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 |
Which characters belong to DC or Marvel Universe? Electro, Black Lightning
|
Electro is Marvel, Black Lightning is DC
|
|
null | false
| null |
Can you write me a haiku to boost my spirits today?
|
Sunlight streaming through,
Lightning in the sky,
My spirits lift.
|
|
null | false
| null |
How many planets orbit the sun?
|
For many years astronomers stated that there were 9 planets that orbited the Sun. In 2006 the 9th planet and furthest from the Sun, Pluto, was demoted from a planet to a dwarf planet. Technically there are now 8 planets and 1 dwarf planet that orbit the Sun.
|
|
null | false
| null |
Who was the most prolific religious leader ever?
|
Some know him as Zoroaster and others as Zarathustra. He lived between the 18th and 16th century BCE. The reputed founder of Zoroastrianism, he started the major religion of the Persian Empire. This empire flourished from 550 BCE to 330 BCE.
Nobody knows his birthplace. But many Arabic sources claim he got born in modern Azerbaijan. To this day, so many countries have tried to lay claim to his birthplace, that we might think he was not a single person.
Followers worshiped Ahura Mazda, the Supreme Being or God of Zoroastrianism. While the religion has gone extinct, we have to note it influenced so many ancient civilizations. Zoroastrianism influenced Classic Greek philosophy, Roman teachings, and many more.
There are still some people practicing Zoroastrianism, namely in India.
|
|
null | false
| null |
List some reasons why cats are the best pets.
|
- They are cuddly
- They are independent
- They very frequently bond with their humans
- They provide emotional support
- They are less demanding than dogs
- They are easy to care for
- They don't need constant attention 24/7
- They can be trained to use a litter box or even a real toilet
- They don't take up a lot of space
- They don't require lots of expensive toys, equipment, or beds
- They don't need to go on walks every day
- They don't make large messes
- They are often much more easier to care for than dogs
- They don't slobber
- They don't lick your face
- They don't chew up your furniture
- They are extremely cute
|
|
null | false
| null |
What do economists do?
|
Economists study a diverse range of social issues whose outcomes are driven by the choices of individuals under resource constraints and limited information. There are various subfields in economics:
- Microeconomists study the supply and demand decision of individuals and firms, such as how profits can be maximized and how much of a good or service consumers will demand at a certain price.
- Industrial/Organizational Economists study the market structure of particular industries in terms of the number of competitors, and the market decisions of competitive firms and monopolies. These economists may also be concerned with antitrust policy and its impact on market structure.
- Macroeconomists study historical trends in the whole economy and forecast future trends in areas such as unemployment, inflation, economic growth, productivity, and investment.
Financial Economists study the money and banking system and the effects of rising interest rates.
- Public Finance Economists primarily are involved in studying the role of the government in the economy and the effects of tax cuts, budget deficits, and welfare policies.
International Economists study international financial markets, exchange rates, and the effects of various trade policies such as tariffs.
- Labor Economists study the supply and demand for labor and the determination of wages. These economists also try to explain the reasons for unemployment, and the effects on labor markets of changing demographic trends such as an aging population and increasing immigration.
- Econometricians are involved in all areas of economics and use mathematical techniques such as calculus, game theory, and regression analysis to formulate economic models. These models help to explain economic relationships and are used to develop forecasts related to the nature and length of business cycles, the effects of a specific rate of inflation on the economy, the effects of tax legislation on unemployment levels, and other economic phenomena. Many economists have applied these fundamental areas of economics to more narrow areas with specific applications such as health, education, agriculture, urban and regional economics, law, history, energy, and the environment.
Economists may work in Business, Finance Government and Academia.
|
|
null | false
| null |
Think up a bunch of different ways to spend time outdoors
|
Hiking, camping, fishing, kayaking, walking, biking, jogging
|
|
null | false
| null |
The Berlin Marathon was established in 1974 by Horst Milde, a baker and running enthusiast in the city. The race had 244 finishers; 234 men and 10 women, and was won by Günter Hallas and Jutta von Haase in times of 2:44:53 and 3:22:01 respectively. In 1977, Christa Vahlensieck established a new women's world record at the Berlin Marathon, running 2:34:47. Four years later, the race started outside the Reichstag for the first time, by which time it had grown to 2,583 finishers, making it the largest city road race in Germany. The following year, before German unification, some East Germans secretly ran the race, registering under false names to avoid recriminations from the East German secret service. In 1989, a children's race was added to the event, and eight years later an inline skating marathon was contested.
In 1998, Ronaldo da Costa set a new men's world record, the only South American to hold a marathon world record. Two years later, one of the race's pacemakers, Simon Biwott won the race, after one of the favourites had pulled out in the second half of the race. In 2001, Naoko Takahashi became the first woman to run a sub-2:20 marathon, completing the Berlin Marathon in 2:19:46. The race had continued to be the venue of world records; between 2003 and 2018, seven men's records were set at the race; the most recent remains as the current world record time; Eliud Kipchoge's 2:01:39.
The event was held in mid-September in 2000 and 2018, because of a conflict with the Women's Olympic Marathon, and due to German Unity Day preparations, respectively.
The 2020 edition of the event was prohibited from being held on its originally scheduled dates due to the coronavirus pandemic, and was eventually cancelled once it was clear that it would not be possible to hold it at a later date in 2020. All registrants were given the option of either transferring their entry to 2021 or obtaining a refund.
|
Extract the names of the winners of the Berlin Marathon from the reference text and provide them as a bulleted list.
|
* Günter Hallas
* Jutta van Hasse
* Christa Vahlensieck
* Ronaldo da Costa
* Simon Biwott
* Naoko Takahashi
* Eliud Kipchoge
|
null | false
| 60
|
Automatic Speech Recognition (ASR) systems have been evolving quickly and reaching human parity in certain cases. The systems usually perform pretty well on reading style and clean speech, however, most of the available systems suffer from situation where the speaking style is conversation and in noisy environments. It is not straight-forward to tackle such problems due to difficulties in data collection for both speech and text. In this paper, we attempt to mitigate the problems using language models combination techniques that allows us to utilize both large amount of writing style text and small number of conversation text data. Evaluation on the VLSP 2019 ASR challenges showed that our system achieved 4.85% WER on the VLSP 2018 and 15.09% WER on the VLSP 2019 data sets.
Evaluation on the VLSP 2019 ASR challenges showed that our system achieved 4.85% WER on the VLSP 2018 and 15.09% WER on the VLSP 2019 data sets.
|
What WER did their system achieve on the VLSP 2019 data sets?
|
15.09% WER.
|
null | false
| null |
The American League of Professional Baseball Clubs, known simply as the American League (AL), is one of two leagues that make up Major League Baseball (MLB) in the United States and Canada.
|
How many teams are in the MLB American League
|
The American League consists of 15 teams (including one Canadian team) aligned in three divisions: the AL East, comprising the Baltimore Orioles, Boston Red Sox, New York Yankees, Tampa Bay Rays, and Toronto Blue Jays; the AL Central, comprising the Chicago White Sox, Cleveland Indians, Detroit Tigers, Kansas City Royals, and Minnesota Twins; and the AL West, comprising the Houston Astros, Los Angeles Angels of Anaheim, Oakland Athletics, Seattle Mariners, and Texas Rangers.
|
null | false
| null |
The proposed league would have consisted of two conferences split into four divisions. The current KHL would have made up one of the conferences. The other conference would have been composed of a Scandinavian division and a Central European division. The winners of the two conferences would have played each other for the overall championship.
|
Based on this paragraph about the potential Kontinental Hockey League expansion, how many divisions and how many conferences would the proposed league have?
|
The league would have two conferences and four divisions.
|
null | false
| null |
Give me a title of one of the "Harry Potter" novels.
|
Harry Potter and the Goblet of Fire
|
|
null | false
| 145
|
All BERT experiments were trained for 500K steps. The code was written in Python and we used the Tensorflow libraries to create the models. The experiments were trained on a single NVIDIA Tesla V100 32 GB graphic card. The data was first processed into Tensorflow records as the input to the model. The set of hyperparameters which we found optimal after experimenting with different sets are in (Table TABREF10).
This set of parameters were chosen as there training performances were better than smaller models on modelling the long sequences of sub-words. We use the Adam optimizer BIBREF16 same as the English BERT. A maximum sequence length of 300 encompasses 98 percent of the training data and also allows us to fit larger models on the GPU card. Hyper-parameter optimization is very difficult in case of these models as they take around 15 days to train given the resources. The hyperparameter choices were therefore more dependant on the original BERT with little tweaks. We assess the training performance of the the model in the (Table TABREF11).
When we train the BERT model we mask some percentage of the input tokens at random, and then predict those masked tokens, this is known as Masked LM. The masked LM loss, refers specifically to the loss when the masked language model predicts on the masked tokens. The masked LM accuracy refers specifically to the accuracy with which the model predicts on the masked tokens. The loss for both the models are far off from the Masked LM loss of the English BERT, key difference being the pre-training data for both the language models are quite different. Google training their model on 3.3 Billion words from BooksCorpus BIBREF17 and the English Wikipedia and our model being trained on 144 million words. Comparing the two Finnish models, the left-marked model has a better training performance than left+right-marked model.
The results of the pseudo-perplexity described in the previous section to evaluate the above models on the test data-set is in table (Table TABREF12).The test dataset is of a different context when compared to the training data, and interestingly both the models are quite confident when it comes to the test dataset. The pseudo-perplexity values of left-marked are lower when compared to left-right-marked signifying that it is more confident.
We cannot directly compare the perplexity scores BERT model with a unidirectional LSTM model as both are calculated in a different manner. We can experiment to compare it with a Bi-directional LSTM or use a downstreaming task to compare both the performances. We could also randomly mask tokens and then compare the prediction accuracy on the masked tokens.
All Transformer-XL experiments are also trained equally for 500K steps. The code was written in Python and we used the PyTorch libraries for model creation. The experiments were trained on a single NVIDIA Tesla V100 32 GB graphic card. Two sets of hyperparameters were chosen to be compared after some initial optimization and are in (Table TABREF14)
From the above parameter choice, we wanted to experiment, whether providing more Segment and Memory length is advantageous (longer context) than a larger model. These parameters where chosen after some hyperparameter optimization. Same as for BERT we use the Adam optimizer, but we also use a cosine annealing learning rate scheduler to speed-up training BIBREF18. The training performance results are in (Table TABREF15)
As opposed to BERT, the left+right-marked models have a better training performance than their counterpart. Interestingly the larger model trains much better when compared to providing larger contexts. The same set of parameters for the 32-32 model cannot be replicated for 150-150 model as the latter takes a lot of space on the GPU card. The test set is same as that used with BERT and the results are in (Table TABREF16). The test performance is similar to that of the training performance with left-right-marked large model(32-32) performing the best. We can directly compare the perplexity scores with the previous best BIBREF19 as both are unidirectional models, Transformer-XL model has outperformed the latter by 27%.
Transformer-XL and BERT both have low perplexity and pseudo-perplexity scores, but both cannot be directly compared as they are calculated quite differently (Eq.DISPLAY_FORM4, Eq.DISPLAY_FORM6). The dramatically low scores of BERT indicate that per word predicted probability is higher than that of a uni-directional model. Thus the predicted word probability distribution is much sharper when compared to the XL model probability distribution. At this point, we cannot say which model architecture has performed better- BERT or Transformer-XL, despite both of them achieving good low perplexity scores. We would need to experiment with a downstreaming task in-order to fairly compare model performances.
At this point, we cannot say which model architecture has performed better- BERT or Transformer-XL, despite both of them achieving good low perplexity scores. We would need to experiment with a downstreaming task in order to fairly compare model performances.
|
Does BERT outperfom Transformer-XL?
|
No, cannot say which model architecture has performed better.
|
null | false
| 4
|
Knowledge Base Question Answering (KBQA) systems answer questions by obtaining information from KB tuples BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 . For an input question, these systems typically generate a KB query, which can be executed to retrieve the answers from a KB. Figure 1 illustrates the process used to parse two sample questions in a KBQA system: (a) a single-relation question, which can be answered with a single $<$ head-entity, relation, tail-entity $>$ KB tuple BIBREF6 , BIBREF7 , BIBREF2 ; and (b) a more complex case, where some constraints need to be handled for multiple entities in the question. The KBQA system in the figure performs two key tasks: (1) entity linking, which links $n$ -grams in questions to KB entities, and (2) relation detection, which identifies the KB relation(s) a question refers to.
The main focus of this work is to improve the relation detection subtask and further explore how it can contribute to the KBQA system. Although general relation detection methods are well studied in the NLP community, such studies usually do not take the end task of KBQA into consideration. As a result, there is a significant gap between general relation detection studies and KB-specific relation detection. First, in most general relation detection tasks, the number of target relations is limited, normally smaller than 100. In contrast, in KBQA even a small KB, like Freebase2M BIBREF2 , contains more than 6,000 relation types. Second, relation detection for KBQA often becomes a zero-shot learning task, since some test instances may have unseen relations in the training data. For example, the SimpleQuestions BIBREF2 data set has 14% of the golden test relations not observed in golden training tuples. Third, as shown in Figure 1 (b), for some KBQA tasks like WebQuestions BIBREF0 , we need to predict a chain of relations instead of a single relation. This increases the number of target relation types and the sizes of candidate relation pools, further increasing the difficulty of KB relation detection. Owing to these reasons, KB relation detection is significantly more challenging compared to general relation detection tasks.
This paper improves KB relation detection to cope with the problems mentioned above. First, in order to deal with the unseen relations, we propose to break the relation names into word sequences for question-relation matching. Second, noticing that original relation names can sometimes help to match longer question contexts, we propose to build both relation-level and word-level relation representations. Third, we use deep bidirectional LSTMs (BiLSTMs) to learn different levels of question representations in order to match the different levels of relation information. Finally, we propose a residual learning method for sequence matching, which makes the model training easier and results in more abstract (deeper) question representations, thus improves hierarchical matching.
