard scaling (zero mean and unit normal) may be beneficial for neural networks as the optimizers work well when the data is centered around origin.
A total of fourteen different feature extractors have been implemented, all of which can be enabled or disabled individually to extract features from a given tweet.
Fourteen such feature extractors have been implemented which can be clubbed into 3 major categories: • Lexicon Features • Word Vectors • Syntax Features
|
What are the three categories of feature extractors?
|
• Lexicon Features • Word Vectors • Syntax Features
|
null | false
| null |
Name some famous Schubert's pieces
|
In his esteemed but rather short life span, Schubert wrote some of the most beautiful and poetic compositions. Some of the well renowned compositions are "Piano Quintet In A", "Winterreise", "Piano Sonata No.21", and "Symphony No.8 In B minor" which is also referred as 'Unfinished' Symphony because Schubert unfortunately died right before completing it.
|
|
null | false
| null |
What is home automation and how can it be used?
|
Home automation is the process of installing various interactive components in a home as a way to centralize the management of these components through an integrated interface such as a phone, table, computer, or dedicated virtual assistant such as Alexa. The advantage of home automation is that the various arrays of smart devices available on the market today can be programmed on a daily, weekly, or hourly schedule. These devices can include lights, switches, plugs, appliances, televisions, garage door openers, locks, and security cameras. Through voice assistants like Alexa you can name your devices and verbally issue commands that the voice assistant will use to action with the targeted smart device. The applications for this technology are extensive including allowing for remote monitoring for security purposes, ensuring minimal energy consumption in your home, providing remote access to your home while away, and simply automating the use of your residence.
|
|
null | false
| 37
|
Recent years have seen unprecedented progress for Natural Language Processing (NLP) on almost every NLP subtask. Even though low-resource settings have also been explored, this progress has overwhelmingly been observed in languages with significant data resources that can be leveraged to train deep neural networks. Low-resource languages still lag behind.
Endangered languages pose an additional challenge. The process of documenting an endangered language typically includes the creation of word lists, audio and video recordings, notes, or grammar fragments, with the created resources then stored in large online linguistics archives. This process is often hindered by the Transcription Bottleneck: the linguistic fieldworker and the language community may not have time to transcribe all of the recordings and may only transcribe segments that are linguistically salient for publication or culturally significant for the creation of community resources.
With this work we make publicly available a large corpus in Mapudungun, a language of the indigenous Mapuche people of southern Chile and western Argentina. We hope to ameliorate the resource gap and the transcription bottleneck in two ways. First, we are providing a larger data set than has previously been available, and second, we are providing baselines for NLP tasks (speech recognition, speech synthesis, and machine translation). In providing baselines and datasets splits, we hope to further facilitate research on low-resource NLP for this language through our data set. Research on low-resource speech recognition is particularly important in relieving the transcription bottleneck, while tackling the research challenges that speech synthesis and machine translation pose for such languages could lead to such systems being deployed to serve more under-represented communities.
With this work we make publicly available a large corpus in Mapudungun, a language of the indigenous Mapuche people of southern Chile and western Argentina. We hope to ameliorate the resource gap and the transcription bottleneck in two ways.
|
What is the problem that the authors hope to ameliorate?
|
The authors hope to ameliorate the resource gap and the transcription bottleneck of Mapudungun.
|
null | false
| 24
|
Reading Comprehension (RC) has become a central task in natural language processing, with great practical value in various industries. In recent years, many large-scale RC datasets in English BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6 have nourished the development of numerous powerful and diverse RC models BIBREF7, BIBREF8, BIBREF9, BIBREF10, BIBREF11. The state-of-the-art model BIBREF12 on SQuAD, one of the most widely used RC benchmarks, even surpasses human-level performance. Nonetheless, RC on languages other than English has been limited due to the absence of sufficient training data. Although some efforts have been made to create RC datasets for Chinese BIBREF13, BIBREF14 and Korean BIBREF15, it is not feasible to collect RC datasets for every language since annotation efforts to collect a new RC dataset are often far from trivial. Therefore, the setup of transfer learning, especially zero-shot learning, is of extraordinary importance.
Existing methods BIBREF16 of cross-lingual transfer learning on RC datasets often count on machine translation (MT) to translate data from source language into target language, or vice versa. These methods may not require a well-annotated RC dataset for the target language, whereas a high-quality MT model is needed as a trade-off, which might not be available when it comes to low-resource languages.
In this paper, we leverage pre-trained multilingual language representation, for example, BERT learned from multilingual un-annotated sentences (multi-BERT), in cross-lingual zero-shot RC. We fine-tune multi-BERT on the training set in source language, then test the model in target language, with a number of combinations of source-target language pair to explore the cross-lingual ability of multi-BERT. Surprisingly, we find that the models have the ability to transfer between low lexical similarity language pair, such as English and Chinese. Recent studies BIBREF17, BIBREF12, BIBREF18 show that cross-lingual language models have the ability to enable preliminary zero-shot transfer on simple natural language understanding tasks, but zero-shot transfer of RC has not been studied. To our knowledge, this is the first work systematically exploring the cross-lingual transferring ability of multi-BERT on RC tasks.
