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 |
List a few interesting ways to make a plant based Alfredo sauce.
|
Some of the keys ways that can vegan-ize an Alfredo sauce are:
1. Use cauliflower: Blanching and making a purée of a cauliflower can make a silky smooth white sauce that can easily replace the daily based Alfredo.
2. Use Tofu: silken tofu can be blended to make a white Alfredo sauce.
3. Use Cashews: overnight soaked cashews can be blended to replace the dairy Alfredo with a plant based Alfredo
4. Use plant based milk: almond milk can replace dairy milk in the recipe of a traditional Alfredo sauce.
|
|
null | false
| null |
Who wrote The Originals?
|
Adam Grant
|
|
1912.13072
| false
| null |
Arab-Tweet. For modeling age and gender, we use Arap-Tweet BIBREF19 , which we will refer to as Arab-Tweet. Arab-tweet is a tweet dataset of 11 Arabic regions from 17 different countries. For each region, data from 100 Twitter users were crawled. Users needed to have posted at least 2,000 and were selected based on an initial list of seed words characteristic of each region. The seed list included words such as <برشة> /barsha/ ‘many’ for Tunisian Arabic and <وايد> /wayed/ ‘many’ for Gulf Arabic. BIBREF19 employed human annotators to verify that users do belong to each respective region. Annotators also assigned gender labels from the set male, female and age group labels from the set under-25, 25-to34, above-35 at the user-level, which in turn is assigned at tweet level. Tweets with less than 3 words and re-tweets were removed. Refer to BIBREF19 for details about how annotation was carried out. We provide a description of the data in Table TABREF10. Table TABREF10 also provides class breakdown across our splits.We note that BIBREF19 do not report classification models exploiting the data.
UBC Twitter Gender Dataset. We also develop an in-house Twitter dataset for gender. We manually labeled 1,989 users from each of the 21 Arab countries. The data had 1,246 “male", 528 “female", and 215 unknown users. We remove the “unknown" category and balance the dataset to have 528 from each of the two `male" and “female" categories. We ended with 69,509 tweets for `male" and 67,511 tweets for “female". We split the users into 80% TRAIN set (110,750 tweets for 845 users), 10% DEV set (14,158 tweets for 106 users), and 10% TEST set (12,112 tweets for 105 users). We, then, model this dataset with BERT-Base, Multilingual Cased model and evaluate on development and test sets. Table TABREF15 shows that fine-tuned model obtains 62.42% acc on DEV and 60.54% acc on TEST.
The dialect identification model in AraNet is based on our winning system in the MADAR shared task 2 BIBREF20 as described in BIBREF12. The corpus is divided into train, dev and test, and the organizers masked test set labels. We lost some tweets from training data when we crawled using tweet ids, ultimately acquiring 2,036 (TRAIN-A), 281 (DEV) and 466 (TEST). We also make use of the task 1 corpus (95,000 sentences BIBREF21). More specifically, we concatenate the task 1 data to the training data of task 2, to create TRAIN-B. Again, note that TEST labels were only released to participants after the official task evaluation. Table TABREF17 shows statistics of the data. We used tweets from 21 Arab countries as distributed by task organizers, except that we lost some tweets when we crawled using tweet ids. We had 2,036 (TRAIN-A), 281 (DEV) and 466 (TEST). For our experiments, we also make use of the task 1 corpus (95,000 sentences BIBREF21). More specifically, we concatenate the task 1 data to the training data of task 2, to create TRAIN-B. Note that both DEV and TEST across our experiments are exclusively the data released in task 2, as described above. TEST labels were only released to participants after the official task evaluation. Table TABREF17 shows statistics of the data. More information about the data is in BIBREF21. We use TRAIN-A to perform supervised modeling with BERT and TRAIN-B for self training, under various conditions. We refer the reader to BIBREF12 for more information about our different experimental settings on dialect id. We acquire our best results with self-training, with a classification accuracy of 49.39% and F1 score at 35.44. This is the winning system model in the MADAR shared task and we showed in BIBREF12 that our tweet-level predictions can be ported to user-level prediction. On user-level detection, our models perform superbly, with 77.40% acc and 71.70% F1 score on unseen MADAR blind test data.