In order to assess how the proposed improved relation detection could benefit the KBQA end task, we also propose a simple KBQA implementation composed of two-step relation detection. Given an input question and a set of candidate entities retrieved by an entity linker based on the question, our proposed relation detection model plays a key role in the KBQA process: (1) Re-ranking the entity candidates according to whether they connect to high confident relations detected from the raw question text by the relation detection model. This step is important to deal with the ambiguities normally present in entity linking results. (2) Finding the core relation (chains) for each topic entity selection from a much smaller candidate entity set after re-ranking. The above steps are followed by an optional constraint detection step, when the question cannot be answered by single relations (e.g., multiple entities in the question). Finally the highest scored query from the above steps is used to query the KB for answers.
Our main contributions include: (i) An improved relation detection model by hierarchical matching between questions and relations with residual learning; (ii) We demonstrate that the improved relation detector enables our simple KBQA system to achieve state-of-the-art results on both single-relation and multi-relation KBQA tasks.
Finally, we propose a residual learning method for sequence matching, which makes the model training easier and results in more abstract (deeper) question representations, thus improves hierarchical matching.
|
Why is the residual learning method used for sequence matching?
|
To make the model training easier and results in more abstract (deeper) question representations, thus improves hierarchical matching.
|
null | false
| null |
Education in Singapore is managed by the Ministry of Education (MOE). It controls the development and administration of state schools receiving taxpayers' funding, but also has an advisory and supervisory role in respect of private schools. For both private and state schools, there are variations in the extent of autonomy in their curriculum, scope of taxpayers' aid and funding, tuition burden on the students, and admission policy.
|
How is education in Singapore
|
Education spending usually makes up about 20 per cent of the annual national budget, which subsidises state education and government-assisted private education for Singaporean citizens and funds the Edusave programme. Non-citizens bear significantly higher costs of educating their children in Singapore government and government-aided schools. In 2000, the Compulsory Education Act codified compulsory education for children of primary school age (excepting those with disabilities), and made it a criminal offence for parents to fail to enroll their children in school and ensure their regular attendance. Exemptions are allowed for homeschooling or full-time religious institutions, but parents must apply for exemption from the Ministry of Education and meet a minimum benchmark.
The main language of instruction in Singapore is English, which was officially designated the first language within the local education system in 1987. English is the first language learned by half the children by the time they reach preschool age and becomes the primary medium of instruction by the time they reach primary school. Although Malay, Mandarin and Tamil are also official languages, English is the language of instruction for nearly all subjects except the official Mother Tongue languages and the literatures of those languages; these are generally not taught in English, although there is provision for the use of English at the initial stages. Certain schools, such as secondary schools under the Special Assistance Plan (SAP), encourage a richer use of the mother tongue and may occasionally teach subjects in Mandarin Chinese.
Singapore's education system has been consistently ranked as one of the highest in the world by the OECD. It is believed that this comes from the style of teaching that is implemented in Singapore. Teachers focus on making sure that each of their students thoroughly move through the syllabus before moving on. By doing this teachers in Singapore teach a much more narrow but deeper type of instruction. Furthermore, it has been described as "world-leading" and in 2010 was among those picked out for commendation by the Conservative former UK Education Secretary Michael Gove. According to PISA, an influential worldwide study on educational systems, Singapore has the highest performance in international education and tops in global rankings. In 2020, Singaporean students made up half of the perfect scorers in the International Baccalaureate (IB) examinations worldwide.
|
null | false
| null |
How many Indian princely states were united by Sardar Patel during 1947-1949
|
565 princely states
|
|
null | false
| null |
Who is the Denver Broncos best player in franchise history?
|
John Elway is widely considered the best player for the Denver Broncos. In addition to leading the team to five Super Bowls, and winning three, he is also a holder of numerous franchise and NFL records. He became a fan favorite for his toughness. His most memorable moment was the "helicopter play" in Super Bowl XXXII, a key moment in the game. Other notable players include Steve Atwater, Shannon Sharpe, Randy Gradishar, Terrell Davis, Rod Smith, Peyton Manning, Jake Plummer, and Tom Jackson.
|
|
null | false
| null |
What is the former prince of Wales is now referred as?
|
King Charles III of United Kingdom
|
|
null | false
| null |
Identify which instrument is string or woodwind: Chrotta, Frula
|
Chrotta is string, Frula is woodwind.
|
|
null | false
| 168
|
Knowledge representation and reasoning (KRR) is the process of representing the domain knowledge in formal languages (e.g., SPARQL, Prolog) such that it can be used by expert systems to execute querying and reasoning services. KRR have been applied in many fields including financial regulations, medical diagnosis, laws, and so on. One major obstacle in KRR is the creation of large-scale knowledge bases with high quality. For one thing, this requires the knowledge engineers (KEs) not only to have the background knowledge in a certain domain but have enough skills in knowledge representation as well. Unfortunately, qualified KEs are also in short supply. Therefore, it would be useful to build a tool that allows the domain experts without any background in logic to construct and query the knowledge base simply from text.
Controlled natural languages (CNLs) BIBREF0 were developed as a technology that achieves this goal. CNLs are designed based on natural languages (NLs) but with restricted syntax and interpretation rules that determine the unique meaning of the sentence. Representative CNLs include Attempto Controlled English BIBREF1 and PENG BIBREF2 . Each CNL is developed with a language parser which translates the English sentences into an intermediate structure, discourse representation structure (DRS) BIBREF3 . Based on the DRS structure, the language parsers further translate the DRS into the corresponding logical representations, e.g., Answer Set Programming (ASP) BIBREF4 programs. One main issue with the aforementioned CNLs is that the systems do not provide enough background knowledge to preserve semantic equivalences of sentences that represent the same meaning but are expressed via different linguistic structures. For instance, the sentences Mary buys a car and Mary makes a purchase of a car are translated into different logical representations by the current CNL parsers. As a result, if the user ask a question who is a buyer of a car, these systems will fail to find the answer.
In this thesis proposal, I will present KALM BIBREF5 , BIBREF6 , a system for knowledge authoring and question answering. KALM is superior to the current CNL systems in that KALM has a complex frame-semantic parser which can standardize the semantics of the sentences that express the same meaning via different linguistic structures. The frame-semantic parser is built based on FrameNet BIBREF7 and BabelNet BIBREF8 where FrameNet is used to capture the meaning of the sentence and BabelNet BIBREF8 is used to disambiguate the meaning of the extracted entities from the sentence. Experiment results show that KALM achieves superior accuracy in knowledge authoring and question answering as compared to the state-of-the-art systems.
The rest parts are organized as follows: Section SECREF2 discusses the related works, Section SECREF3 presents the KALM architecture, Section SECREF4 presents KALM-QA, the question answering part of KALM, Section SECREF5 shows the evaluation results, Section SECREF6 shows the future work beyond the thesis, and Section SECREF7 concludes the paper.
One major obstacle in KRR is the creation of large-scale knowledge bases with high quality.
|
What is the major obstacle in KRR?
|
One major obstacle in KRR is the creation of large-scale knowledge bases with high quality.
|
null | false
| null |
What are four well known movies with Samuel L Jackson?
|
Four well known movies with Samuel L Jackson are Pulp Fiction, Goodfellas, The Incredibles, and Marvel's the Avengers
|
|
null | false
| null |
Why is Pickleball a popular sport?
|
Pickleball's popularity has been on the rise, thanks in large part to its low entry barrier. With elements of tennis, ping-pong, and badminton, players with experience in racket sports can easily pick up the game. The court is smaller than a tennis court, and the slower-moving ball allows for players of varying athletic abilities to participate with ease
|
|
null | false
| null |
Kenneth John Ortega (born April 18, 1950) is an American filmmaker, touring manager, and choreographer.
|
Who is Kenny Ortega?
|
Kenneth John Ortega (born April 18, 1950) is an American filmmaker, touring manager, and choreographer. He is known for directing the films Newsies, Hocus Pocus, The Cheetah Girls 2, High School Musical, Michael Jackson's This Is It, Descendants, and Julie and the Phantoms. He also choreographed the films St. Elmo's Fire, Pretty in Pink, Ferris Bueller's Day Off, and Dirty Dancing; and directed multiple concert tours for Cher, Gloria Estefan, Miley Cyrus, and Michael Jackson. For his work, Ortega was honored with both a star on the Hollywood Walk of Fame and a Disney Legend Award in 2019.
|
null | false
| null |
What are some good body weight exercises for shoulders?
|
Good body weight exercises for shoulders include plank front arm reaches, pike pushups, plank shoulder taps, arm circles, and dolphin presses.
|
|
null | false
| null |
How do you make a good cappuccino?
|
There are multiple steps involved in making a good cappuccino:
Step 1: Find fresh coffee beans and grind them finely
For optimal taste, the coffee beans should be fresh and only slightly oxidized. It is important to use a high quality grinder that grinds the beans uniformly and finely. You should fill a portafilter with 18-21g of grounds and tamp the grounds.
Step 2: Pull the shot
Turn on the espresso machine, wait for it to heat up, and press the button to start the shot. The goal is to make ~40g of espresso in 25-30 seconds starting from when you initiated the pull.
Step 3: Steam the milk and pour
Pour milk into a carafe and use the steam wand to steam the milk. You want to aerate the milk, but not too much or too little. You want to avoid making macrobubbles and also heat up the milk so that the carafe is hot to the touch. Now pour the milk into the espresso, and you're done!
|
|
1611.02550
| false
| null |
The data used for this task is drawn from the Switchboard conversational English corpus BIBREF31 . The word segments range from 50 to 200 frames in length. The acoustic features in each frame (the input to the word embedding models INLINEFORM0 ) are 39-dimensional MFCCs+ INLINEFORM1 + INLINEFORM2 . We use the same train, development, and test partitions as in prior work BIBREF13 , BIBREF11 , and the same acoustic features as in BIBREF13 , for as direct a comparison as possible. The train set contains approximately 10k example segments, while dev and test each contain approximately 11k segments (corresponding to about 60M pairs for computing the dev/test AP). As in BIBREF13 , when training the classification-based embeddings, we use a subset of the training set containing all word types with a minimum of 3 occurrences, reducing the training set size to approximately 9k segments.
The data used for this task is drawn from the Switchboard conversational English corpus BIBREF31 .
|
Which dataset do they use?
|
The answers are shown as follows:
* Switchboard conversational English corpus
|
null | false
| null |
Why does adding water to a hot pan of oil cause fire?
|
Oil and water do not mix. Water is denser than oil and will settle in the bottom. When you have hot oil in a pan and add a droplet of water the water will sink to the bottom and rapidly heat up to boiling point and vaporize into steam. The steam escapes through the oil leading to a steam explosion. The explosion takes the oil with it and the hot oil becomes fuel which ignites on the nearby stove flame causing a fire.
|
|
null | false
| 242
|
Automatic Text Summarization deals with the task of condensing documents into a summary, whose level is similar to a human-generated summary. It is mostly distributed into two distinct domains, i.e., Abstractive Summarization and Extractive Summarization. Abstractive summarization( Dejong et al. ,1978) involves models to deduce the crux of the document. It then presents a summary consisting of words and phrases that were not there in the actual document, sometimes even paraphrasing BIBREF1 . A state of art method proposed by Wenyuan Zeng BIBREF2 produces such summaries with length restricted to 75. There have been many recent developments that produce optimal results, but it is still in a developing phase. It highly relies on natural language processing techniques, which is still evolving to match human standards. These shortcomings make abstractive summarization highly domain selective. As a result, their application is skewed to the areas where NLP techniques have been superlative. Extractive Summarization, on the other hand, uses different methods to identify the most informative/dominant sentences through the text, and then present the results, ranking them accordingly. In this paper, we have proposed two novel stand-alone summarization methods.The first method is based on Glove Model BIBREF3 ,and other is based on Facebook's InferSent BIBREF4 . We have also discussed how we can effectively subdue shortcomings of one model by using it in coalition with models which capture the view that other faintly held.
In this paper, we have proposed two novel stand-alone summarization methods.The first method is based on Glove Model,and other is based on Facebook's InferSent.
|
What models are their methods based on?
|
Glove Model and InferSent model.
|
null | false
| 206
|
Keyphrases are single or multi-word lexical units that best summarise a document BIBREF0. As such, they are of great importance for indexing, categorising and browsing digital libraries BIBREF1. Yet, very few documents have keyphrases assigned, thus raising the need for automatic keyphrase generation systems. This task falls under the task of automatic keyphrase extraction which can also be the subtask of finding keyphrases that only appear in the input document. Generating keyphrases can be seen as a particular instantiation of text summarization, where the goal is not to produce a well-formed piece of text, but a coherent set of phrases that convey the most salient information. Those phrases may or may not appear in the document, the latter requiring some form of abstraction to be generated. State-of-the-art systems for this task rely on recurrent neural networks BIBREF2, BIBREF3, BIBREF4, and hence require large amounts of annotated training data to achieve good performance. As gold annotated data is expensive and difficult to obtain BIBREF5, previous works focused on readily available scientific abstracts and used author-assigned keyphrases as a proxy for expert annotations. However, this poses two major issues: 1) neural models for keyphrase generation do not generalize well across domains, thus limiting their use in practice; 2) author-assigned keyphrases exhibit strong consistency issues that negatively impacts the model's performance. There is therefore a great need for annotated data from different sources, that is both sufficiently large to support the training of neural-based models and that comprises gold-standard labels provided by experts. In this study, we address this need by providing KPTimes, a dataset made of 279 923 news articles that comes with editor-assigned keyphrases.
Online news are particularly relevant to keyphrase generation since they are a natural fit for faceted navigation BIBREF6 or topic detection and tracking BIBREF7. Also, and not less importantly, they are available in large quantities and are sometimes accompanied by metadata containing human-assigned keyphrases initially intended for search engines. Here, we divert these annotations from their primary purpose, and use them as gold-standard labels to automatically build our dataset. More precisely, we collect data by crawling selected news websites and use heuristics to draw texts paired with gold keyphrases. We then explore the resulting dataset to better understand how editors tag documents, and how these expert annotations differ from author-assigned keyphrases found in scholarly documents. Finally, we analyse the performance of state-of-the-art keyphrase generation models and investigate their transferability to the news domain and the impact of domain shift.