In this paper, we leverage pre-trained multilingual language representation, for example, BERT learned from multilingual un-annotated sentences (multi-BERT), in cross-lingual zero-shot RC. We fine-tune multi-BERT on the training set in source language, then test the model in target language, with a number of combinations of source-target language pair to explore the cross-lingual ability of multi-BERT.
|
What representation is used in the paper?
|
Pre-trained multilingual language representation.
|
1811.01088
| true
| null |
FLOAT SELECTED: Table 1: GLUE results with and without STILTs, fine-tuning on full training data of each target task. Bold marks the best within each section. Strikethrough indicates cases where the intermediate task is the same as the target task—we substitute the baseline result for that cell. A.Ex is the average excluding MNLI and QQP because of the overlap with intermediate tasks. See text for discussion of WNLI results. Test results on STILTs uses the supplementary training regime for each task based on the performance on the development set, corresponding to the numbers shown in Best of Each. The aggregated GLUE scores differ from the public leaderboard because we report performance on QNLIv1.
FLOAT SELECTED: Table 1: GLUE results with and without STILTs, fine-tuning on full training data of each target task. Bold marks the best within each section. Strikethrough indicates cases where the intermediate task is the same as the target task—we substitute the baseline result for that cell. A.Ex is the average excluding MNLI and QQP because of the overlap with intermediate tasks. See text for discussion of WNLI results. Test results on STILTs uses the supplementary training regime for each task based on the performance on the development set, corresponding to the numbers shown in Best of Each. The aggregated GLUE scores differ from the public leaderboard because we report performance on QNLIv1.
|
Is the new model evaluated on the tasks that BERT and ELMo are evaluated on?
|
Yes.
|
null | false
| null |
Which is a species of fish? Slipmouth or Slipknot
|
Slipmouth
|
|
null | false
| null |
What does BAFTA stands for ?
|
British Academy of Film and Television Arts
|
|
null | false
| null |
Are you going for a walk today?
|
These rainy days haven't been the greatest, but now with that shining sun, it'd be a shame to stay in!
|
|
null | false
| null |
Who prepares the Global Hunger Index?
|
Global Hunger Index is jointly prepared by Welhunger Life and Concern Worldwide
|
|
null | false
| null |
The Supermarine Spitfire is a British single-seat fighter aircraft used by the Royal Air Force and other Allied countries before, during, and after World War II. Many variants of the Spitfire were built, from the Mk 1 to the Rolls-Royce Griffon-engined Mk 24 using several wing configurations and guns. It was the only British fighter produced continuously throughout the war. The Spitfire remains popular among enthusiasts; around 70 remain airworthy, and many more are static exhibits in aviation museums throughout the world.
|
How many Spitfire fighter aircraft remain airworthy
|
around 70 remain airworthy
|
null | false
| 77
|
The corpus consists of 53 documents, which contain an average number of 156.1 sentences per document, each with 19.55 tokens on average. The corpus comprises 8,275 sentences and 167,739 words in total. However, as mentioned above, only case presentation sections, headings and abstracts are annotated. The numbers of annotated entities are summarized in Table TABREF24.
Findings are the most frequently annotated type of entity. This makes sense given that findings paint a clinical picture of the patient's condition. The number of tokens per entity ranges from one token for all types to 5 tokens for cases (average length 3.1), nine tokens for conditions (average length 2.0), 16 tokens for factors (average length 2.5), 25 tokens for findings (average length 2.6) and 18 tokens for modifiers (average length 1.4) (cf. Table TABREF24). Examples of rather long entities are given in Table TABREF25.
Entities can appear in a discontinuous way. We model this as a relation between two spans which we call “discontinuous” (cf. Figure FIGREF26). Especially findings often appear as discontinuous entities, we found 543 discontinuous finding relations. The numbers for conditions and factors are lower with seven and two, respectively. Entities can also be nested within one another. This happens either when the span of one annotation is completely embedded in the span of another annotation (fully-nested; cf. Figure FIGREF12), or when there is a partial overlapping between the spans of two different entities (partially-nested; cf. Figure FIGREF12). There is a high number of inter-sentential relations in the corpus (cf. Table TABREF27). This can be explained by the fact that the case entity occurs early in each document; furthermore, it is related to finding and factor annotations that are distributed across different sentences.
The most frequently annotated relation in our corpus is the has-relation between a case entity and the findings related to that case. This correlates with the high number of finding entities. The relations contained in our corpus are summarized in Table TABREF27.
The corpus consists of 53 documents, which contain an average number of 156.1 sentences per document, each with 19.55 tokens on average. The corpus comprises 8,275 sentences and 167,739 words in total.
|
How large is the corpus?
|
The corpus consists of 53 documents, which contain an average number of 156.1 sentences per document, each with 19.55 tokens on average. The corpus comprises 8,275 sentences and 167,739 words in total.
|
null | false
| 98
|
Lexical analysis, syntactic analysis, semantic analysis, disclosure analysis and pragmatic analysis are five main steps in natural language processing BIBREF0 , BIBREF1 . While morphology is a basic task in lexical analysis of English, word segmentation is considered a basic task in lexical analysis of Vietnamese and other East Asian languages processing. This task is to determine borders between words in a sentence. In other words, it is segmenting a list of tokens into a list of words such that words are meaningful.