We make use of two datasets, the LAMA-DINA dataset from BIBREF22, a Twitter dataset with a combination of gold labels from BIBREF23 and distant supervision labels. The tweets are labeled with the Plutchik 8 primary emotions from the set: {anger, anticipation, disgust, fear, joy, sadness, surprise, trust}. The distant supervision approach depends on use of seed phrases with the Arabic first person pronoun انا> (Eng. “I") + a seed word expressing an emotion, e.g., فرحان> (Eng. “happy"). The manually labeled part of the data comprises tweets carrying the seed phrases verified by human annotators $9,064$ tweets for inclusion of the respective emotion. The rest of the dataset is only labeled using distant supervision (LAMA-DIST) ($182,605$ tweets) . For more information about the dataset, readers are referred to BIBREF22. The data distribution over the emotion classes is in Table TABREF20. We combine LAMA+DINA and LAMA-DIST training set and refer to this new training set as LAMA-D2 (189,903 tweets). We fine-tune BERT-Based, Multilingual Cased on the LAMA-D2 and evaluate the model with same DEV and TEST sets from LAMA+DINA. On DEV set, the fine-tuned BERT model obtains 61.43% on accuracy and 58.83 on $F_1$ score. On TEST set, we acquire 62.38% acc and 60.32% $F_1$ score.
We use the dataset for irony identification on Arabic tweets released by IDAT@FIRE2019 shared-task BIBREF24. The shared task dataset contains 5,030 tweets related to different political issues and events in the Middle East taking place between 2011 and 2018. Tweets are collected using pre-defined keywords (i.e., targeted political figures or events) and the positive class involves ironic hashtags such as #sokhria, #tahakoum, and #maskhara (Arabic variants for “irony"). Duplicates, retweets, and non-intelligible tweets are removed by organizers. Tweets involve both MSA as well as dialects at various degrees of granularity such as Egyptian, Gulf, and Levantine.
We collect 15 datasets related to sentiment analysis of Arabic, including MSA and dialects BIBREF25, BIBREF26, BIBREF27, BIBREF1, BIBREF28, BIBREF29, BIBREF30, BIBREF31, BIBREF32, BIBREF33, BIBREF34. Table TABREF28 shows all the corpora we use. These datasets involve different types of sentiment analysis tasks such as binary classification (i.e., negative or positive), 3-way classification (i.e., negative, neutral, or positive), and subjective language detection. To combine these datasets for binary sentiment classification, we normalize different types of label to binary labels in the set $\lbrace `positive^{\prime }, `negative^{\prime }\rbrace $ by following rules:
Arab-Tweet. For modeling age and gender, we use Arap-Tweet BIBREF19 , which we will refer to as Arab-Tweet.
UBC Twitter Gender Dataset. We also develop an in-house Twitter dataset for gender. We manually labeled 1,989 users from each of the 21 Arab countries.
The dialect identification model in AraNet is based on our winning system in the MADAR shared task 2 BIBREF20 as described in BIBREF12. The corpus is divided into train, dev and test, and the organizers masked test set labels.
We make use of two datasets, the LAMA-DINA dataset from BIBREF22, a Twitter dataset with a combination of gold labels from BIBREF23 and distant supervision labels.
The rest of the dataset is only labeled using distant supervision (LAMA-DIST) ($182,605$ tweets) . For more information about the dataset, readers are referred to BIBREF22.
We use the dataset for irony identification on Arabic tweets released by IDAT@FIRE2019 shared-task BIBREF24.