More precisely, we collect data by crawling selected news websites and use heuristics to draw texts paired with gold keyphrases.
|
How to collect the data?
|
By crawling selected news websites.
|
null | false
| null |
Classify each of the following as either a stringed instrument or a percussion instrument: violin, guitar, drums, congas, bongos, bass guitar.
|
The violin, guitar and bass guitar are stringed instruments. Drums, congas, and bongos are percussion instruments.
|
|
null | false
| 102
|
Following lebret2016neural, we used BLEU-4, NIST-4 and ROUGE-4 as the evaluation metrics. We first make a few observations based on the results on the English dataset (Table TABREF15 ). The basic seq2seq model, as well as the model proposed by weather16, perform better than the model proposed by lebret2016neural. Our final model with bifocal attention and gated orthogonalization gives the best performance and does 10% (relative) better than the closest baseline (basic seq2seq) and 21% (relative) better than the current state of the art method BIBREF0 . In Table TABREF16 , we show some qualitative examples of the output generated by different models.
Following Lebret et al. (2016), we used BLEU- 4, NIST-4 and ROUGE-4 as the evaluation metrics.
|
Which evaluation metric was chosen?
|
They used BLEU- 4, NIST-4 and ROUGE-4 as the evaluation metrics.
|
1704.08960
| false
| null |
FLOAT SELECTED: Table 3: Statistics of external data.
Neural network models for NLP benefit from pretraining of word/character embeddings, learning distributed sementic information from large raw texts for reducing sparsity. The three basic elements in our neural segmentor, namely characters, character bigrams and words, can all be pretrained over large unsegmented data. We pretrain the five-character window network in Figure FIGREF13 as an unit, learning the MLP parameter together with character and bigram embeddings. We consider four types of commonly explored external data to this end, all of which have been studied for statistical word segmentation, but not for neural network segmentors.
Raw Text. Although raw texts do not contain explicit word boundary information, statistics such as mutual information between consecutive characters can be useful features for guiding segmentation BIBREF11 . For neural segmentation, these distributional statistics can be implicitly learned by pretraining character embeddings. We therefore consider a more explicit clue for pretraining our character window network, namely punctuations BIBREF10 .
Automatically Segmented Text. Large texts automatically segmented by a baseline segmentor can be used for self-training BIBREF13 or deriving statistical features BIBREF12 . We adopt a simple strategy, taking automatically segmented text as silver data to pretrain the five-character window network. Given INLINEFORM0 , INLINEFORM1 is derived using the MLP in Figure FIGREF13 , and then used to classify the segmentation of INLINEFORM2 into B(begining)/M(middle)/E(end)/S(single character word) labels. DISPLAYFORM0
Heterogenous Training Data. Multiple segmentation corpora exist for Chinese, with different segmentation granularities. There has been investigation on leveraging two corpora under different annotation standards to improve statistical segmentation BIBREF16 . We try to utilize heterogenous treebanks by taking an external treebank as labeled data, training a B/M/E/S classifier for the character windows network. DISPLAYFORM0
POS Data. Previous research has shown that POS information is closely related to segmentation BIBREF14 , BIBREF15 . We verify the utility of POS information for our segmentor by pretraining a classifier that predicts the POS on each character, according to the character window representation INLINEFORM0 . In particular, given INLINEFORM1 , the POS of the word that INLINEFORM2 belongs to is used as the output. DISPLAYFORM0
FLOAT SELECTED: Table 3: Statistics of external data.
We consider four types of commonly explored external data to this end, all of which have been studied for statistical word segmentation, but not for neural network segmentors.
Raw Text.
Automatically Segmented Text.
Heterogenous Training Data.
POS Data.
|
What external sources are used?
|
Raw data from Gigaword, Automatically segmented text from Gigaword, Heterogenous training data from People's Daily, POS data from People's Daily
|
null | false
| null |
According to popular belief, the macaron has been introduced in France by the Italian chef of queen Catherine de Medici during the Renaissance but this appears to be a legend as the macaron is already mentioned in France in the 8th century in the Cormery Abbey.
|
Where do Macarons originate from?
|
The Macaron originates from France and is generally thought to have been created by an Italian chef of Catherine de Medici when she was the Queen of France during the 16th century. However the time period and original creator maybe a legend as there are documents referring to the Macaron in the 8th century from Cormery Abbey in France.
|
1909.08402
| false
| null |
In this paper we presented a way of enriching BERT with knowledge graph embeddings and additional metadata. Exploiting the linked knowledge that underlies Wikidata improves performance for our task of document classification. With this approach we improve the standard BERT models by up to four percentage points in accuracy. Furthermore, our results reveal that with task-specific information such as author names and publication metadata improves the classification task essentially compared a text-only approach. Especially, when metadata feature engineering is less trivial, adding additional task-specific information from an external knowledge source such as Wikidata can help significantly. The source code of our experiments and the trained models are publicly available.
With this approach we improve the standard BERT models by up to four percentage points in accuracy.
|
By how much do they outperform standard BERT?
|
The answers are shown as follows:
* up to four percentage points in accuracy
|
null | false
| null |
Pink Floyd are an English rock band formed in London in 1965. Gaining an early following as one of the first British psychedelic groups, they were distinguished by their extended compositions, sonic experimentation, philosophical lyrics and elaborate live shows. They became a leading band of the progressive rock genre, cited by some as the greatest progressive rock band of all time.
Pink Floyd were founded in 1965 by Syd Barrett (guitar, lead vocals), Nick Mason (drums), Roger Waters (bass guitar, vocals), and Richard Wright (keyboards, vocals). Under Barrett's leadership, they released two charting singles and the successful debut album The Piper at the Gates of Dawn (1967). The guitarist and vocalist David Gilmour joined in January 1968; Barrett left in April 1968 due to deteriorating mental health. Waters became the primary lyricist and thematic leader, devising the concepts behind Pink Floyd's most successful albums, The Dark Side of the Moon (1973), Wish You Were Here (1975), Animals (1977) and The Wall (1979). The musical film based on The Wall, Pink Floyd – The Wall (1982), won two BAFTA Awards. Pink Floyd also composed several film scores.
Following personal tensions, Wright left Pink Floyd in 1981, followed by Waters in 1985. Gilmour and Mason continued as Pink Floyd, rejoined later by Wright. They produced the albums A Momentary Lapse of Reason (1987) and The Division Bell (1994), backed by major tours, before entering a long hiatus. In 2005, all but Barrett reunited for a performance at the global awareness event Live 8. Barrett died in 2006, and Wright in 2008. The last Pink Floyd studio album, The Endless River (2014), was based on unreleased material from the Division Bell recording sessions. In 2022, Gilmour and Mason reformed Pink Floyd to release the song "Hey, Hey, Rise Up!" in protest of the Russo-Ukrainian War.
By 2013, Pink Floyd had sold more than 250 million records worldwide, making them one of the best-selling music artists of all time. The Dark Side of the Moon and The Wall were inducted into the Grammy Hall of Fame, and these albums and Wish You Were Here are among the best-selling albums of all time. Four Pink Floyd albums topped the US Billboard 200, and five topped the UK Albums Chart. Pink Floyd's hit singles include "See Emily Play" (1967), "Money" (1973), "Another Brick in the Wall, Part 2" (1979), "Not Now John" (1983), "On the Turning Away" (1987) and "High Hopes" (1994). They were inducted into the US Rock and Roll Hall of Fame in 1996 and the UK Music Hall of Fame in 2005. In 2008, Pink Floyd were awarded the Polar Music Prize in Sweden for their contribution to modern music.
|
Extract the name and the release year of the albums of the rock band Pink Floyd. Place them in a bullet list and separate name and year with a comma
|
- The Piper at the Gates of Dawn,1967
- The Dark Side of the Moon,1973
- Wish You Were Here,1975
- Animals,1977
- The Wall,1979
- A Momentary Lapse of Reason,1987
- The Division Bell,1994
- The Endless River,2014
|
null | false
| null |
Which is a bird or fish: Redfin perch, Red-breasted merganser
|
Red-breasted merganser is a bird, Redfin perch is a fish
|
|
1907.08501
| false
| null |
The same 36 questions were answered using four QALD tools: WDAqua BIBREF0 , QAKiS BIBREF7 , gAnswer BIBREF6 and Platypus BIBREF8 .
The same 36 questions were answered using four QALD tools: WDAqua BIBREF0 , QAKiS BIBREF7 , gAnswer BIBREF6 and Platypus BIBREF8 .
|
Which four QA systems do they use?
|
The answers are shown as follows:
* WDAqua BIBREF0 , QAKiS BIBREF7 , gAnswer BIBREF6 and Platypus BIBREF8
|
1911.00523
| false
| null |
Non-contextual properties of a word. These features are derived directly from the word and capture the general tendency of a word being echoed in explanations.
Non-contextual properties of a word. These features are derived directly from the word and capture the general tendency of a word being echoed in explanations.
|
What non-contextual properties do they refer to?
|
The answers are shown as follows:
* These features are derived directly from the word and capture the general tendency of a word being echoed in explanations.
|
null | false
| 213
|
According to The National Institute on Deafness, one in thousand infants is born deaf. An additional one to six per thousand are born with hearing loss at different levels BIBREF0. Sign language is commonly used by Deaf and Hard-of-Hearing (DHH) persons to communicate via hand gestures. An automatic sign language recognizer enables an ASL user to translate the sign language to written text or speech, allowing them to communicate with people who are not familiar with ASL. There is a tremendous rise in the popularity of personal digital assistants; available on user's personal and wearable devices (Google Now, Amazon Alexa and Apple Siri, etc.) and also in the form of standalone devices (Amazon Echo and Google Home smart speakers). These devices are primarily controlled through voice, and hence, their functionality is not readily available to DHH users. An automatic sign recognizer can also enable the interaction between a DHH user and a digital assistant.
Most current systems have capability of ASL recognition with RGB video data BIBREF1, BIBREF2, BIBREF3. An ASL sign is performed by a combination of hand gestures, facial expressions and postures of the body. Sequential motion of specific body locations (such as hand-tip, neck and arm) provide informative cues about a sign. Using video data, it is difficult to extract different body locations and associated motion sequences from a series of RGB frames. Microsoft Kinect is a 3D camera sensor which can use the depth information of a person to capture 3D coordinates of his/her body location across a video. This sequence of 3D body location is referred by skeletal data BIBREF4. To the best of our knowledge, there is no publicly available skeletal dataset in literature for ASL recognition.
With skeletal data, an ASL sign can be seen as a sequence of 3D coordinates or a 3D time series BIBREF5. Recurrent neural networks (RNN) have shown strong performance for sequential modeling BIBREF6. In this work, we investigate the impact of RGB video data in recognition accuracy when combined with skeletal data. We also propose a combined RNN network with a simple spatial data augmentation technique. In summary, the contributions of this work are:
We propose an RNN architecture with a novel spatial data augmentation technique.
We propose an architecture which uses both RGB and skeletal data to improve recognition accuracy.
We introduce and publicly release a multi–modal dataset for ASL called GMU-ASL51.
We propose an architecture which uses both RGB and skeletal data to improve recognition accuracy.
|
What kind of architecture do they propose?
|
An architecture which uses both RGB and skeletal data to improve recognition accuracy.
|
null | false
| 350
|
Machine reading comprehension and question answering has becomes a crucial application problem in evaluating the progress of AI system in the realm of natural language processing and understanding BIBREF0 . The computational linguistics communities have devoted significant attention to the general problem of machine reading comprehension and question answering.
However, most of existing reading comprehension tasks only focus on shallow QA tasks that can be tackled very effectively by existing retrieval-based techniques BIBREF1 . For example, recently we have seen increased interest in constructing extractive machine reading comprehension datasets such as SQuAD BIBREF2 and NewsQA BIBREF3 . Given a document and a question, the expected answer is a short span in the document. Question context usually contains sufficient information for identifying evidence sentences that entail question-answer pairs. For example, 90.2% questions in SQuAD reported by Min BIBREF4 are answerable from the content of a single sentence. Even in some multi-turn conversation tasks, the existing models BIBREF5 mostly focus on retrieval-based response matching.
In this paper, we focus on multiple-choice reading comprehension datasets such as RACE BIBREF6 in which each question comes with a set of answer options. The correct answer for most questions may not appear in the original passage which makes the task more challenging and allow a rich type of questions such as passage summarization and attitude analysis. This requires a more in-depth understanding of a single document and leverage external world knowledge to answer these questions. Besides, comparing to traditional reading comprehension problem, we need to fully consider passage-question-answer triplets instead of passage-question pairwise matching.
In this paper, we propose a new model, Dual Co-Matching Network, to match a question-answer pair to a given passage bidirectionally. Our network leverages the latest breakthrough in NLP: BERT BIBREF7 contextual embedding. In the origin BERT paper, the final hidden vector corresponding to first input token ([CLS]) is used as the aggregation representation and then a standard classification loss is computed with a classification layer. We think this method is too rough to handle the passage-question-answer triplet because it only roughly concatenates the passage and question as the first sequence and uses question as the second sequence, without considering the relationship between the question and the passage. So we propose a new method to model the relationship among the passage, the question and the candidate answer.
Firstly we use BERT as our encode layer to get the contextual representation of the passage, question, answer options respectively. Then a matching layer is constructed to get the passage-question-answer triplet matching representation which encodes the locational information of the question and the candidate answer matched to a specific context of the passage. Finally we apply a hierarchical aggregation method over the matching representation from word-level to sequence-level and then from sequence level to document-level. Our model improves the state-of-the-art model by 2.6 percentage on the RACE dataset with BERT base model and further improves the result by 3 percentage with BERT large model.