Word segmentation is the primary step in prior to other natural language processing tasks i. e., term extraction and linguistic analysis (as shown in Figure 1). It identifies the basic meaningful units in input texts which will be processed in the next steps of several applications. For named entity recognization BIBREF2 , word segmentation chunks sentences in input documents into sequences of words before they are further classified in to named entity classes. For Vietnamese language, words and candidate terms can be extracted from Vietnamese copora (such as books, novels, news, and so on) by using a word segmentation tool. Conformed features and context of these words and terms are used to identify named entity tags, topic of documents, or function words. For linguistic analysis, several linguistic features from dictionaries can be used either to annotating POS tags or to identifying the answer sentences. Moreover, language models can be trained by using machine learning approaches and be used in tagging systems, like the named entity recognization system of Tran et al. BIBREF2 .
Many studies forcus on word segmentation for Asian languages, such as: Chinese, Japanese, Burmese (Myanmar) and Thai BIBREF3 , BIBREF4 , BIBREF5 , BIBREF6 . Approaches for word segmentation task are variety, from lexicon-based to machine learning-based methods. Recently, machine learning-based methods are used widely to solve this issue, such as: Support Vector Machine or Conditional Random Fields BIBREF7 , BIBREF8 . In general, Chinese is a language which has the most studies on the word segmentation issue. However, there is a lack of survey of word segmentation studies on Asian languages and Vietnamese as well. This paper aims reviewing state-of-the-art word segmentation approaches and systems applying for Vietnamese. This study will be a foundation for studies on Vietnamese word segmentation and other following Vietnamese tasks as well, such as part-of-speech tagger, chunker, or parser systems.
There are several studies about the Vietnamese word segmentation task over the last decade. Dinh et al. started this task with Weighted Finite State Transducer (WFST) approach and Neural Network approach BIBREF9 . In addition, machine learning approaches are studied and widely applied to natural language processing and word segmentation as well. In fact, several studies used support vector machines (SVM) and conditional random fields (CRF) for the word segmentation task BIBREF7 , BIBREF8 . Based on annotated corpora and token-based features, studies used machine learning approaches to build word segmentation systems with accuracy about 94%-97%.
According to our observation, we found that is lacks of complete review approaches, datasets and toolkits which we recently used in Vietnamese word segmentation. A all sided review of word segmentation will help next studies on Vietnamese natural language processing tasks have an up-to-date guideline and choose the most suitable solution for the task. The remaining part of the paper is organized as follows. Section II discusses building corpus in Vietnamese, containing linguistic issues and the building progress. Section III briefly mentions methods to model sentences and text in machine learning systems. Next, learning models and approaches for labeling and segmenting sequence data will be presented in Section IV. Section V mainly addresses two existing toolkits, vnTokenizer and JVnSegmenter, for Vietnamese word segmentation. Several experiments based on mentioned approaches and toolkits are described in Section VI. Finally, conclusions and future works are given in Section VII.
According to our observation, we found that is lacks of complete review approaches, datasets and toolkits which we recently used in Vietnamese word segmentation.
|
What is there lack of according to their observation?
|
Complete review approaches, datasets and toolkits which are used in Vietnamese word segmentation.
|
null | false
| 328
|
We compare performances of our model with the implementation of BIBREF13 and the results are shown in Table TABREF43 . Our model obtains better performances in Multi-Domain scenario with an average improvement of 4.5%, where datasets are product reviews on different domains with similar sequence lengths and the same class number, thus producing stronger correlations. Multi-Cardinality scenario also achieves significant improvements of 2.77% on average, where datasets are movie reviews with different cardinalities.
However, Multi-Objective scenario benefits less from multi-task learning due to lacks of salient correlation among sentiment, topic and question type. The QC dataset aims to classify each question into six categories and its performance even gets worse, which may be caused by potential noises introduced by other tasks. In practice, the structure of our model is flexible, as couplings and fusions between some empirically unrelated tasks can be removed to alleviate computation costs.
We further explore the influence of INLINEFORM0 in TOS on our model, which can be any positive integer. A higher value means larger and more various samples combinations, while requires higher computation costs.
Figure FIGREF45 shows the performances of datasets in Multi-Domain scenario with different INLINEFORM0 . Compared to INLINEFORM1 , our model can achieve considerable improvements when INLINEFORM2 as more samples combinations are available. However, there are no more salient gains as INLINEFORM3 gets larger and potential noises from other tasks may lead to performance degradations. For a trade-off between efficiency and effectiveness, we determine INLINEFORM4 as the optimal value for our experiments.
In order to measure the correlation strength between two task INLINEFORM0 and INLINEFORM1 , we learn them jointly with our model and define Pair-wise Performance Gain as INLINEFORM2 , where INLINEFORM3 are the performances of tasks INLINEFORM4 and INLINEFORM5 when learned individually and jointly.
We calculate PPGs for every two tasks in Table TABREF35 and illustrate the results in Figure FIGREF47 , where darkness of colors indicate strength of correlation. It is intuitive that datasets of Multi-Domain scenario obtain relatively higher PPGs with each other as they share similar cardinalities and abundant low-level linguistic characteristics. Sentences of QC dataset are much shorter and convey unique characteristics from other tasks, thus resulting in quite lower PPGs.