We collect 15 datasets related to sentiment analysis of Arabic, including MSA and dialects BIBREF25, BIBREF26, BIBREF27, BIBREF1, BIBREF28, BIBREF29, BIBREF30, BIBREF31, BIBREF32, BIBREF33, BIBREF34. Table TABREF28 shows all the corpora we use.
|
What datasets are used in training?
|
The answers are shown as follows:
* Arap-Tweet BIBREF19
* an in-house Twitter dataset for gender
* the MADAR shared task 2 BIBREF20
* the LAMA-DINA dataset from BIBREF22
* LAMA-DIST
* Arabic tweets released by IDAT@FIRE2019 shared-task BIBREF24
* BIBREF25, BIBREF26, BIBREF27, BIBREF1, BIBREF28, BIBREF29, BIBREF30, BIBREF31, BIBREF32, BIBREF33, BIBREF34
|
null | false
| null |
Who wrote the most episodes of season four of Game of Thrones?
|
Co-writers David Benioff and D.B. Weiss wrote seven of the ten episodes of Game of Thrones's fourth season.
|
|
null | false
| null |
In the United States, the Recording Industry Association of America (RIAA) awards certification based on the number of albums and singles sold through retail and other ancillary markets. Other countries have similar awards (see music recording certification). Certification is not automatic; for an award to be made, the record label must first request certification. The audit is conducted against net shipments after returns (most often an artist's royalty statement is used), which includes albums sold directly to retailers and one-stops, direct-to-consumer sales (music clubs and mail order) and other outlets.
Description and qualifications
A platinum award for the album GHV2 of Madonna
A Gold record is a song or album that sells 500,000 units (records, tapes, and compact discs). The award was launched in 1958; originally, the requirement for a Gold single was one million units sold and a Gold album represented $1 million in sales (at wholesale value, around a third of the list price). In 1975, the additional requirement of 500,000 units sold was added for Gold albums. Reflecting growth in record sales, the Platinum award was added in 1976, for albums able to sell one million units, and singles selling two million units. The Multi-Platinum award was introduced in 1984, signifying multiple Platinum levels of albums and singles. In 1989, the sales thresholds for singles were reduced to 500,000 for Gold and 1,000,000 for Platinum, reflecting a decrease in sales of singles. In 1992, RIAA began counting each disc in a multi-disc set as one unit toward certification. Reflecting additional growth in music sales, the Diamond award was instituted in 1999 for albums or singles selling ten million units. Because of these changes in criteria, the sales level associated with a particular award depends on when the award was made.
|
Given a reference text about the Recording Industry Association of America (RIAA) awards, tell me how many albums need to be sold to get an award.
|
As of 1989, a song or album needed to exceed a sales threshold of 500,000 for Gold and 1,000,000 for Platinum. In 1999 after a growth in music sales, the Diamond award was introduced for songs or albums that sold over 10,000,000 units.
|
null | false
| null |
The game's English name derives from the Balti language, from its word for 'ball', polo. It is cognate with the Standard Tibetan pulu, also meaning 'ball'.
Although the exact origins of the game are not certain, many scholars suggest it most likely began as a simple game played by Iranic and Turkic equestrian nomads in Central Asia. An archaic variation of polo, regionally referred to as buzkashi or kokpar, is still played in parts of Central Asia. It was developed and formalised in Ancient Iran (Persia) as "chovgan" (čowgān), becoming a national sport played extensively by the nobility. Women played as well as men. During the period of the Parthian Empire (247 BC to 224 AD), the sport had great patronage under the kings and noblemen. According to The Oxford Dictionary of Late Antiquity, the Persian ball game was an important pastime in the court of the Sasanian Empire (224–651). It was also part of the royal education for the Sasanian ruling class. Emperor Shapur II learnt to play polo at age seven in 316 AD.
|
During which year Emperor Shapur II learned to play polo?
|
316 AD
|
null | false
| null |
What are 3 to 5 creative ways to prevent deer from walking through your yard?