Especially, our single model performs even better than human turkers on the RACE full dataset, which is the first milestone achievement ever since the RACE challenge has been set up.
|
Whose performance is better on the RACE full dataset? Their model or human turkers?
|
Their single model.
|
null | false
| null |
Should a man get married when he is young, or wait until he is older?
|
Most importantly, a person should get married when they find the right person. But assuming one has found the right person, there are comparative advantages and disadvantages to marrying at a young age or an older age. Men have the advantage of not needing to worry about the 'biological clock' - unlike women, their reproductive age can extend into quite a late age, whereas women typically lose the ability to reproduce once they hit around the age of 40. An advantage of having children when one is young is that a man will have more energy and physical dexterity to be able to engage in physical activities with his children, like camping, biking, hiking, sports, skiing, etc. Bending over to change diapers will not be as painful on the back and knees. As a younger man, travel with children may be less discomforting. As an older man, one might have more time or flexibility of time to spend with one's children. An older man may also have greater financial resources to help make child raising easier. For example, an older man might have a larger house, or a greater ability to afford in-house child care. On the other hand, there are studies that show having an older father may increase the odds of a child having autism or attention deficit disorder. Anecdotally, it seems people who marry at a young age are more likely to divorce. But perhaps having two marriages, one when one is young and another when one is older, is not a bad thing. Men have the advantage of being able to sire two sets of children at different times in their lives, making this possible. But ultimately the decision should be based on whether or not you have met the person you know without a doubt you want to spend the rest of your life with. If you have found that person, you should stay with them.
|
|
1910.07181
| false
| null |
We evalute Bertram on the WNLaMPro dataset of BIBREF0. This dataset consists of cloze-style phrases like
We evalute Bertram on the WNLaMPro dataset of BIBREF0.
|
What is dataset for word probing task?
|
The answers are shown as follows:
* WNLaMPro dataset
|
null | false
| null |
How can I become the best person?
|
To want to be the best person, you have to know who or what you want to compare yourself to. It is very advisable to compare yourself only with yourself. Where am I currently in life, what do I want to change and what goals do I want to achieve? Examples are: to have more friends, to be healthier, to develop a better contact with the children, to work less etc.. So the question should be: how can I become the best person I can be? It makes no sense to compare yourself with others. Only compare to yourself.
|
|
null | false
| 295
|
First, we evaluated the standard HMM model with an uninformative prior (this will be our baseline) for the two different input features: MFCC (and derivatives) and MBN. Results are shown in Table TABREF20 . Surprisingly, the MBN features perform relatively poorly compared to the standard MFCC. These results are contradictory to those reported in BIBREF3 . Two factors may explain this discrepancy: the Mboshi5k data being different from the training data of the MBN neural network, the neural network may not generalize well. Another possibility may be that the initialization scheme of the model is not suitable for this type of features. Indeed, Variational Bayesian Inference algorithm converges only to a local optimum of the objective function and is therefore dependent of the initialization. We believe the second explanation is the more likely since, as we shall see shortly, the best results in term of word segmentation and NMI are eventually obtained with the MBN features when the inference is done with the informative prior. Next, we compared the HMM and the SVAE models when trained with an uninformative prior (lines with "Inf. Prior" set to "no" in Table TABREF23 ). The SVAE significantly improves the NMI and the precision showing that it extracts more consistent units than the HMM model. However, it also degrades the segmentation in terms of recall. We further investigated this behavior by looking at the duration of the units found by both models compared to the true phones (Table TABREF22 ). We observe that the SVAE model favors longer units than the HMM model hence leading to fewer boundaries and consequently smaller recall.
We then evaluated the effect of the informative prior on the acoustic unit discovery (Table TABREF23 ). On all 4 combinations (2 features sets INLINEFORM0 2 models) we observe an improvement in terms of precision and NMI but a degradation of the recall. This result is encouraging since the informative prior was trained on English data (TIMIT) which is very different from Mboshi. Indeed, this suggests that even speech from an unrelated language can be of some help in the design of an ASR for a very low resource language. Finally, similarly to the SVAE/HMM case described above, we found that the degradation of the recall is due to longer units discovered for models with an informative prior (numbers omitted due to lack of space).
Word discovery results are given in Table TABREF21 for the Boundary metric BIBREF20 , BIBREF21 . We observe that i) the best word boundary detection (F-score) is obtained with MBN features, an informative prior and the SVAE model; this confirms the results of table TABREF23 and shows that better AUD leads to better word segmentation ii) word segmentation from AUD graph Lattices is slightly better than from flat sequences of AUD symbols (1-best); iii) our results outperform a pure speech based baseline based on segmental DTW BIBREF22 (F-score of 19.3% on the exact same corpus).
First, we evaluated the standard HMM model with an uninformative prior (this will be our baseline) for the two different input features: MFCC (and derivatives) and MBN.
|
What are the two different input features?
|
MFCC (and derivatives) and MBN.
|
1701.03051
| false
| null |
The baseline model for our experiments is explained in the paper by Alec Go [1]. The model uses the Naive Bayes, SVM, and the Maximum Entropy classifiers for their experiment. Their feature vector is either composed of Unigrams, Bigrams, Unigrams + Bigrams, or Unigrams + POS tags.
The baseline model for our experiments is explained in the paper by Alec Go [1]. The model uses the Naive Bayes, SVM, and the Maximum Entropy classifiers for their experiment.
|
What previously proposed methods is this method compared against?
|
The answers are shown as follows:
* Naive Bayes
* SVM
* Maximum Entropy classifiers
|
null | false
| null |
Because endocrinology encompasses so many conditions and diseases, there are many organizations that provide education to patients and the public. The Hormone Foundation is the public education affiliate of The Endocrine Society and provides information on all endocrine-related conditions. Other educational organizations that focus on one or more endocrine-related conditions include the American Diabetes Association, Human Growth Foundation, American Menopause Foundation, Inc., and Thyroid Foundation of America.[citation needed]
In North America the principal professional organizations of endocrinologists include The Endocrine Society, the American Association of Clinical Endocrinologists, the American Diabetes Association, the Lawson Wilkins Pediatric Endocrine Society, and the American Thyroid Association.
In Europe, the European Society of Endocrinology (ESE) and the European Society for Paediatric Endocrinology (ESPE) are the main organisations representing professionals in the fields of adult and paediatric endocrinology, respectively.
In the United Kingdom, the Society for Endocrinology and the British Society for Paediatric Endocrinology and Diabetes are the main professional organisations.
The European Society for Paediatric Endocrinology is the largest international professional association dedicated solely to paediatric endocrinology. There are numerous similar associations around the world.
|
List some of the endocrinology societies and organizations within Europe and specifically talk about the United Kingdom.
|
Specifically in the United Kingdom, the Society for Endocrinology and the British Society for Paediatric Endocrinology and Diabetes are the main professional endocrinology organisations. Europe also has the following organisations: the European Society of Endocrinology (ESE) and the European Society for Paediatric Endocrinology (ESPE).
|
null | false
| null |
What is Jon Snow's true name?
|
Aegon Targaryen
|
|
null | false
| 168
|
Knowledge representation and reasoning (KRR) is the process of representing the domain knowledge in formal languages (e.g., SPARQL, Prolog) such that it can be used by expert systems to execute querying and reasoning services. KRR have been applied in many fields including financial regulations, medical diagnosis, laws, and so on. One major obstacle in KRR is the creation of large-scale knowledge bases with high quality. For one thing, this requires the knowledge engineers (KEs) not only to have the background knowledge in a certain domain but have enough skills in knowledge representation as well. Unfortunately, qualified KEs are also in short supply. Therefore, it would be useful to build a tool that allows the domain experts without any background in logic to construct and query the knowledge base simply from text.
Controlled natural languages (CNLs) BIBREF0 were developed as a technology that achieves this goal. CNLs are designed based on natural languages (NLs) but with restricted syntax and interpretation rules that determine the unique meaning of the sentence. Representative CNLs include Attempto Controlled English BIBREF1 and PENG BIBREF2 . Each CNL is developed with a language parser which translates the English sentences into an intermediate structure, discourse representation structure (DRS) BIBREF3 . Based on the DRS structure, the language parsers further translate the DRS into the corresponding logical representations, e.g., Answer Set Programming (ASP) BIBREF4 programs. One main issue with the aforementioned CNLs is that the systems do not provide enough background knowledge to preserve semantic equivalences of sentences that represent the same meaning but are expressed via different linguistic structures. For instance, the sentences Mary buys a car and Mary makes a purchase of a car are translated into different logical representations by the current CNL parsers. As a result, if the user ask a question who is a buyer of a car, these systems will fail to find the answer.
In this thesis proposal, I will present KALM BIBREF5 , BIBREF6 , a system for knowledge authoring and question answering. KALM is superior to the current CNL systems in that KALM has a complex frame-semantic parser which can standardize the semantics of the sentences that express the same meaning via different linguistic structures. The frame-semantic parser is built based on FrameNet BIBREF7 and BabelNet BIBREF8 where FrameNet is used to capture the meaning of the sentence and BabelNet BIBREF8 is used to disambiguate the meaning of the extracted entities from the sentence. Experiment results show that KALM achieves superior accuracy in knowledge authoring and question answering as compared to the state-of-the-art systems.
The rest parts are organized as follows: Section SECREF2 discusses the related works, Section SECREF3 presents the KALM architecture, Section SECREF4 presents KALM-QA, the question answering part of KALM, Section SECREF5 shows the evaluation results, Section SECREF6 shows the future work beyond the thesis, and Section SECREF7 concludes the paper.
KALM is superior to the current CNL systems in that KALM has a complex frame-semantic parser which can standardize the semantics of the sentences that express the same meaning via different linguistic structures.
|
Why the KALM is superior to the current CNL systems?
|
Because the KALM has a complex frame-semantic parser which can standardize the semantics of the sentences that express the same meaning via different linguistic structures.
|
null | false
| null |
What are things people do in the morning?
|
Things that people tend to do in the morning are to wake up, drink coffee and have breakfast.
|
|
null | false
| null |
What is India’s rank in UN Human Development Index 2021?
|
132nd (total nations 191)
|
|
null | false
| null |
Hispania (Latin: Hispānia [hɪsˈpaːnia], Spanish: [isˈpanja]; nearly identically pronounced in Spanish, Portuguese, Catalan, and Italian) was the Roman name for the Iberian Peninsula and its provinces. Under the Roman Republic, Hispania was divided into two provinces: Hispania Citerior and Hispania Ulterior. During the Principate, Hispania Ulterior was divided into two new provinces, Baetica and Lusitania, while Hispania Citerior was renamed Hispania Tarraconensis. Subsequently, the western part of Tarraconensis was split off, first as Hispania Nova, later renamed "Callaecia" (or Gallaecia, whence modern Galicia). From Diocletian's Tetrarchy (AD 293) onwards, the south of the remainder of Tarraconensis was again split off as Carthaginensis, and all of the mainland Hispanic provinces, along with the Balearic Islands and the North African province of Mauretania Tingitana, were later grouped into a civil diocese headed by a vicarius. The name Hispania was also used in the period of Visigothic rule.
The modern place names Spain and Hispaniola are both derived from Hispania.
|
What is Hispania
|
Hispania (Latin: Hispānia [hɪsˈpaːnia], Spanish: [isˈpanja]; nearly identically pronounced in Spanish, Portuguese, Catalan, and Italian) was the Roman name for the Iberian Peninsula and its provinces. Under the Roman Republic, Hispania was divided into two provinces: Hispania Citerior and Hispania Ulterior. During the Principate, Hispania Ulterior was divided into two new provinces, Baetica and Lusitania, while Hispania Citerior was renamed Hispania Tarraconensis.
Subsequently, the western part of Tarraconensis was split off, first as Hispania Nova, later renamed "Callaecia" (or Gallaecia, whence modern Galicia). From Diocletian's Tetrarchy (AD 293) onwards, the south of the remainder of Tarraconensis was again split off as Carthaginensis, and all of the mainland Hispanic provinces, along with the Balearic Islands and the North African province of Mauretania Tingitana, were later grouped into a civil diocese headed by a vicarius. The name Hispania was also used in the period of Visigothic rule.
The modern place names Spain and Hispaniola are both derived from Hispania.
|
null | false
| null |
Give me a list of the greatest basketball players of all time.
|
Michael Jordan and Lebron James are often considered as the two greatest basketball players of all time. In addition, Kareem Abdul-Jabbar, Bill Russell, Magic Johnson, Kobe Bryant, and Larry Bird are all considered amongst the best basketball players to play the game.
|
|
null | false
| 306
|
Recent years, great efforts have been made on extracting relational fact from unstructured raw texts to build large structural knowledge bases. A relational fact is often represented as a triplet which consists of two entities (subject and object) and semantic relation between them. Early works BIBREF0, BIBREF1, BIBREF2 mainly focused on the task of relation classification which assumes the entity pair are identified beforehand. This limits their practical application since they neglect the extraction of entities. To extract both entities and their relation, existing methods can be divided into two categories : the pipelined framework, which first uses sequence labeling models to extract entities, and then uses relation classification models to identify the relation between each entity pair; and the joint approach, which combines the entity model and the relation model through different strategies, such as constraints or parameters sharing.