Our model obtains better performances in MultiDomain scenario with an average improvement of 4.5%, where datasets are product reviews on different domains with similar sequence lengths and the same class number, thus producing stronger correlations.
|
Does the authors' model obtain better performances in the MultiDomain scenario with an average improvement of 4.5%?
|
Yes, it does.
|
null | false
| null |
The arrival of the 20th century brought a convergence of economic factors that fueled rapid growth in Houston, including a burgeoning port and railroad industry, the decline of Galveston as Texas's primary port following a devastating 1900 hurricane, the subsequent construction of the Houston Ship Channel, and the Texas oil boom. In the mid-20th century, Houston's economy diversified, as it became home to the Texas Medical Center—the world's largest concentration of healthcare and research institutions—and NASA's Johnson Space Center, home to the Mission Control Center.
|
What professions are likely to be pursued in Houston?
|
Houston appears to be a good place to be a doctor, aerospace engineer, or researcher. In the past it may have been a hotbed for workers in the oil or transportation industries.
|
1606.00189
| false
| null |
Recently, continuous space representations of words and phrases have been incorporated into SMT systems via neural networks. Specifically, addition of monolingual neural network language models BIBREF13 , BIBREF14 , neural network joint models (NNJM) BIBREF4 , and neural network global lexicon models (NNGLM) BIBREF3 have been shown to be useful for SMT. Neural networks have been previously used for GEC as a language model feature in the classification approach BIBREF15 and as a classifier for article error correction BIBREF16 . Recently, a neural machine translation approach has been proposed for GEC BIBREF17 . This method uses a recurrent neural network to perform sequence-to-sequence mapping from erroneous to well-formed sentences. Additionally, it relies on a post-processing step based on statistical word-based translation models to replace out-of-vocabulary words. In this paper, we investigate the effectiveness of two neural network models, NNGLM and NNJM, in SMT-based GEC. To the best of our knowledge, there is no prior work that uses these two neural network models for SMT-based GEC.
Grammatical error correction (GEC) is a challenging task due to the variability of the type of errors and the syntactic and semantic dependencies of the errors on the surrounding context. Most of the grammatical error correction systems use classification and rule-based approaches for correcting specific error types. However, these systems use several linguistic cues as features. The standard linguistic analysis tools like part-of-speech (POS) taggers and parsers are often trained on well-formed text and perform poorly on ungrammatical text. This introduces further errors and limits the performance of rule-based and classification approaches to GEC. As a consequence, the phrase-based statistical machine translation (SMT) approach to GEC has gained popularity because of its ability to learn text transformations from erroneous text to correct text from error-corrected parallel corpora without any additional linguistic information. They are also not limited to specific error types. Currently, many state-of-the-art GEC systems are based on SMT or use SMT components for error correction BIBREF0 , BIBREF1 , BIBREF2 . In this paper, grammatical error correction includes correcting errors of all types, including word choice errors and collocation errors which constitute a large class of learners' errors.
We conduct experiments by incorporating NNGLM and NNJM both independently and jointly into our baseline system. The results of our experiments are described in Section SECREF23 . The evaluation is performed similar to the CoNLL 2014 shared task setting using the the official test data of the CoNLL 2014 shared task with annotations from two annotators (without considering alternative annotations suggested by the participating teams). The test dataset consists of 1,312 error-annotated sentences with 30,144 tokens on the source side. We make use of the official scorer for the shared task, M INLINEFORM0 Scorer v3.2 BIBREF19 , for evaluation. We perform statistical significance test using one-tailed sign test with bootstrap resampling on 100 samples.
On top of our baseline system described above, we incorporate the two neural network components, neural network global lexicon model (NNGLM) and neural network joint model (NNJM) as features. Both NNGLM and NNJM are trained using the parallel data used to train the translation model of our baseline system.
Recently, continuous space representations of words and phrases have been incorporated into SMT systems via neural networks. Specifically, addition of monolingual neural network language models BIBREF13 , BIBREF14 , neural network joint models (NNJM) BIBREF4 , and neural network global lexicon models (NNGLM) BIBREF3 have been shown to be useful for SMT
Grammatical error correction (GEC) is a challenging task due to the variability of the type of errors and the syntactic and semantic dependencies of the errors on the surrounding context. Most of the grammatical error correction systems use classification and rule-based approaches for correcting specific error types. However, these systems use several linguistic cues as features. The standard linguistic analysis tools like part-of-speech (POS) taggers and parsers are often trained on well-formed text and perform poorly on ungrammatical text. This introduces further errors and limits the performance of rule-based and classification approaches to GEC. As a consequence, the phrase-based statistical machine translation (SMT) approach to GEC has gained popularity because of its ability to learn text transformations from erroneous text to correct text from error-corrected parallel corpora without any additional linguistic information.
We conduct experiments by incorporating NNGLM and NNJM both independently and jointly into our baseline system
On top of our baseline system described above, we incorporate the two neural network components, neural network global lexicon model (NNGLM) and neural network joint model (NNJM) as features. Both NNGLM and NNJM are trained using the parallel data used to train the translation model of our baseline system.
|
How do they combine the two proposed neural network models?
|
The answers are shown as follows:
* ncorporating NNGLM and NNJM both independently and jointly into
* baseline system
|
1809.04960
| false
| null |
TF-IDF (Lexical, Non-Neural) is an important unsupervised baseline. We use the concatenation of the title and the body as the query to retrieve the candidate comment set by means of the similarity of the tf-idf value. The model is trained on unpaired articles and comments, which is the same as our proposed model.
NVDM (Lexical, Neural) is a VAE-based approach for document modeling BIBREF10 . We compare our model with this baseline to demonstrate the effect of modeling topic.