|
Deer can be a menace, but here are 5 ways to keep them out of your yard:
1. Install motion sensors that trigger sprinklers when activated
2. Completely enclose your property with a solid fence; nothing cheap that will just be pushed over
3. Get a guard dog, and train it to bark at intruding deer
4. Plant only deer-resistant shrubs and flowers; the deer won't stop by if they have nothing to eat
5. Move to a location without deer!
|
|
null | false
| null |
Identify which instrument is string or percussion: Maram, Phin
|
Phin is string, Maram is percussion.
|
|
null | false
| null |
Give me a list of 5 key challenges faced by users moving from a Windows laptop to a Macbook
|
1. Incompatibility between MS office documents and MacOS native applications like pages and numbers
2. Inability to copy content on and off of the laptop easily
3. Learning the differences between the windows keyboard and the macbook keyboard commands to do simple tasks - like right click or taking a screenshot
4. The change in process to deploy software.
5. Navigating the difference between the layout and structure of WIndows explorer vs Macbook finder
|
|
null | false
| null |
When giving examples of what is a verb and what is a noun, explain which of these words falls into which category. Slowly, wipe, tidily, fully, jump, yell, sleep, talk, hopelessly.
|
Wipe, jump, yell, sleep and talk are all examples of verbs. Slowly, tidily, fully and hopelessly are adverbs.
|
|
null | false
| null |
When was the last financial crisis and what caused it?
|
The last financial crisis was from 2007-2008. It was dubbed the Global Financial Crisis (GFC) because it had severe worldwide economic impact. The Global Financial Crisis was caused by the predatory lending of low-income homebuyers, excessive risk-taking of Wall Street and the bursting of the United States hosing bubble.
|
|
null | false
| null |
Make me a list of all the different ways I could tell my sister about a recent vacation
|
You could tell her about your vacation by writing a letter, mailing a postcard, sending an email, giving her a call, sending a text or visiting her house
|
|
null | false
| null |
What is the list of ingredients for the Lebanese tabbouleh recipe?
|
- Parsley: finely chopped
- Tomatoes: cut in small cubes
- Bulgur wheat: soaked in room temperature water
- White and Green Onions: finely chopped
- Mint: finely chopped
- Lemon: squeezed
- Olive oil
- Salt
- Pepper
|
|
1810.09774
| false
| null |
Also including a pretrained ELMo language model did not improve the results significantly. The overall performance of BERT was significantly better than the other models, having the lowest average difference in accuracy of 22.5 points. Our baselines for SNLI (90.4%) and SNLI + MultiNLI (90.6%) outperform the previous state-of-the-art accuracy for SNLI (90.1%) by BIBREF24 .
The overall performance of BERT was significantly better than the other models, having the lowest average difference in accuracy of 22.5 points.
|
Which model generalized the best?
|
The answers are shown as follows:
* BERT
|
null | false
| null |
The Ansonia (formerly the Ansonia Hotel) is a condominium building at 2109 Broadway, between 73rd and 74th Streets, on the Upper West Side of Manhattan in New York City. The 17-story structure was designed by French architect Paul Emile Duboy in the Beaux-Arts style. It was built between 1899 and 1903 as a residential hotel by William Earle Dodge Stokes, who named it after his grandfather, the industrialist Anson Greene Phelps. Over the years, the Ansonia has housed many conductors, opera singers, baseball players, and other famous and wealthy people. The Ansonia is a New York City designated landmark and is listed on the National Register of Historic Places.
|
Given this paragraph about a hotel, how many stories is the Ansonia?
|
17 stories
|
null | false
| null |
classify each of the following as either a city or a county: usa, mexico, cananda, syracuse, austin, mumbai, japan, tokyo
|
Of the list your presented, following are countries: usa, mexico, canada and japan. The following are cities: syracuse, austin, mumbai, tokyo.
|
|
null | false
| 385
|
Neural sequence models BIBREF0, BIBREF1 typically generate outputs in an autoregressive left-to-right manner. These models have been successfully applied to a range of task, for example machine translation BIBREF2. They often rely on an encoder that processes the source sequence, and a decoder that generates the output sequence conditioned on the output of the encoder. The decoder will typically generate the target sequence one token at a time, in an autoregressive left-to-right fashion.