This tagging based method is better than most of the existing pipelined methods, but its flexibility is limited and can not tackle the situations when (1) one entity belongs to multiple triplets (2) multiple entities have overlaps.
|
What situations can not be tackled by tagging-based method?
|
(1) one entity belongs to multiple triplets (2) multiple entities have overlaps.
|
null | false
| null |
SoftBank Group Corp. (ソフトバンクグループ株式会社, SofutoBanku Gurūpu Kabushiki-gaisha) is a Japanese multinational conglomerate holding company headquartered in Minato, Tokyo which focuses on investment management. The Group primarily invests in companies operating in technology that offer goods and services to customers in a multitude of markets and industries ranging from the internet to automation. With over $100 billion in capital at its onset, SoftBank’s Vision Fund is the world's largest technology-focused venture capital fund. Fund investors included sovereign wealth funds from countries in the Middle East.The company is known for the leadership of its controversial founder and largest shareholder Masayoshi Son. Its investee companies, subsidiaries and divisions, including several unprofitable unicorns, operate in robotics, artificial intelligence, logistics, transportation, proptech, real estate, hospitality, broadband, fixed-line telecommunications, e-commerce, information technology, finance, media and marketing, and other areas. Among its biggest and most internationally recognizable current stockholdings are sizeable stakes in Arm (semiconductors), Alibaba (e-commerce), OYO Rooms (hospitality), WeWork (coworking) and Deutsche Telekom (telecommunications). SoftBank Corporation, its spun-out affiliate and former flagship business, is the third-largest wireless carrier in Japan, with 45.621 million subscribers as of March 2021.SoftBank was ranked in the 2017 Forbes Global 2000 list as the 36th largest public company in the world and the second-largest publicly traded company in Japan after Toyota.The logo of SoftBank is based on the flag of the Kaientai, a naval trading company founded in 1865, near the end of the Tokugawa shogunate, by Sakamoto Ryōma.Although SoftBank does not affiliate itself to any traditional keiretsu, it has close ties with Mizuho Financial Group, its primary lender.
|
From the passage mention the different verticals in which SoftBank Group Corp. invests in. Display the results in comma separated format.
|
robotics, artificial intelligence, logistics, transportation, proptech, real estate, hospitality, broadband, fixed-line telecommunications, e-commerce, information technology, finance, media and marketing
|
null | false
| null |
Which ones of the following are considered nutrient-dense foods: soda, vegetables, fruits, chips, whole grains, seafood, cookies, eggs, beans, peas, lentils, unsalted nuts and seeds, fat-free and low-fat dairy products, fried chicken, lean meats and poultry.
|
Vegetables, fruits, whole grains, seafood, eggs, beans, peas, and lentils, unsalted nuts and seeds, fat-free and low-fat dairy products, and lean meats and poultry are nutrient-dense foods.
|
|
null | false
| null |
What is the largest land animal in the world?
|
African Elephant
|
|
null | false
| null |
What is considered a good meal?
|
Have plenty and balanced intake of carbs, proteins and fats. Have various ingredients including greens.
|
|
null | false
| null |
How to write a college essay?
|
A college essay is likely the most important part of your college application. While it's important to be factual about the writing, the essay is meant to demonstrate you as a unique person, that's set apart from the other applicants. Examples are unique experiences that changed the path of your life; special activities you have involved with that made a difference to others' lives, and interests that you have devoted much of your spare time on. The goal is help the admission officer to know you from aspects other than your grades and SAT scores, and evaluate whether you are good fit for the culture of the college you are applying for.
|
|
null | false
| null |
Away from his expeditions, Shackleton's life was generally restless and unfulfilled. In his search for rapid pathways to wealth and security, he launched business ventures which failed to prosper, and he died heavily in debt. Upon his death, he was lauded in the press but was thereafter largely forgotten, while the heroic reputation of his rival Scott was sustained for many decades. Later in the 20th century, Shackleton was "rediscovered", and became a role model for leadership in extreme circumstances.
|
Besides his expeditions, what was Shackleton's life like?
|
Despite being considered today a role model for leadership in extreme circumstances, Shackleton was largely forgotten for decades after his death. He was focused on and participated in several get rich quick schemes but they all ended in failure. Shackleton died heavily in debt.
|
null | false
| null |
Both the modern free software movement and the Open Source Initiative were born from a common history of Unix, Internet free software, and the hacker culture, but their basic goals and philosophy differ, the free software movement being more focused on the ethics of software, and their open source counterparts being more focused on practical benefits. The Open Source Initiative chose the term "open source," in founding member Michael Tiemann's words, to "dump the moralizing and confrontational attitude that had been associated with 'free software'" and instead promote open source ideas on "pragmatic, business-case grounds."
|
Which is better the Free Software Movement or the Open Source Initiative?
|
The Free Software Movement is the better organization for me since I also hold a high value on ethics and believe in that vision.
|
null | false
| null |
What are foundational models?
|
Foundational Models as the name says can be the foundation for many applications of the AI Model, using self-supervised learning and transfer learning, the model can apply information it learnt about one situation to another. As an example. You learn to drive one car and without too much effort you can drive most of the other cars, even trucks or a bus.
|
|
1912.07976
| false
| null |
We build a joint model for the multi-task of ATE and APC based on the BERT-BASE model. After optimizing the model parameters according to the empirical result, the joint model based on BERT-BASE achieved hopeful performance on all three datasets and even surpassed other proposed BERT based improved models on some datasets, such as BERT-PT, AEN-BERT, SDGCN-BERT, and so on. Meanwhile, we implement the joint-task model based on BERT-SPC. Compared with the BERT-BASE model, BERT-SPC significantly improves the accuracy and F1 score of aspect polarity classification. In addition, for the first time, BERT-SPC has increased the F1 score of ATE subtask on three datasets up to 99%.
Compared with the BERT-BASE model, BERT-SPC significantly improves the accuracy and F1 score of aspect polarity classification.
|
How much better is performance of the proposed model compared to the state of the art in these various experiments?
|
The answers are shown as follows:
* significantly improves the accuracy and F1 score of aspect polarity classification
|
null | false
| null |
What is the smallest country in the world?
|
The smallest country in the world is Vatican City, with an area of less than .5 square kilometers. It is fully encircled by the Italian capital of Rome, and has existed as a sovereign nation since 1929. Due to its small size it is considered to be a microstate.
|
|
null | false
| null |
Are universities becoming obsolete ?
|
The combination of skyrocketing tuition fees and the increase in adoption in remote learning might make most universities obsolete in the future. It’s completely dependent on the individual’s expectation from a University level education but the goals of students are changing. The ability to network or gain popularity to excel your potential career path or brand recognition is now possible outside these institutes. Even the access to venture capital and access to business, athletic and other titans of industry, can be accomplished via other paths.
|
|
2004.04435
| false
| null |
In this section, we empirically compare automatic differentiation (AD, our implementation based on Clad) and numerical differentiation (ND, based on finite difference method) in ROOT. We show that AD can drastically improve accuracy and performance of derivative evaluation, compared to ND.
In this section, we empirically compare automatic differentiation (AD, our implementation based on Clad) and numerical differentiation (ND, based on finite difference method) in ROOT. We show that AD can drastically improve accuracy and performance of derivative evaluation, compared to ND.
|
How is correctness of automatic derivation proved?
|
The answers are shown as follows:
* empirically compare automatic differentiation (AD, our implementation based on Clad) and numerical differentiation (ND, based on finite difference method)
|
null | false
| null |
write a short essay on day spent in central park
|
Central Park is one of the most iconic landmarks in New York City, and a popular destination for locals and tourists alike. Spending a day in Central Park is an unforgettable experience, filled with beauty, excitement, and relaxation.
The park offers a wide range of activities and attractions, from boating on the lake and exploring the many walking paths, to visiting the Central Park Zoo and the famous Alice in Wonderland statue. There are also numerous opportunities for outdoor sports and recreation, including baseball fields, basketball courts, and running trails.
One of the most popular activities in Central Park is picnicking. Visitors can spread out a blanket on the grass and enjoy a meal with friends or family while taking in the stunning views of the park's natural beauty. There are also plenty of food vendors and cafes throughout the park, offering everything from hot dogs and ice cream to gourmet sandwiches and salads.
Central Park is also home to many cultural events and performances throughout the year. From concerts and plays to art exhibits and film screenings, there is always something exciting happening in the park. Visitors can check the park's schedule to find out about upcoming events and plan their visit accordingly.
Whether you're looking for a relaxing day in nature or an action-packed adventure, a day spent in Central Park has something for everyone. The park's beauty and charm make it a must-visit destination for anyone visiting New York City, and a favorite spot for locals to escape the hustle and bustle of city life.
|
|
null | false
| null |
Can you tell me what of these are capital cities of their respective countries? Manchester, Madrid, Paris, Milan, London and Munich
|
A capital city is the city where the government of the country resides. From the previous lists only Madrid, Paris and London are capital cities. Manchester, Milan and Munich are large and important cities in Europe but they are not their country's capital. Some cities like Milan might be consider their country's economic capital as they are the main economic hub of Italy but they are still not their capital city which is Rome.
|
|
null | false
| 270
|
Princeton WordNet BIBREF0 is one of the most important resources used in many different tasks across linguistics and natural language processing, however the resource is only available for English and is limited in its coverage of real world concepts. To cross the language barrier, huge efforts have been made to extend the Princeton WordNet with multilingual information in projects, such as EuroWordNet BIBREF1 , BalkaNet BIBREF2 and MultiWordNet BIBREF3 , mostly following the extend approach, where the structure of the Princeton WordNet is preserved and only the words in each synset are translated and new synsets are added for concepts. Furthermore, the Princeton WordNet has many fewer concepts than large scale encyclopedias such as Wikipedia and resources derived from it such as DBpedia BIBREF4 and BabelNet BIBREF5 . This problem is even worse for many non-English wordnets, due to the extend approach, as these resources have even fewer synsets than Princeton WordNet. Furthermore, there are still many languages for which a wordnet does not exist or is not available to all potential users due to licensing restrictions.
To address these deficiencies we propose two approaches. Firstly, we apply high-quality statistical machine translation (SMT) to automatically translate the WordNet entries into several different European languages. While an SMT system can only return the most frequent translation when given a term by itself, we propose a novel method to provide strong word sense disambiguation when translating wordnet entries. In addition, our method can handle fundamental complexities such as the need to translate all senses of a word including low-frequency senses, which is very challenging for current SMT approaches. For these reasons, we leverage existing translations of Princeton WordNet entries in other languages to identify contextual information for wordnet senses from a large set of generic parallel corpora. The goal is to identify sentences that share the same semantic information in respect to the synset of the Princeton WordNet entry that we want to translate. Secondly, we describe a novel system based on state-of-the-art semantic textual similarity and ontology alignment to establish a new linking between Princeton WordNet and DBpedia. This method uses a multi-feature approach to establish similarities between synsets and DBpedia entities based on analysis of the definitions using a variety of methods from simple string statistics, to methods based on explicit semantic analysis as well as deep learning methods including long short-term memory (LSTM) networks. These statistics are created based on the Princeton WordNet synset gloss as well as the neighbouring words in the WordNet graph. These are combined using a constraint-based solver that considers not only the semantic similarity of the synsets but also the overall structure of the alignment and its consistency, following the best practices in ontology alignment.
This work has led to the development of a large multilingual WordNet in more than 20 European languages, which we call Polylingual WordNet BIBREF6 , which is available under an open (CC-BY) license. Finally, we describe how this resource is published, firstly as linked data in the linguistic linked open data cloud, as well as published in all the formats of the Global WordNet Association Interlingual Index.
These statistics are created based on the Princeton WordNet synset gloss as well as the neighbouring words in the WordNet graph.
|
What dataset is used when creating the statistics?
|
The Princeton WordNet synset.
|
null | false
| 365
|
Words with multiple senses commonly exist in many languages. For example, the word bank can either mean a “financial establishment” or “the land alongside or sloping down to a river or lake”, based on different contexts. Such a word is called a “polyseme”. The task to identify the meaning of a polyseme in its surrounding context is called word sense disambiguation (WSD). Word sense disambiguation is a long-standing problem in natural language processing (NLP), and has broad applications in other NLP problems such as machine translation BIBREF0 . Lexical sample task and all-word task are the two main branches of WSD problem. The former focuses on only a pre-selected set of polysemes whereas the later intends to disambiguate every polyseme in the entire text. Numerous works have been devoted in WSD task, including supervised, unsupervised, semi-supervised and knowledge based learning BIBREF1 . Our work focuses on using supervised learning to solve all-word WSD problem.
Most supervised approaches focus on extracting features from words in the context. Early approaches mostly depend on hand-crafted features. For example, IMS by BIBREF2 uses POS tags, surrounding words and collections of local words as features. These approaches are later improved by combining with word embedding features BIBREF0 , which better represents the words' semantic information in a real-value space. However, these methods neglect the valuable positional information between the words in the sequence BIBREF3 . The bi-directional Long-Short-Term-Memory (LSTM) approach by BIBREF3 provides one way to leverage the order of words. Recently, BIBREF4 improved the performance by pre-training a LSTM language model with a large unlabelled corpus, and using this model to generate sense vectors for further WSD predictions. However, LSTM significantly increases the computational complexity during the training process.
The development of the so called “fixed-size ordinally forgetting encoding” (FOFE) has enabled us to consider more efficient method. As firstly proposed in BIBREF5 , FOFE provides a way to encode the entire sequence of words of variable length into an almost unique fixed-size representation, while also retain the positional information for words in the sequence. FOFE has been applied to several NLP problems in the past, such as language model BIBREF5 , named entity recognition BIBREF6 , and word embedding BIBREF7 . The promising results demonstrated by the FOFE approach in these areas inspired us to apply FOFE in solving the WSD problem. In this paper, we will first describe how FOFE is used to encode sequence of any length into a fixed-size representation. Next, we elaborate on how a pseudo language model is trained with the FOFE encoding from unlabelled data for the purpose of context abstraction, and how a classifier for each polyseme is built from context abstractions of its labelled training data. Lastly, we provide the experiment results of our method on several WSD data sets to justify the equivalent performance as the state-of-the-art approach.
Next, we elaborate on how a pseudo language model is trained with the FOFE encoding from unlabelled data for the purpose of context abstraction, and how a classifier for each polyseme is built from context abstractions of its labelled training data.
|
What model do they train for context abstraction?
|
A pseudo language model.
|
null | false
| 284
|
With the massive increase in social interactions on online social networks, there has also been an increase of hateful activities that exploit such infrastructure. On Twitter, hateful tweets are those that contain abusive speech targeting individuals (cyber-bullying, a politician, a celebrity, a product) or particular groups (a country, LGBT, a religion, gender, an organization, etc.). Detecting such hateful speech is important for analyzing public sentiment of a group of users towards another group, and for discouraging associated wrongful activities. It is also useful to filter tweets before content recommendation, or learning AI chatterbots from tweets.