Table TABREF31 shows the performance of our models and the baselines in retrieval evaluation. We first compare our proposed model with other popular unsupervised methods, including TF-IDF, LDA, and NVDM. TF-IDF retrieves the comments by similarity of words rather than the semantic meaning, so it achieves low scores on all the retrieval metrics. The neural variational document model is based on the neural VAE framework. It can capture the semantic information, so it has better performance than the TF-IDF model. LDA models the topic information, and captures the deeper relationship between the article and comments, so it achieves improvement in all relevance metrics. Finally, our proposed model outperforms all these unsupervised methods, mainly because the proposed model learns both the semantics and the topic information.
Table TABREF32 shows the performance for our models and the baselines in generative evaluation. Similar to the retrieval evaluation, our proposed model outperforms the other unsupervised methods, which are TF-IDF, NVDM, and LDA, in generative evaluation. Still, the supervised IR achieves better scores than the seq2seq model. With the help of our proposed model, both IR and S2S achieve an improvement under the semi-supervised scenarios.
FLOAT SELECTED: Table 2: The performance of the unsupervised models and supervised models under the retrieval evaluation settings. (Recall@k, MRR: higher is better; MR: lower is better.)
FLOAT SELECTED: Table 3: The performance of the unsupervised models and supervised models under the generative evaluation settings. (METEOR, ROUGE, CIDEr, BLEU: higher is better.)
TF-IDF (Lexical, Non-Neural) is an important unsupervised baseline.
NVDM (Lexical, Neural) is a VAE-based approach for document modeling BIBREF10 . We compare our model with this baseline to demonstrate the effect of modeling topic.
Table TABREF31 shows the performance of our models and the baselines in retrieval evaluation.
Table TABREF32 shows the performance for our models and the baselines in generative evaluation.
FLOAT SELECTED: Table 2: The performance of the unsupervised models and supervised models under the retrieval evaluation settings. (Recall@k, MRR: higher is better; MR: lower is better.)
FLOAT SELECTED: Table 3: The performance of the unsupervised models and supervised models under the generative evaluation settings. (METEOR, ROUGE, CIDEr, BLEU: higher is better.)
|
By how much does their system outperform the lexicon-based models?
|
Proposed model is better than both lexical based models by significan margin in all metrics: BLEU 0.261 vs 0.250, ROUGLE 0.162 vs 0.155 etc.
|
null | false
| null |
How many continents are there in total?
|
There are 7 continents in total.
|
|
null | false
| null |
Elvis Aaron Presley (January 8, 1935 – August 16, 1977), often referred to mononymously as Elvis, was an American singer, actor and sergeant in the United States Army. Dubbed the "King of Rock and Roll", he is regarded as one of the most significant cultural figures of the 20th century. His energized interpretations of songs and sexually provocative performance style, combined with a singularly potent mix of influences across color lines during a transformative era in race relations, led him to both great success and initial controversy.
|
What rank did Elvis Presley hold in the US Army?
|
Elvis Presley held the rank of sergeant in the United States Army.
|
null | false
| 172
|
We describe a dataset developed for Named Entity Recognition in German federal court decisions. It consists of approx. 67,000 sentences with over 2 million tokens. The resource contains 54,000 manually annotated entities, mapped to 19 fine-grained semantic classes: person, judge, lawyer, country, city, street, landscape, organization, company, institution, court, brand, law, ordinance, European legal norm, regulation, contract, court decision, and legal literature. The legal documents were, furthermore, automatically annotated with more than 35,000 TimeML-based time expressions. The dataset, which is available under a CC-BY 4.0 license in the CoNNL-2002 format, was developed for training an NER service for German legal documents in the EU project Lynx.
The resource contains 54,000 manually annotated entities, mapped to 19 fine-grained semantic classes: person , judge , lawyer , country , city , street , landscape , organization , company , institution , court , brand , law , ordinance , European legal norm , regulation , contract , court decision, and legal literature.
|
What are the 19 fine-grained semantic classes?
|
Person , judge , lawyer , country , city , street , landscape , organization , company , institution , court , brand , law , ordinance , European legal norm , regulation , contract , court decision, and legal literature.
|
null | false
| 407
|
Zhang and proposed the post-nonlinear (PNL) causal model taking account of real-world data generating processes. In the PNL model, variables are generated as follows:
where X pa i is a set of direct cause variables, or parents, of x i and e i is an unobserved noise variable.
The inner function f i : R |Xpa i | → R represents nonlinear mixture of the causes and the outer one g i : R → R represents nonlinear distortion of sensors that often observed in real-world situations. The noises are independent of each other, e i ⊥ e j (∀i ̸ = j), and thus, a noise is independent of each of the corresponding causes, e i ⊥ x j ∈ X pa i . The distortion g i is assumed to be invertible. The PNL model is identifiable except for some special combinations of the functions and the noise distributions. The PNL model is known as the most general identifiable causal model and is equivalent to LiNGAM when f is linear and g is the identity and ANM when g is the identity.
The PNL model is identifiable except for some special combinations of the functions and the noise
distributions (Zhang and Hyvarinen ¨ , 2009; Peters et al., 2011, 2014). The PNL model is known as
the most general identifiable causal model and is equivalent to LiNGAM when f is linear and g is
the identity and ANM when g is the identity
|
How do you prove model identifiability?
|
The identifiability of the multivariate PNL model is already shown in the existing studies as stated in Section 2.2. We focused on the development of its estimation method in this paper.
|
null | false
| null |
classify the following items as income or spending: salary, movie, tax return, car rent, food
|
salary is income, car rent is income, movie is spending, car rent is spending, food is spending
|
|
null | false
| null |
What is linear programming and how to solve?
|
Linear programming is a method to achieve the best objective in an optimization problem whose objective function and requirements are represented by linear relationships. The simplex method and the interior-point method are common algorithms to solve linear programming problems.