Recently, research in insertion-based non- or partially- autoregressive models has spiked BIBREF3, BIBREF4, BIBREF5, BIBREF6. These model are more flexible than their autoregressive counterparts. They can generate sequences in any order, and can benefit from parallel token generation. They can learn complex orderings (e.g., tree orderings) and may be more applicable to task like cloze question answering BIBREF6 and text simplification, where the order of generation is not naturally left to right, and the source sequence might not be fully observed. One recently proposed approach is the Insertion Transformer BIBREF3, where the target sequence is modelled with insertion-edits. As opposed to traditional sequence-to-sequence models, the Insertion Transformer can generate sequences in any arbitrary order, where left-to-right is a special case. Additionally, during inference, the model is endowed with parallel token generation capabilities. The Insertion Transformer can be trained to follow a soft balanced binary tree order, thus allowing the model to generate $n$ tokens in $O(\log _2 n)$ iterations.
In this work we propose to generalize this insertion-based framework, we present a framework which emits both insertions and deletions. Our Insertion-Deletion Transformer consists of an insertion phase and a deletion phase that are executed iteratively. The insertion phase follows the typical insertion-based framework BIBREF3. However, in the deletion phase, we teach the model to do deletions with on-policy training. We sample an input sequence on-policy from the insertion model (with on-policy insertion errors), and teach the deletion model its appropriate deletions.
This insertion-deletion framework allows for flexible sequence generation, parallel token generation and text editing. In a conventional insertion-based model, if the model makes a mistake during generation, this cannot be undone. Introducing the deletion phase makes it possible to undo the mistakes made by the insertion model, since it is trained on the on-policy errors of the insertion phase. The deletion model extension also enables the framework to efficiently handle tasks like text simplification and style transfer by starting the decoding process from the original source sequence.
A concurrent work was recently proposed, called the Levenshtein Transformer (LevT) BIBREF7. The LevT framework also generates sequences with insertion and deletion operations. Our approach has some important distinctions and can be seen as a simplified version, for both the architecture and the training algorithm. The training algorithm used in the LevT framework uses an expert policy. This expert policy requires dynamic programming to minimize Levenshtein distance between the current input and the target. This approach was also explored by BIBREF8, BIBREF9. Their learning algorithm arguably adds more complexity than needed over the simple on-policy method we propose. The LevT framework consists of three stages, first the number of tokens to be inserted is predicted, then the actual tokens are predicted, and finally the deletion actions are emitted. The extra classifier to predict the number of tokens needed to be inserted adds an additional Transformer pass to each generation step. In practice, it is also unclear whether the LevT exhibits speedups over an insertion-based model following a balanced binary tree order. In contrast, our Insertion-Deletion framework only has one insertion phase and one deletion phase, without the need to predict the number of tokens needed to be inserted. This greatly simplifies the model architecture, training procedure and inference runtime.
An alternative approach for text editing is proposed by BIBREF10, which they dub Deliberation Networks. This work also acknowledges the potential benefits from post-editing output sequences and proposes a two-phase decoding framework to facilitate this.
In this paper, we present the insertion-deletion framework as a proof of concept by applying it to two synthetic character-based translation tasks and showing it can significantly increase the BLEU score over the insertion-only framework.
We sample an input sequence on-policy from the insertion model (with on-policy insertion errors), and teach the deletion model its appropriate deletions.
|
What does the author sample from the insertion model?
|
An input sequence on-policy.
|
null | false
| null |
Twitch is an American video live streaming service that focuses on video game live streaming, including broadcasts of esports competitions, in addition to offering music broadcasts, creative content, and "in real life" streams. Twitch is operated by Twitch Interactive, a subsidiary of Amazon.com, Inc. It was introduced in June 2011 as a spin-off of the general-interest streaming platform Justin.tv. Content on the site can be viewed either live or via video on demand. The games shown on Twitch's current homepage are listed according to audience preference and include genres such as real-time strategy games (RTS), fighting games, racing games, and first-person shooters.