The manual way of filtering out hateful tweets is not scalable, motivating researchers to identify automated ways. In this work, we focus on the problem of classifying a tweet as racist, sexist or neither. The task is quite challenging due to the inherent complexity of the natural language constructs – different forms of hatred, different kinds of targets, different ways of representing the same meaning. Most of the earlier work revolves either around manual feature extraction BIBREF0 or use representation learning methods followed by a linear classifier BIBREF1 , BIBREF2 . However, recently deep learning methods have shown accuracy improvements across a large number of complex problems in speech, vision and text applications. To the best of our knowledge, we are the first to experiment with deep learning architectures for the hate speech detection task.
In this paper, we experiment with multiple classifiers such as Logistic Regression, Random Forest, SVMs, Gradient Boosted Decision Trees (GBDTs) and Deep Neural Networks(DNNs). The feature spaces for these classifiers are in turn defined by task-specific embeddings learned using three deep learning architectures: FastText, Convolutional Neural Networks (CNNs), Long Short-Term Memory Networks (LSTMs). As baselines, we compare with feature spaces comprising of char n-grams BIBREF0 , TF-IDF vectors, and Bag of Words vectors (BoWV).
Main contributions of our paper are as follows: (1) We investigate the application of deep learning methods for the task of hate speech detection. (2) We explore various tweet semantic embeddings like char n-grams, word Term Frequency-Inverse Document Frequency (TF-IDF) values, Bag of Words Vectors (BoWV) over Global Vectors for Word Representation (GloVe), and task-specific embeddings learned using FastText, CNNs and LSTMs. (3) Our methods beat state-of-the-art methods by a large margin ( INLINEFORM0 18 F1 points better).
As baselines, we compare with feature spaces comprising of char n-grams [6], TF-IDF vectors, and Bag of Words vectors (BoWV).
|
What are the baselines?
|
Char n-grams, TF-IDF vectors, and Bag of Words vectors (BoWV).
|
null | false
| 203
|
We base our experiments on a large collection of Bible translations crawled from the web, coming from various sources and periods of times. Any other multilingual data collection would work as well, but with the selected corpus we have the advantage that we cover the same genre and roughly the same coverage for each language involved. It is also easy to divide the data into training and test sets by using Bible verse numbers, which allows us to control for semantic similarity between languages in a way that would have been difficult in a corpus that is not multi-parallel. Altogether we have 1,303 translations in 990 languages that we can use for our purposes. These were chosen so that the model alphabet size is below 1000 symbols, which was satisfied by choosing only translations in Latin, Cyrillic or Greek script.
Certainly, there are disadvantages as well, such as the limited size (roughly 500 million tokens in total, with most languages having only one translation of the New Testament each, with roughly 200 thousand tokens), the narrow domain and the high overlap of named entities. The latter can lead to some unexpected effects when using nonsensical language vectors, as the model will then generate a sequence of random names.
The corpus deviates in some ways from an ideal multi-parallel corpus. Most translations are of the complete New Testament, whereas around 300 also contain the Old Testament (thus several times longer), and around ten contain only portions of the New Testament. Additionally, several languages have multiple translations, which are then concatenated. These translations may vary in age and style, but historical versions of languages (with their own ISO 639-3 code) are treated as distinct languages. During training we enforce a uniform distribution between languages when selecting training examples.
We base our experiments on a large collection of Bible translations crawled from the web, coming from various sources and periods of times.
|
What data do their experiments base on?
|
A large collection of Bible translations crawled from the web.
|
null | false
| null |
Name some celebrities who have been diagnosed with multiple sclerosis.
|
Jack Osbourne, Selma Blair, Jason DaSilva, Richard Pryor
|
|
null | false
| 254
|
Every day textual information is being added/updated on Wikipedia, as well as other social media platforms like Facebook, Twitter, etc. These platforms receive a huge amount of unverified textual data from all its users such as News Channels, Bloggers, Journalists, Field-Experts which ought to be verified before other users start consuming it. This information boom has increased the demand of information verification also known as Fact Checking. Apart from the encyclopedia and other platforms, domains like scientific publications and e-commerce also require information verification for reliability purposes. Generally, Wikipedia authors, bloggers, journalists and scientists provide references to support their claims. Providing referenced text against the claims makes the fact checking task a little easier as the verification system no longer needs to search for the relevant documents.
Wikipedia manages to verify all this new information with a number of human reviewers. Manual review processes introduce delays in publishing and is not a well scalable approach. To address this issue, researchers have launched relevant challenges, such as the Fake News Challenge (BIBREF0), Fact Extraction and VERification (FEVER) (BIBREF1) challenge along with the datasets. Moreover, Thorne and Vlachos (BIBREF2) released a survey on the current models for automated fact-checking. FEVER is the largest dataset and contains around 185k claims from the corpus of 5.4M Wikipedia articles. The claims are labeled as “SUPPORTS”, “REFUTES”, or “NOT ENOUGH INFO”, based on the evidence set.
In this paper, we propose an unsupervised question-answering based approach for solving the fact-checking problem. This approach is inspired from the memory-based reading comprehension task that humans perform at an early age. As we know that kids in schools, first read and learn the syllabus content so that they can answer the questions in the exam. Similarly, our model learns a language model and linguistics features in unsupervised fashion from the provided Wikipedia pages.
To transform the FEVER dataset into the above-mentioned task, we first generate the questions from the claims. In literature, there are majorly two types of Question Generation systems: Rule-based and Neural Question Generation (NQG) model based. Ali et al. (BIBREF3) proposed a rule-based pipeline to automate the question generation using POS (Part-of-speech) tagging and Named Entity Recognition (NER) tagging from the sentences. Recently, many NQG models have been introduced to generate questions in natural language. Serban et al. (BIBREF4) achieved better performance for question generation utilizing (passage, question, answer) triplets as training data and an encoder-decoder based architecture as their learning model.
Du et al. (BIBREF5) introduced a sequence-to-sequence model with an attention mechanism, outperforming rule-base question generation systems. Although the models proposed in (BIBREF6; BIBREF7) are effective, they require a passage to generate the plausible questions which is not readily available in the FEVER dataset. To resolve the issues and to keep the system simple but effective, we chose to generate questions similar to a Cloze-task or masked language modeling task. Such a task makes the problem more tractable as the masked entities are already known (i.e. named entities) and tight as there is only one correct answer for a given question. Later when the answers are generated, due to the question generation process, it becomes very easy to identify the correct answers.
We use the BERT's (Bidirectional Encoder Representations from Transformers) (BIBREF8) masked language model, that is pre-trained on Wikipedia articles for predicting the masked entities. Currently, neither the claim verification process nor the question generation process mandates explicit reasoning. For the same reason, it is difficult to put “REFUTES” or “NOT ENOUGH INFO” labels. To resolve this issue, we classify the unsupported claims as “MANUAL_REVIEW” instead of labeling them as “NOT ENOUGH INFO” or “REFUTES”.
In the literature, the shared task has been tackled using pipeline-based supervised models (BIBREF9; BIBREF10; BIBREF11). To our knowledge, only BIBREF10 has provided the confusion matrix for each of the labels for their supervised system. For the same reason, we are only providing the comparison of the label accuracy on the “SUPPORTS” label in the results section.
We use the BERT's (Bidirectional Encoder Representations from Transformers) masked language model, that is pre-trained on Wikipedia articles for predicting the masked entities.
|
What model do they use?
|
BERT's (Bidirectional Encoder Representations from Transformers) masked language model.
|
null | false
| null |
How do you subtract two numbers?
|
Subtraction is an operation that identifies the total count of objects in a collection of items after a specific count of items are removed from the starting total.
You can subtract two numbers by starting with an initial count, which represents the number to subtract from, and a second count which represents the number to subtract. Then remove one from the starting count repeating this operation for the number of items in the subtract value.
Subtraction is not associative, which means that the order in which subtraction is performed matters. For instance, seven minus five (7 - 3) equals four (= 4) but four minus seven (4 - 7) equals minus three (= -3).
|
|
null | false
| null |
What are some ways of traveling from Washington D.C to San Francisco?
|
- take a direct flight
- drive cross country on paved roads with a car
- drive cross country on dirt roads with a car
- ride cross country on dirt roads with a motorcycle
- ride cross country with a bicycle
- take a ship through the panama canal
- walk cross country
- take a bus
- take a train
- hitchhike
- combine any of the following methods
|
|
1908.06606
| false
| null |
To implement deep neural network models, we utilize the Keras library BIBREF36 with TensorFlow BIBREF37 backend. Each model is run on a single NVIDIA GeForce GTX 1080 Ti GPU. The models are trained by Adam optimization algorithm BIBREF38 whose parameters are the same as the default settings except for learning rate set to $5\times 10^{-5}$. Batch size is set to 3 or 4 due to the lack of graphical memory. We select BERT-base as the pre-trained language model in this paper. Due to the high cost of pre-training BERT language model, we directly adopt parameters pre-trained by Google in Chinese general corpus. The named entity recognition is applied on both pathology report texts and query texts.
Due to the high cost of pre-training BERT language model, we directly adopt parameters pre-trained by Google in Chinese general corpus. The named entity recognition is applied on both pathology report texts and query texts.
|
What data is the language model pretrained on?
|
The answers are shown as follows:
* Chinese general corpus
|
1902.10525
| false
| null |
We present an extensive comparison of the differences in recognition accuracy for eight languages (Sec. SECREF5 ) and compare the accuracy of models trained on publicly available datasets where available (Sec. SECREF4 ). In addition, we propose a new standard experimental protocol for the IBM-UB-1 dataset BIBREF25 (Sec. SECREF50 ) to enable easier comparison between approaches in the future.
The IAM-OnDB dataset BIBREF42 is probably the most used evaluation dataset for online handwriting recognition. It consists of 298 523 characters in 86 272 word instances from a dictionary of 11 059 words written by 221 writers. We use the standard IAM-OnDB dataset separation: one training set, two validations sets and a test set containing 5 363, 1 438, 1 518 and 3 859 written lines, respectively. We tune the decoder weights using the validation set with 1 438 items and report error rates on the test set.
We provide an evaluation of our production system trained on our in-house datasets applied to a number of publicly available benchmark datasets from the literature. Note that for all experiments presented in this section we evaluate our current live system without any tuning specifec to the tasks at hand.
The ICDAR-2013 Competition for Online Handwriting Chinese Character Recognition BIBREF45 introduced a dataset for classifying the most common Chinese characters. We report the error rates in comparison to published results from the competition and more recent work done by others in Table TABREF56 .
In the ICFHR2018 Competition on Vietnamese Online Handwritten Text Recognition using VNOnDB BIBREF50 , our production system was evaluated against other systems. The system used in the competition is the one reported and described in this paper. Due to licensing restrictions we were unable to do any experiments on the competition training data, or specific tuning for the competition, which was not the case for the other systems mentioned here.
In addition, we propose a new standard experimental protocol for the IBM-UB-1 dataset BIBREF25 (Sec. SECREF50 ) to enable easier comparison between approaches in the future.
The IAM-OnDB dataset BIBREF42 is probably the most used evaluation dataset for online handwriting recognition. It consists of 298 523 characters in 86 272 word instances from a dictionary of 11 059 words written by 221 writers. We use the standard IAM-OnDB dataset separation: one training set, two validations sets and a test set containing 5 363, 1 438, 1 518 and 3 859 written lines, respectively. We tune the decoder weights using the validation set with 1 438 items and report error rates on the test set.
We provide an evaluation of our production system trained on our in-house datasets applied to a number of publicly available benchmark datasets from the literature. Note that for all experiments presented in this section we evaluate our current live system without any tuning specifec to the tasks at hand.
The ICDAR-2013 Competition for Online Handwriting Chinese Character Recognition BIBREF45 introduced a dataset for classifying the most common Chinese characters. We report the error rates in comparison to published results from the competition and more recent work done by others in Table TABREF56 .
In the ICFHR2018 Competition on Vietnamese Online Handwritten Text Recognition using VNOnDB BIBREF50 , our production system was evaluated against other systems. The system used in the competition is the one reported and described in this paper. Due to licensing restrictions we were unable to do any experiments on the competition training data, or specific tuning for the competition, which was not the case for the other systems mentioned here.
|
What datasets did they use?
|
The answers are shown as follows:
* IBM-UB-1 dataset BIBREF25
* IAM-OnDB dataset BIBREF42
* The ICDAR-2013 Competition for Online Handwriting Chinese Character Recognition BIBREF45
* ICFHR2018 Competition on Vietnamese Online Handwritten Text Recognition using VNOnDB BIBREF50
|
null | false
| null |
What's the difference between facts and truth?
|
Facts are pieces of information that are verifiable and based on empirical evidence whereas truth, although a related concept is related to a subjective perception or belief of what is a fact and does not always require evidence. An example would be a question about whether aliens exists. Many have a personal belief that aliens do exists, but since no empirical evidence exists upon which society can reach a consensus that they do truly exist. Someone may say that they are speaking truth when they say that aliens exists, however this is more likely just based on their belief which is unsupported by evidence (assuming in this example that the person making this statement has no empirical evidence). In this example someone who says they believe aliens do exist, believes they are speaking the truth but that does not make it a fact.
|
|
null | false
| null |
Tell me whether each of these wines is red or white: Merlot, Pinot Noir, Pinot Gris, Malbec, Syrah, Riesling, Chardonnay
|
Merlot: Red
Pinot Noir: Red
Pinot Gris: White
Malbec: Red
Syrah: Red
Riesling: White
Chardonnay: White
|
|
null | false
| 247
|
We conduct experiments on a simplified Chinese emotion cause corpus BIBREF31 , the only publicly available dataset on this task to the best of our knowledge. The corpus contains 2,105 documents from SINA city news. Each document has only one emotion word and one or more emotion causes. The documents are segmented into clauses manually. The main task is to identify which clause contains the emotion cause.