You can visit https://en.wikipedia.org/wiki/Linear_programming to learn more.
|
|
null | false
| null |
What year did the hindenburg disaster occur?
|
The Hindenburg disaster occurred on May 6, 1937, when the Hindenburg airship caught fire and was destroyed during its attempt to dock with its mooring mast at Naval Air Station Lakehurst.
|
|
null | false
| null |
What links Brazil, Uruguay, Mozambique and Angola
|
Colonies of Portugal
|
|
1911.06191
| false
| null |
We participated in the WMT19 shared news translation task in 11 translation directions. We achieved first place for 8 directions: German$\leftrightarrow $English, German$\leftrightarrow $French, Chinese$\leftrightarrow $English, English$\rightarrow $Lithuanian, English$\rightarrow $Finnish, and Russian$\rightarrow $English, and three other directions were placed second (ranked by teams), which included Lithuanian$\rightarrow $English, Finnish$\rightarrow $English, and English$\rightarrow $Kazakh.
We achieved first place for 8 directions: German$\leftrightarrow $English, German$\leftrightarrow $French, Chinese$\leftrightarrow $English, English$\rightarrow $Lithuanian, English$\rightarrow $Finnish, and Russian$\rightarrow $English, and three other directions were placed second (ranked by teams), which included Lithuanian$\rightarrow $English, Finnish$\rightarrow $English, and English$\rightarrow $Kazakh.
|
Which language directions are machine translation systems of WMT evaluated on?
|
The answers are shown as follows:
* German$\leftrightarrow $English, German$\leftrightarrow $French, Chinese$\leftrightarrow $English, English$\rightarrow $Lithuanian, English$\rightarrow $Finnish, and Russian$\rightarrow $English
* Lithuanian$\rightarrow $English, Finnish$\rightarrow $English, and English$\rightarrow $Kazakh
|
null | false
| null |
What are the different ways in which a batsman in cricket can be dismissed
|
There are 11 ways of getting out or dismissed in cricket: caught, leg before wicket (LBW), stumped, hit wicket, run out, bowled, handled the ball, obstructing the field, stumped, hitting the ball twice, and retired.
|
|
null | false
| null |
What would you consider when preparing a Dj set?
|
There are many aspects to consider. Initially having an understanding of the audience and setting, can help in thinking about energy levels (BPMs), and duration.
It is also helpful to plan the direction of the DJ set, ramping from slow BPMs to high ones, perhaps peaking mid-set or just a different series of peaks along the way.
Track selection is key. Having a pre-defined list of potential tracks to play is a great idea.
Another consideration is mixing in key or following a harmonic mixing style for a neat flow.
All of the above require preparation and programming for a well-crafted DJ mix.
|
|
null | false
| null |
In 2003, Yang Liwei was launched aboard Shenzhou 5, becoming the first person sent into space by the Chinese space program. This achievement made China the third country to independently send humans into space.
|
Who is the first astronaut to get into space in China?
|
Yang Liwei, who reached space in 2003 onboard the Shenzhou 5 space ship, is the first astronaut in China to reach space.
|
null | false
| 157
|
The task of interpreting and following natural language (NL) navigation instructions involves interleaving different signals, at the very least the linguistic utterance and the representation of the world. For example, in turn right on the first intersection, the instruction needs to be interpreted, and a specific object in the world (the intersection) needs to be located in order to execute the instruction. In NL navigation studies, the representation of the world may be provided via visual sensors BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4 or as a symbolic world representation. This work focuses on navigation based on a symbolic world representation (referred to as a map).
Previous datasets for NL navigation based on a symbolic world representation, HCRC BIBREF5, BIBREF6, BIBREF7 and SAIL BIBREF8, BIBREF9, BIBREF10, BIBREF11, BIBREF12, BIBREF13, BIBREF14, BIBREF15 present relatively simple worlds, with a small fixed set of entities known to the navigator in advance. Such representations bypass the great complexity of real urban navigation, which consists of long paths and an abundance of previously unseen entities of different types.
In this work we introduce Realistic Urban Navigation (RUN), where we aim to interpret navigation instructions relative to a rich symbolic representation of the world, given by a real dense urban map. To address RUN, we designed and collected a new dataset based on OpenStreetMap, in which we align NL instructions to their corresponding routes. Using Amazon Mechanical Turk, we collected 2515 instructions over 3 regions of Manhattan, all specified (and verified) by (respective) sets of humans workers. This task raises several challenges. First of all, we assume a large world, providing long routes, vulnerable to error propagation; secondly, we assume a rich environment, with entities of various different types, most of which are unseen during training and are not known in advance; finally, we evaluate on the full route intended, rather than on last-position only.
We then propose a strong neural baseline for RUN where we augment a standard encoder-decoder architecture with an entity abstraction layer, attention over words and worlds, and a constantly updating world-state. Our experimental results and ablation study show that this architecture is indeed better-equipped to treat grounding in realistic urban settings than standard sequence-to-sequence architectures. Given this RUN benchmark, empirical results, and evaluation procedure, we hope to encourage further investigation into the topic of interpreting NL instructions in realistic and previously unseen urban domains.