The popularity of Twitch eclipsed that of its general-interest counterpart. In October 2013, the website had 45 million unique viewers, and by February 2014, it was considered the fourth-largest source of peak Internet traffic in the United States. At the same time, Justin.tv's parent company was re-branded as Twitch Interactive to represent the shift in focus – Justin.tv was shut down in August 2014. That month, the service was acquired by Amazon for US$970 million, which later led to the introduction of synergies with the company's subscription service Amazon Prime. Twitch acquired Curse LLC in 2016, an operator of online video gaming communities and introduced means to purchase games through links on streams along with a program allowing streamers to receive commissions on the sales of games that they play.
By 2015, Twitch had more than 100 million viewers per month. In 2017, Twitch remained the leading live streaming video service for video games in the US, and had an advantage over YouTube Gaming, which shut down its standalone app in May 2019. As of February 2020, it had 3 million broadcasters monthly and 15 million daily active users, with 1.4 million average concurrent users. As of May 2018, Twitch had over 27,000 partner channels.
|
What is Twitch Interactive?
|
Twitch is a video live streaming service that gained popularity by focusing on video game live streaming broadcasts.
|
null | false
| null |
What is a good score to shoot in golf?
|
A good score in golf highly depends on the skill of the player. A professional can easily shoot under par by many strokes. Conversely, an amateur player can shoot anywhere from a few shots over par to over 30 shots over par. A player can also use tools that calculate their handicap, which will show them on average how much they shoot over or under par. This is a good tool for an individual to track a good or bad score during their round.
|
|
null | false
| 210
|
BioASQ is a semantic indexing, question answering (QA) and information extraction challenge BIBREF0 . We participated in Task B of the challenge which is concerned with biomedical QA. More specifically, our system participated in Task B, Phase B: Given a question and gold-standard snippets (i.e., pieces of text that contain the answer(s) to the question), the system is asked to return a list of answer candidates.
The fifth BioASQ challenge is taking place at the time of writing. Five batches of 100 questions each were released every two weeks. Participating systems have 24 hours to submit their results. At the time of writing, all batches had been released.
The questions are categorized into different question types: factoid, list, summary and yes/no. Our work concentrates on answering factoid and list questions. For factoid questions, the system's responses are interpreted as a ranked list of answer candidates. They are evaluated using mean-reciprocal rank (MRR). For list questions, the system's responses are interpreted as a set of answers to the list question. Precision and recall are computed by comparing the given answers to the gold-standard answers. F1 score, i.e., the harmonic mean of precision and recall, is used as the official evaluation measure .
Most existing biomedical QA systems employ a traditional QA pipeline, similar in structure to the baseline system by weissenborn2013answering. They consist of several discrete steps, e.g., named-entity recognition, question classification, and candidate answer scoring. These systems require a large amount of resources and feature engineering that is specific to the biomedical domain. For example, OAQA BIBREF1 , which has been very successful in last year's challenge, uses a biomedical parser, entity tagger and a thesaurus to retrieve synonyms.
Our system, on the other hand, is based on a neural network QA architecture that is trained end-to-end on the target task. We build upon FastQA BIBREF2 , an extractive factoid QA system which achieves state-of-the-art results on QA benchmarks that provide large amounts of training data. For example, SQuAD BIBREF3 provides a dataset of $\approx 100,000$ questions on Wikipedia articles. Our approach is to train FastQA (with some extensions) on the SQuAD dataset and then fine-tune the model parameters on the BioASQ training set.
Note that by using an extractive QA network as our central component, we restrict our system's responses to substrings in the provided snippets. This also implies that the network will not be able to answer yes/no questions. We do, however, generalize the FastQA output layer in order to be able to answer list questions in addition to factoid questions.