[id=lq]Details of the corpus are shown in Table 1. The metrics we used in evaluation follows lee2010text. It is commonly accepted so that we can compare our results with others. If a proposed emotion cause clause covers the annotated answer, the word sequence is considered correct. The precision, recall, and F-measure are defined by INLINEFORM0
In the experiments, we randomly select 90% of the dataset as training data and 10% as testing data. In order to obtain statistically credible results, we evaluate our method and baseline methods 25 times with different train/test splits.
We conduct experiments on a simplified Chinese emotion cause corpus (Gui et al., 2016) ∗ , the only publicly available dataset on this task to the best of our knowledge.
|
What corpus is used to conduct experiments?
|
A simplified Chinese emotion cause corpus.
|
1908.08593
| false
| null |
We use the following subset of GLUE tasks BIBREF4 for fine-tuning:
MRPC: the Microsoft Research Paraphrase Corpus BIBREF13
STS-B: the Semantic Textual Similarity Benchmark BIBREF14
SST-2: the Stanford Sentiment Treebank, two-way classification BIBREF15
QQP: the Quora Question Pairs dataset
RTE: the Recognizing Textual Entailment datasets
QNLI: Question-answering NLI based on the Stanford Question Answering Dataset BIBREF3
MNLI: the Multi-Genre Natural Language Inference Corpus, matched section BIBREF16
We use the following subset of GLUE tasks BIBREF4 for fine-tuning:
MRPC: the Microsoft Research Paraphrase Corpus BIBREF13
STS-B: the Semantic Textual Similarity Benchmark BIBREF14
SST-2: the Stanford Sentiment Treebank, two-way classification BIBREF15
QQP: the Quora Question Pairs dataset
RTE: the Recognizing Textual Entailment datasets
QNLI: Question-answering NLI based on the Stanford Question Answering Dataset BIBREF3
MNLI: the Multi-Genre Natural Language Inference Corpus, matched section BIBREF16
|
What subset of GLUE tasks is used?
|
The answers are shown as follows:
* MRPC
* STS-B
* SST-2
* QQP
* RTE
* QNLI
* MNLI
|
null | false
| null |
Tell me whether these Pokemon are fairy or poison type: Clefable, Jigglypuff, Rapidash, Arbok, Nidorino, Zubat
|
Clefable: Fairy
Jigglypuff: Fairy
Rapidash: Fairy
Arbok: Poison
Nidorino: Poison
Zubat: Poison
|
|
null | false
| 66
|
The convolutional language model (LM) is the GCNN-14B from BIBREF0 , which achieved competitive results on several language modeling benchmarks. The network contains 14 convolutional residual blocks BIBREF21 with a growing number of channels, and uses gated linear units as activation function.
The language model is used to score candidate transcriptions in addition to the acoustic model in the beam search decoder described in the next section. Compared to n-gram LMs, convolutional LMs allow for much larger context sizes. Our detailed experiments study the effect of context size on the final speech recognition performance.
Compared to n-gram LMs, convolutional LMs allow the decoder to look at longer context with better perplexity.
|
Why are convolutional LMs preferred than n-gram LMs？
|
Because convolutional LMs allow the decoder to look at longer context with better perplexity.
|
null | false
| null |
Apartment buildings are multi-story buildings where three or more residences are contained within one structure. Such a building may be called an apartment building, apartment complex, flat complex, block of flats, tower block, high-rise or, occasionally, mansion block (in British English), especially if it consists of many apartments for rent. A high-rise apartment building is commonly referred to as a residential tower, apartment tower, or block of flats in Australia.
|
What is an apartment building?
|
An apartment building is a multi-story building with at least three residences. Each residential unit within the building has many different names depending on the country or region; flat is commonly used in British English, while the term apartment is most popular in North America.
|
null | false
| null |
Resettlement Administration
Lange's iconic 1936 photograph of Florence Owens Thompson, Migrant Mother
"Broke, baby sick, and car trouble!" (1937)
Lange and Dixon divorced on October 28, 1935 and on December 6 she married economist Paul Schuster Taylor, professor of economics at the University of California, Berkeley. For the next five years they traveled through the California coast and the midwest. Throughout their travels they documented rural poverty, in particular the exploitation of sharecroppers and migrant laborers. Taylor interviewed subjects and gathered economic data while Lange produced photographs and accompanying data. They lived and worked from Berkeley for the rest of her life.
Working for the Resettlement Administration and Farm Security Administration, Lange's images brought to public attention the plight of the poor and forgotten—particularly sharecroppers, displaced farm families, and migrant workers. Lange's work was distributed to newspapers across the country, and the poignant images became icons of the era.
One of Lange's most recognized works is Migrant Mother, published in 1936. The woman in the photograph is Florence Owens Thompson. In 1960, Lange spoke about her experience taking the photograph:
"I saw and approached the hungry and desperate mother, as if drawn by a magnet. I do not remember how I explained my presence or my camera to her, but I do remember she asked me no questions. I made five exposures, working closer and closer from the same direction. I did not ask her name or her history. She told me her age, that she was thirty-two. She said that they had been living on frozen vegetables from the surrounding fields, and birds that the children killed. She had just sold the tires from her car to buy food. There she sat in that lean-to tent with her children huddled around her, and seemed to know that my pictures might help her, and so she helped me. There was a sort of equality about it."
Lange reported the conditions at the camp to the editor of a San Francisco newspaper, showing him her photographs. The editor informed federal authorities and published an article that included some of the images. In response, the government rushed aid to the camp to prevent starvation.
According to Thompson's son, while Lange got some details of the story wrong, the impact of the photograph came from an image that projected both the strengths and needs of migrant workers. Twenty-two of Lange's photographs produced for the FSA were included in John Steinbeck's The Harvest Gypsies when it was first published in 1936 in The San Francisco News. According to an essay by photographer Martha Rosler, Migrant Mother became the most reproduced photograph in the world.
|
What is the most known picture from Dorothea Lange
|
The most famous picture taken by Dorothea Lange is the "migrant mother" where a women is shown with her child during the great depression. The image symbolises the suffering of the american population at the time following years of bad crop that led to the economic crisis of 1929 and mid 30's depression.
|
null | false
| 22
|
To the best of our knowledge, we are not aware of any knowledge learning system that can learn new knowledge in the conversation process. This section thus discusses other related work.
Among existing KB completion approaches, BIBREF20 extended the vector space model for zero-shot KB inference. However, the model cannot handle unknown entities and can only work on fixed set of unknown relations with known embeddings. Recently, BIBREF22 proposed a method using external text corpus to perform inference on unknown entities. However, the method cannot handle unknown relations. Thus, these methods are not suitable for our open-world setting. None of the existing KB inference methods perform interactive knowledge learning like LiLi. NELL BIBREF23 continuously updates its KB using facts extracted from the Web. Our task is very different as we do not do Web fact extraction (which is also useful). We focus on user interactions in this paper. Our work is related to interactive language learning (ILL) BIBREF24 , BIBREF25 , but these are not about KB completion. The work in BIBREF26 allows a learner to ask questions in dialogue. However, this work used RL to learn about whether to ask the user or not. The “what to ask aspect" was manually designed by modeling synthetic tasks. LiLi formulates query-specific inference strategies which embed interaction behaviors. Also, no existing dialogue systems BIBREF4 , BIBREF27 , BIBREF28 , BIBREF29 , BIBREF30 employ lifelong learning to train prediction models by using information/knowledge retained in the past.
Our work is related to general lifelong learning in BIBREF31 , BIBREF32 , BIBREF33 , BIBREF34 , BIBREF35 , BIBREF36 . However, they learn only one type of tasks, e.g., supervised, topic modeling or reinforcement learning (RL) tasks. None of them is suitable for our setting, which involves interleaving of RL, supervised and interactive learning. More details about lifelong learning can be found in the book BIBREF31 .
Among existing KB completion approaches, (Neelakantan et al., 2015) extended the vector space model for zero-shot KB inference. However, the model cannot handle unknown entities and can only work on fixed set of unknown relations with known embeddings.
|
what is the shortcoming of the model of Neelakantan et al., 2015?
|
It cannot handle unknown entities and can only work on fixed set of unknown relations with known embeddings.
|
null | false
| 58
|
[0]leftmargin=* [0]leftmargin=*
Automatic systems have had a significant and beneficial impact on all walks of human life. So much so that it is easy to overlook their potential to benefit society by promoting equity, diversity, and fairness. For example, machines do not take bribes to do their jobs, they can determine eligibility for a loan without being influenced by the color of the applicant's skin, and they can provide access to information and services without discrimination based on gender or sexual orientation. Nonetheless, as machine learning systems become more human-like in their predictions, they can also perpetuate human biases. Some learned biases may be beneficial for the downstream application (e.g., learning that humans often use some insect names, such as spider or cockroach, to refer to unpleasant situations). Other biases can be inappropriate and result in negative experiences for some groups of people. Examples include, loan eligibility and crime recidivism prediction systems that negatively assess people belonging to a certain pin/zip code (which may disproportionately impact people of a certain race) BIBREF0 and resumé sorting systems that believe that men are more qualified to be programmers than women BIBREF1 . Similarly, sentiment and emotion analysis systems can also perpetuate and accentuate inappropriate human biases, e.g., systems that consider utterances from one race or gender to be less positive simply because of their race or gender, or customer support systems that prioritize a call from an angry male over a call from the equally angry female.
Predictions of machine learning systems have also been shown to be of higher quality when dealing with information from some groups of people as opposed to other groups of people. For example, in the area of computer vision, gender classification systems perform particularly poorly for darker skinned females BIBREF2 . Natural language processing (NLP) systems have been shown to be poor in understanding text produced by people belonging to certain races BIBREF3 , BIBREF4 . For NLP systems, the sources of the bias often include the training data, other corpora, lexicons, and word embeddings that the machine learning algorithm may leverage to build its prediction model.
Even though there is some recent work highlighting such inappropriate biases (such as the work mentioned above), each such past work has largely focused on just one or two systems and resources. Further, there is no benchmark dataset for examining inappropriate biases in natural language systems. In this paper, we describe how we compiled a dataset of 8,640 English sentences carefully chosen to tease out biases towards certain races and genders. We will refer to it as the Equity Evaluation Corpus (EEC). We used the EEC as a supplementary test set in a recent shared task on predicting sentiment and emotion intensity in tweets, SemEval-2018 Task 1: Affect in Tweets BIBREF5 . In particular, we wanted to test a hypothesis that a system should equally rate the intensity of the emotion expressed by two sentences that differ only in the gender/race of a person mentioned. Note that here the term system refers to the combination of a machine learning architecture trained on a labeled dataset, and possibly using additional language resources. The bias can originate from any or several of these parts. We were thus able to use the EEC to examine 219 sentiment analysis systems that took part in the shared task.
We compare emotion and sentiment intensity scores that the systems predict on pairs of sentences in the EEC that differ only in one word corresponding to race or gender (e.g., `This man made me feel angry' vs. `This woman made me feel angry'). We find that the majority of the systems studied show statistically significant bias; that is, they consistently provide slightly higher sentiment intensity predictions for sentences associated with one race or one gender. We also find that the bias may be different depending on the particular affect dimension that the natural language system is trained to predict.
Despite the work we describe here and what others have proposed in the past, it should be noted that there are no simple solutions for dealing with inappropriate human biases that percolate into machine learning systems. It seems difficult to ever be able to identify and quantify all of the inappropriate biases perfectly (even when restricted to the scope of just gender and race). Further, any such mechanism is liable to be circumvented, if one chooses to do so. Nonetheless, as developers of sentiment analysis systems, and NLP systems more broadly, we cannot absolve ourselves of the ethical implications of the systems we build. Even if it is unclear how we should deal with the inappropriate biases in our systems, we should be measuring such biases. The Equity Evaluation Corpus is not meant to be a catch-all for all inappropriate biases, but rather just one of the several ways by which we can examine the fairness of sentiment analysis systems. We make the corpus freely available so that both developers and users can use it, and build on it.
For NLP systems, the sources of the bias often include the training data, other corpora, lexicons, and word embeddings that the machine learning algorithm may leverage to build its prediction model.
|
What are the sources of NLP system deviation?
|
For NLP systems, the sources of the bias often include the training data, other corpora, lexicons, and word embeddings that the machine learning algorithm may leverage to build its prediction model.
|
null | false
| null |
What is the difference between a day on Earth and Mars?
|
A day on mars is called a Sol whose name is borrowed from the Latin word for Sun. Compared to a day on Earth, a sol is slightly longer at 24 hours, 39 minutes, and 35 seconds which is equivalent to 1.0275 Earth days. Spacecraft that have landed on Mars conventionally use a 24 hour Mars clock where the hours, minutes, and seconds are 2.75% longer than the standard Earth durations.
|
|
2004.01694
| false
| null |
The human evaluation of MT output in research scenarios is typically conducted by crowd workers in order to minimise costs. BIBREF13 shows that aggregated assessments of bilingual crowd workers are very similar to those of MT developers, and BIBREF14, based on experiments with data from WMT 2012, similarly conclude that with proper quality control, MT systems can be evaluated by crowd workers. BIBREF3 also use bilingual crowd workers, but the studies supporting the use of crowdsourcing for MT evaluation were performed with older MT systems, and their findings may not carry over to the evaluation of contemporary higher-quality neural machine translation (NMT) systems. In addition, the MT developers to which crowd workers were compared are usually not professional translators. We hypothesise that expert translators will provide more nuanced ratings than non-experts, and that their ratings will show a higher difference between MT outputs and human translations.
MT has been evaluated almost exclusively at the sentence level, owing to the fact that most MT systems do not yet take context across sentence boundaries into account. However, when machine translations are compared to those of professional translators, the omission of linguistic context—e. g., by random ordering of the sentences to be evaluated—does not do justice to humans who, in contrast to most MT systems, can and do take inter-sentential context into account BIBREF15, BIBREF16. We hypothesise that an evaluation of sentences in isolation, as applied by BIBREF3, precludes raters from detecting translation errors that become apparent only when inter-sentential context is available, and that they will judge MT quality less favourably when evaluating full documents.