[1]The task defined by BIBREF6 is of moving between entities only.
In this work we introduce Realistic Urban Navigation (RUN), where we aim to interpret navigation instructions relative to a rich symbolic representation of the world, given by a real dense urban map.
|
What do the authors introduce in the paper?
|
Realistic Urban Navigation (RUN).
|
null | false
| null |
Which of the following are beach wear vs what one would wear on a cold winter trek? If there are common items, create a third category called 'Common':
Socks
Hats
Neck gaiter
Sunglasses
Swimsuits
Undershirts
Cover-ups
Down jacket
Trek poles
Water Shoes
Beanie
Sandals
|
Beach wear:
Swimsuits
Cover-ups
Sandals
Water Shoes
Cold winter trek wear:
Neck gaiter
Down jacket
Beanie
Trek poles
Common:
Socks
Hats
Sunglasses
Undershirts
|
|
null | false
| null |
Songkran is a term derived from Sanskrit संक्रान्ति saṅkrānti meaning 'to move' or 'movement'. It derives from the movement of the sun from one position to another in the zodiac. According to its literal meaning in Sanskrit, a Songkran occurs every month. However, the period that Thai people refer to as Songkran happens when the sun moves from Pisces to Aries in the zodiac. The correct name for this period should actually be Maha Songkran ('great Songkran) because it coincides with the arrival of a New Year. The Songkran festival is, therefore, a celebration of the New Year in accordance with the solar calendar. The celebration covers a period of three days: 13 April is regarded as Maha Songkran, the day that the sun moves into Aries on the zodiac or the last day of the old year. The next day, 14 April is called Wan Nao, the transitional day between the old and the new years, and 15 April is called Wan Thaloeng Sok (Thai: วันเถลิงศก 'to begin a new era or year'), New Year's day itself.
|
What does Thai Songkran festival represents and how many days does it last?
|
Thai Songkran festival is a New Year celebration in Thailand according to solar calendar. Songkran happens when the sun moves from Pisces to Aries in the zodiac. The festival covers three days period from 13 April to 15 April.
|
null | false
| 402
|
The YummlyBIBREF8 dataset is used to understand how ingredients can be used to determine the cuisine. The dataset consists of 39,774 recipes. Each recipe is associated with a particular cuisine and a particular set of ingredients. Initial analysis of the data-set revealed a total of 20 different cuisines and 6714 different ingredients. Italian cuisine, with 7383 recipes, overshadows the dataset.
The numbers of recipes for the 19 cuisines is quite imbalanced.BIBREF9 The following graph shows the count of recipes per cuisine.
User specific data is collected from social media platforms such as Facebook and Instagram with the users permission. These images are then undergo a series of pre processing tasks. This helps in cleaning the data.
The Yummly dataset is used to understand how ingredients can be used to determine the cuisine. The dataset consists of 39,774 recipes. Each recipe is associated with a particular cuisine and a particular set of ingredients. Initial analysis of the data-set revealed a total of 20 different cuisines and 6714 different ingredients.
|
What is their dataset?
|
The Yummly dataset.
|
null | false
| 462
|
At the start of this section we set out to address three questions, which can be answered as: (1) The dependence of cross-domain generalisation on complexity can be directly and precisely when using linear models.
(2) The erratic performance of state of the art methods in DomainBed can be largely explained in terms of implied model complexity.
(3) If the goal is to optimise for domain generalisation, then then domain-wise validation is preferred to instance-wise validation as a modelselection objective.
Our analysis suggests that several existing methods that tried to tune learnable DG hyper-parameters by performance on a held-out domain were broadly on the right track, and concurs with Gulrajani & Lopez-Paz (2021) that those methods with underspecified hyperparameter and model selection procedures are unhelpful. However, given that most neural methods have many more complexity hyperparameters than the single hyperparameter that we were able to carefully control for linear models, obtaining accurate tuning and reliable performance evaluation is likely to be a challenge. Gradient-based hyper-parameter estimation methods, as initially attempted in; , together with efficient methods for long-inner loop hypergradient calculation, may benefit the former problem by making search in larger numbers of hyperparameters more feasible. Alternatively, using general purpose pre-trained features as we did here, and focusing on learning shallow models that can be accurately tuned for DG may be another promising avenue in practice. Although achieving state of the art performance is not our focus, we note that our results in Table are quite competitive with end-to-end trained state of the art, despite using fixed features and shallow models.
Our analysis suggests that several existing methods that tried to tune learnable DG hyper-parameters by performance on a held-out domain (Balaji et al., 2018; Li et al., 2019) were broadly on the right track, and concurs with Gulrajani & Lopez-Paz (2021) that those methods with underspecified hyperparameter and model selection procedures are unhelpful
|
What result of [Gulrajani & Lopez-Paz, 2020] do we dispute?
|
[Gulrajani & Lopez-Paz, 2020] observed that instance-wise was a better validation criterion than domain-wise. We have shown the opposite is the case when using better hyperparameter tuning.
|
1602.01208
| false
| null |
Accuracy of acquired phoneme sequences representing the names of places
We evaluated whether the names of places were properly learned for the considered teaching places. This experiment assumes a request for the best phoneme sequence INLINEFORM0 representing the self-position INLINEFORM1 for a robot. The robot moves close to each teaching place. The probability of a word INLINEFORM2 when the self-position INLINEFORM3 of the robot is given, INLINEFORM4 , can be obtained by using equation ( EQREF37 ). The word having the best probability was selected. We compared the PAR with the correct phoneme sequence and a selected name of the place. Because “kiqchiN” and “daidokoro” were taught for the same place, the word whose PAR was the higher score was adopted.