The human reference translations with which machine translations are compared within the scope of a human–machine parity assessment play an important role. BIBREF3 used all source texts of the WMT 2017 Chinese–English test set for their experiments, of which only half were originally written in Chinese; the other half were translated from English into Chinese. Since translated texts are usually simpler than their original counterparts BIBREF17, they should be easier to translate for MT systems. Moreover, different human translations of the same source text sometimes show considerable differences in quality, and a comparison with an MT system only makes sense if the human reference translations are of high quality. BIBREF3, for example, had the WMT source texts re-translated as they were not convinced of the quality of the human translations in the test set. At WMT 2018, the organisers themselves noted that the manual evaluation included several reports of ill-formed reference translations BIBREF5. We hypothesise that the quality of the human translations has a significant effect on findings of human–machine parity, which would indicate that it is necessary to ensure that human translations used to assess parity claims need to be carefully vetted for their quality.
BIBREF3 also use bilingual crowd workers, but the studies supporting the use of crowdsourcing for MT evaluation were performed with older MT systems, and their findings may not carry over to the evaluation of contemporary higher-quality neural machine translation (NMT) systems. In addition, the MT developers to which crowd workers were compared are usually not professional translators.
We hypothesise that an evaluation of sentences in isolation, as applied by BIBREF3, precludes raters from detecting translation errors that become apparent only when inter-sentential context is available, and that they will judge MT quality less favourably when evaluating full documents.
BIBREF3 used all source texts of the WMT 2017 Chinese–English test set for their experiments, of which only half were originally written in Chinese;
|
What was the weakness in Hassan et al's evaluation design?
|
MT developers to which crowd workers were compared are usually not professional translators, evaluation of sentences in isolation prevents raters from detecting translation errors, used not originally written Chinese test set
|
null | false
| null |
Waddell's career wound through a number of teams. He was notably unpredictable; early in his career, he once left in the middle of a game to go fishing. He also had a longstanding fascination with fire trucks and ran off the field to chase after them during games on multiple occasions. He would disappear for months at a time during the offseason, and it was not known where he went until it was discovered that he was wrestling alligators in a circus. He was easily distracted by opposing fans who held up puppies, which caused him to run over to play with them, and shiny objects, which seemed to put him in a trance. An alcoholic for much of his short life, he reportedly spent his entire first signing bonus on a drinking binge; as a pun of the baseball term "southpaw" denoting a left-handed pitcher, the Sporting News dubbed him a "sousepaw". His eccentric behavior led to constant battles with his managers and scuffles with bad-tempered teammates.
|
Extract which animals Waddell played with and which ones he fought with from the text. Separate them with a comma, putting the ones he played with first.
|
puppies, alligators
|
null | false
| null |
Who created the first chocolate bar?
|
Joseph Fry created the first chocolate bar in 1847. Soon after he created the Fry's Chocolate Cream bar in 1866. The first chocolate bar was made of bittersweet chocolate.
|
|
null | false
| null |
Vallabhbhai Jhaverbhai Patel (Gujarati: [ʋəlːəbːʰɑi dʒʰəʋeɾbʰɑi pəʈel]; pə-TEL; 31 October 1875 – 15 December 1950), commonly known as Sardar Patel, was an Indian lawyer, influential political leader, barrister and statesman who served as the first Deputy Prime Minister and Home Minister of India from 1947 to 1950. He was a barrister and a senior leader of the Indian National Congress, who played a leading role in the country's struggle for independence, guiding its integration into a united, independent nation. In India and elsewhere, he was often called Sardar, meaning "chief" in Hindi, Urdu, Bengali and Persian. He acted as the Home Minister during the political integration of India and the Indo-Pakistani War of 1947.
Patel was born in Nadiad, Kheda district, and raised in the countryside of the state of Gujarat. He was a successful lawyer. One of Mahatma Gandhi's earliest political lieutenants, he organised peasants from Kheda, Borsad, and Bardoli in Gujarat in non-violent civil disobedience against the British Raj, becoming one of the most influential leaders in Gujarat. He was appointed as the 49th President of Indian National Congress, organising the party for elections in 1934 and 1937 while promoting the Quit India Movement.
As the first Home Minister and Deputy Prime Minister of India, Patel organised relief efforts for partition refugees fleeing to Punjab and Delhi from Pakistan and worked to restore peace. He led the task of forging a united India, successfully integrating into the newly independent nation those British colonial provinces that formed the Dominion of India. Besides those provinces that had been under direct British rule, approximately 565 self-governing princely states had been released from British suzerainty by the Indian Independence Act of 1947. Patel persuaded almost every princely state to accede to India. His commitment to national integration in the newly independent country was total and uncompromising, earning him the sobriquet "Iron Man of India". He is also remembered as the "patron saint of India's civil servants" for having established the modern All India Services system.
The Statue of Unity, the world's tallest statue which was erected by the Indian government at a cost of US$420 million, was dedicated to him on 31 October 2018 and is approximately 182 metres (597 ft) in height.
|
Who is called Iron Man of India and why?
|
Sardar Vallabhbhai Jhaverbhai Patel is known as Iron Man of India. Born in 31 October 1875 in Gujarat, Patel was an Indian lawyer, influential political leader, barrister and statesman who served as the first Deputy Prime Minister and Home Minister of India from 1947 to 1950. After the Indian independence as the first Home Minister and Deputy Prime Minister of India, Patel organized relief efforts for partition refugees fleeing to Punjab and Delhi from Pakistan and worked to restore peace. He led the task of creating a united India, successfully integrating into the newly independent nation those British colonial provinces that formed the Dominion of India. Approximately 565 self-governing princely states had been released from British suzerainty by the Indian Independence Act of 1947 and became part of the Indian Union. Patel persuaded almost every princely state to accede to India. His commitment to national integration in the newly independent country was total and uncompromising, earning him the sobriquet "Iron Man of India"
|
1710.02772
| false
| null |
Using the fine-grained gating mechanism conditioned on the lexical features, we can accurately control the information flows between word-level and char-level. Intuitively, the formulation is as follows: INLINEFORM0
where INLINEFORM0 is the element-wise multiplication operator. when the gate has high value, more information flows from the word-level representation; otherwise, char-level will take the dominating place. It is practical in real scenarios. For example, for unfamiliar noun entities, the gates tend to bias towards char-level representation in order to care richer morphological structure. Besides, we not only utilize the lexical properties as the gating feature, we also concatenate them as a supplement of lexical information. Therefore, the final representation of words are computed as follows: INLINEFORM1
Using the fine-grained gating mechanism conditioned on the lexical features, we can accurately control the information flows between word-level and char-level. Intuitively, the formulation is as follows: INLINEFORM0
where INLINEFORM0 is the element-wise multiplication operator. when the gate has high value, more information flows from the word-level representation; otherwise, char-level will take the dominating place. It is practical in real scenarios. For example, for unfamiliar noun entities, the gates tend to bias towards char-level representation in order to care richer morphological structure.
|
How does the gatint mechanism combine word and character information?
|
The answers are shown as follows:
* when the gate has high value, more information flows from the word-level representation; otherwise, char-level will take the dominating place
* for unfamiliar noun entities, the gates tend to bias towards char-level representation in order to care richer morphological structure
|
null | false
| null |
Albert Einstein (14 March 1879 – 18 April 1955) was a German-born theoretical physicist, widely acknowledged to be one of the greatest and most influential physicists of all time. Einstein is best known for developing the theory of relativity, but he also made important contributions to the development of the theory of quantum mechanics. Relativity and quantum mechanics are the two pillars of modern physics. His mass–energy equivalence formula E = mc2, which arises from relativity theory, has been dubbed "the world's most famous equation". His work is also known for its influence on the philosophy of science. He received the 1921 Nobel Prize in Physics "for his services to theoretical physics, and especially for his discovery of the law of the photoelectric effect", a pivotal step in the development of quantum theory. His intellectual achievements and originality resulted in "Einstein" becoming synonymous with "genius". Einsteinium, one of the synthetic elements in the periodic table, was named in his honor.
|
Provide three bullets about Einstein from this paragraph that does not relate to physics
|
- Einstein was born in Germany
- Einstein lived to 76 years old
- One part of Einstein’s legacy was that his name became synonymous with “genius”
|
null | false
| null |
Peloton Interactive, Inc. is an American exercise equipment and media company based in New York City. The company's products are stationary bicycles, treadmills, and indoor rowers equipped with Internet-connected touch screens that stream live and on-demand fitness classes through a subscription service. The equipment includes built-in sensors that track metrics such as power output, providing users with real-time feedback on their performance and leaderboard rankings to compete with other users.
|
Where is Peloton Interactive based?
|
Peloton Interactive is based in New York City, NY, US. They are an American exercise and media company. Their products include stationary spin bicycles, treadmills, and rowers. Their products feature internet-connected touch screens to stream live and on-demand fitness classes. Sensors in the equipment track metrics such as power, cadence, and resistance to provide real-time feedback to the user. Users must subscribe to Peloton Interactive services to use classes on the equipment.
|
null | false
| null |
When was the NFL founded? Which NFL team won the most Super Bowls? Which quarterback has won the most Super Bowls?
|
The NFL was officially founded in 1920.
The New England Patriots have won 6 titles.
Tom Brady has won 7 superbowls
|
|
null | false
| null |
Who gave the slogan of Quit India and Simon Go Back?
|
Yusuf Meherally
|
|
null | false
| null |
Tell me whether you eat each of these with a spoon or fork: soup, ice cream, cereal, salad, pasta
|
Soup: Spoon
Ice Cream: Spoon
Cereal: Spoon
Salad: Fork
Pasta: Fork
|
|
null | false
| null |
Norway, officially the Kingdom of Norway, is a Nordic country in Northern Europe, the mainland territory of which comprises the western and northernmost portion of the Scandinavian Peninsula. The remote Arctic island of Jan Mayen and the archipelago of Svalbard also form part of Norway.[note 5] Bouvet Island, located in the Subantarctic, is a dependency of Norway; it also lays claims to the Antarctic territories of Peter I Island and Queen Maud Land. The capital and largest city in Norway is Oslo.
Norway has a total area of 385,207 square kilometres (148,729 sq mi) and had a population of 5,488,984 in January 2023. The country shares a long eastern border with Sweden at a length of 1,619 km (1,006 mi). It is bordered by Finland and Russia to the northeast and the Skagerrak strait to the south, on the other side of which are Denmark and the United Kingdom. Norway has an extensive coastline, facing the North Atlantic Ocean and the Barents Sea. The maritime influence dominates Norway's climate, with mild lowland temperatures on the sea coasts; the interior, while colder, is also significantly milder than areas elsewhere in the world on such northerly latitudes. Even during polar night in the north, temperatures above freezing are commonplace on the coastline. The maritime influence brings high rainfall and snowfall to some areas of the country.
Harald V of the House of Glücksburg is the current King of Norway. Jonas Gahr Støre has been prime minister since 2021, replacing Erna Solberg. As a unitary sovereign state with a constitutional monarchy, Norway divides state power between the parliament, the cabinet and the supreme court, as determined by the 1814 constitution. The kingdom was established in 872 as a merger of many petty kingdoms and has existed continuously for 1,151 years. From 1537 to 1814, Norway was a part of the Kingdom of Denmark–Norway, and, from 1814 to 1905, it was in a personal union with the Kingdom of Sweden. Norway was neutral during the First World War, and also in World War II until April 1940 when the country was invaded and occupied by Nazi Germany until the end of the war.
Norway has both administrative and political subdivisions on two levels: counties and municipalities. The Sámi people have a certain amount of self-determination and influence over traditional territories through the Sámi Parliament and the Finnmark Act. Norway maintains close ties with both the European Union and the United States. Norway is also a founding member of the United Nations, NATO, the European Free Trade Association, the Council of Europe, the Antarctic Treaty, and the Nordic Council; a member of the European Economic Area, the WTO, and the OECD; and a part of the Schengen Area. In addition, the Norwegian languages share mutual intelligibility with Danish and Swedish.
Norway maintains the Nordic welfare model with universal health care and a comprehensive social security system, and its values are rooted in egalitarian ideals. The Norwegian state has large ownership positions in key industrial sectors, having extensive reserves of petroleum, natural gas, minerals, lumber, seafood, and fresh water. The petroleum industry accounts for around a quarter of the country's gross domestic product (GDP). On a per-capita basis, Norway is the world's largest producer of oil and natural gas outside of the Middle East.
The country has the fourth-highest per-capita income in the world on the World Bank and IMF lists.It has the world's largest sovereign wealth fund, with a value of US$1 trillion. Although the majority of Norway's population is ethnic Norwegian, in the 21st century immigration has accounted for more than half of population growth; in 2021, the five largest minority groups in the country were the descendants of Polish, Lithuanian, Somali, Pakistani, and Swedish immigrants.
|
Please give me a short bulleted list of the most important facts about the country Norway
|
- Population: 5.488 million
- Size: 148729 sq mi
- GDP: 425 billion USD
- Neighbouring countries: Sweden, Finland, Russia
- Unitary parliamentary constitutional monarchy
- Fourth highest per-capita income in the world
- Norway is the world's largest producer of oil and natural gas outside of the Middle East.
- Norway maintains the Nordic welfare model with universal health care and a comprehensive social security system
|
1603.01417
| false
| null |
We have proposed new modules for the DMN framework to achieve strong results without supervision of supporting facts. These improvements include the input fusion layer to allow interactions between input facts and a novel attention based GRU that allows for logical reasoning over ordered inputs. Our resulting model obtains state of the art results on both the VQA dataset and the bAbI-10k text question-answering dataset, proving the framework can be generalized across input domains.
the input fusion layer to allow interactions between input facts and a novel attention based GRU that allows for logical reasoning over ordered inputs.
|
How does the model circumvent the lack of supporting facts during training?
|
The answers are shown as follows:
* the input fusion layer to allow interactions between input facts and a novel attention based GRU that allows for logical reasoning over ordered inputs.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.