Fig. FIGREF63 shows the results of PAR for the word considered the name of a place. SpCoA (latticelm), the proposed method using the results of unsupervised word segmentation on the basis of the speech recognition results in the lattice format, showed the best PAR score. In the 1-best and BoS methods, a part syllable sequence of the name of a place was more minutely segmented as shown in Table TABREF55 . Therefore, the robot could not learn the name of the teaching place as a coherent phoneme sequence. In contrast, the robot could learn the names of teaching places more accurately by using the proposed method.
Accuracy of acquired phoneme sequences representing the names of places
We evaluated whether the names of places were properly learned for the considered teaching places. This experiment assumes a request for the best phoneme sequence INLINEFORM0 representing the self-position INLINEFORM1 for a robot. The robot moves close to each teaching place. The probability of a word INLINEFORM2 when the self-position INLINEFORM3 of the robot is given, INLINEFORM4 , can be obtained by using equation ( EQREF37 ). The word having the best probability was selected. We compared the PAR with the correct phoneme sequence and a selected name of the place. Because “kiqchiN” and “daidokoro” were taught for the same place, the word whose PAR was the higher score was adopted.
Fig. FIGREF63 shows the results of PAR for the word considered the name of a place. SpCoA (latticelm), the proposed method using the results of unsupervised word segmentation on the basis of the speech recognition results in the lattice format, showed the best PAR score. In the 1-best and BoS methods, a part syllable sequence of the name of a place was more minutely segmented as shown in Table TABREF55 . Therefore, the robot could not learn the name of the teaching place as a coherent phoneme sequence. In contrast, the robot could learn the names of teaching places more accurately by using the proposed method.
|
How do they evaluate how their model acquired words?
|
The answers are shown as follows:
* PAR score
|
null | false
| null |
How do you prepare lentil soup?
|
There are many ways to prepare lentil soup. However a common method involves cooking onions, celery, and carrots in olive oil for several minutes. Then you would add aromatics, such as garlic and spices, and cook for a minute. Then you would add lentils and broth (or water) and bring to a simmer. The soup should simmer for at least 30 minutes until the lentils are fully cooked. Once finished cooking, you would garnish the soup with croutons, herbs, or drizzle with oil.
|
|
1706.01723
| true
| null |
We evaluate our method on three tagging tasks: POS tagging (Pos), morphological tagging (Morph) and supertagging (Stag).
We select these tasks as examples for tagging applications because they differ strongly in tag set sizes. Generally, the Pos set sizes for all the languages are no more than 17 and Stag set sizes are around 200. When treating morphological features as a string (i.e. not splitting into key-value pairs), the sizes of the Morph tag sets range from about 100 up to 2000.
The test results for the three tasks are shown in Table TABREF17 in three groups. The first group of seven columns are the results for Pos, where both LSTM and CNN have three variations of input features: word only ( INLINEFORM0 ), character only ( INLINEFORM1 ) and both ( INLINEFORM2 ). For Morph and Stag, we only use the INLINEFORM3 setting for both LSTM and CNN.
We evaluate our method on three tagging tasks: POS tagging (Pos), morphological tagging (Morph) and supertagging (Stag).
We select these tasks as examples for tagging applications because they differ strongly in tag set sizes.
The test results for the three tasks are shown in Table TABREF17 in three groups. The first group of seven columns are the results for Pos, where both LSTM and CNN have three variations of input features: word only ( INLINEFORM0 ), character only ( INLINEFORM1 ) and both ( INLINEFORM2 ). For Morph and Stag, we only use the INLINEFORM3 setting for both LSTM and CNN.
|
Do they jointly tackle multiple tagging problems?
|
No.
|
null | false
| null |
Carol is used as a female name, Carl or Karl is the male version of the name, although in contemporary Western usage it is more commonly used for females. It is a European continental spelling of the English Charles, the Germanic Carl, and the Latin Carolus, and can also be a short version of the name Caroline. Spelling variations include Carole.
|
Given the reference text, what is the female equivalent of Karl?
|
Carol (sometimes Carole and other variations) is used a female name for Carl or Karl.
|
1701.08118
| false
| null |
In order to assess the reliability of the hate speech definitions on social media more comprehensively, we developed two online surveys in a between-subjects design. They were completed by 56 participants in total (see Table TABREF7 ). The main goal was to examine the extent to which non-experts agree upon their understanding of hate speech given a diversity of social media content. We used the Twitter definition of hateful conduct in the first survey. This definition was presented at the beginning, and again above every tweet. The second survey did not contain any definition. Participants were randomly assigned one of the two surveys.
We used the Twitter definition of hateful conduct in the first survey. This definition was presented at the beginning, and again above every tweet.
|
What definition was one of the groups was shown?
|
The answers are shown as follows:
* Twitter definition of hateful conduct
|
null | false
| null |
Is Slovenia a nice place to take a food fueled vacation?
|
Yeah, for sure, Slovenia regularly ranks super-highly in terms of safety, sustainability, and friendliness. The whole country was also awarded the “European Region of Gastronomy” title for 2021, and a few years before that, chef Ana Roš was named World’s Best Female Chef
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.