BioASQ is a semantic indexing, question answering (QA) and information extraction challenge.
|
What is BioASQ?
|
A semantic indexing, question answering (QA) and information extraction challenge.
|
null | false
| null |
How soon should you make coffee after grinding the beans?
|
For best results, no matter the brewing method, you should make your coffee within twenty minutes of grinding the coffee beans.
|
|
null | false
| null |
What is the most dangerous motorcycle race in the world?
|
The Isle of Man TT is most likely the deadliest motorcycle race in the world. The race takes place once a year on a small island between Ireland and the United Kingdom called Isle of Man. The very first race was held in 1907 and since then 265 riders have lost their lives.
|
|
null | false
| null |
Who is the winningest college women's basketball coach?
|
Geno Auriemma has the most wins in NCAA women's basketball history for a head coach. He has 1180 career wins, all with the University of Connecticut.
|
|
null | false
| 42
|
Neural machine translation (NMT) emerged in the last few years as a very successful paradigm BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 . While NMT is generally more fluent than previous statistical systems, adequacy is still a major concern BIBREF4 : common mistakes include dropping source words and repeating words in the generated translation.
Previous work has attempted to mitigate this problem in various ways. BIBREF5 incorporate coverage and length penalties during beam search—a simple yet limited solution, since it only affects the scores of translation hypotheses that are already in the beam. Other approaches involve architectural changes: providing coverage vectors to track the attention history BIBREF6 , BIBREF7 , using gating architectures and adaptive attention to control the amount of source context provided BIBREF8 , BIBREF9 , or adding a reconstruction loss BIBREF10 . BIBREF11 also use the notion of fertility implicitly in their proposed model. Their “fertility conditioned decoder” uses a coverage vector and an “extract gate” which are incorporated in the decoding recurrent unit, increasing the number of parameters.
In this paper, we propose a different solution that does not change the overall architecture, but only the attention transformation. Namely, we replace the traditional softmax by other recently proposed transformations that either promote attention sparsity BIBREF12 or upper bound the amount of attention a word can receive BIBREF13 . The bounds are determined by the fertility values of the source words. While these transformations have given encouraging results in various NLP problems, they have never been applied to NMT, to the best of our knowledge. Furthermore, we combine these two ideas and propose a novel attention transformation, constrained sparsemax, which produces both sparse and bounded attention weights, yielding a compact and interpretable set of alignments. While being in-between soft and hard alignments (Figure FIGREF20 ), the constrained sparsemax transformation is end-to-end differentiable, hence amenable for training with gradient backpropagation.
To sum up, our contributions are as follows:
In this paper, we propose a different solution that does not change the overall architecture, but only the attention transformation.
|
How does this approach differ from previous ones？
|
Their solution does not change the overall architecture, but only the attention transformation.
|
null | false
| null |
The game involves using figures mounted on rotating bars to kick a ball into the opposing goal. Table football tables can vary in size, but a typical table is about 120 centimetres (3.9 ft) long and 61 centimetres (2.00 ft) wide. The table usually contains eight rows of foos men, which are plastic, metal, wooden, or sometimes carbon-fibre figures mounted on horizontal metal bars. Each team of one or two human players controls four rows of foos men, one row each for the goalkeeper, defenders, midfield and strikers. Players manipulate the rods to control the figures, using them to hold up, pass or 'kick' the ball. Games begin when the ball is served through a hole at the side of the table, or simply placed by hand at the feet of a figure in the centre of the table. A coin toss is usually used to determine which player or team serves first. If it is not the first match or if the ball gets out of play or if the ball suddenly stops out, then the team scored last would get the serve after he scores one. The term for when the ball leaves the table or gets stopped somewhere out of the men-like figures’ reach is dead ball. Expert players have been known to move balls at speeds up to 56 km/h (35 mph) in competition.
The rules prohibit "over 360-degree shots", or "spinning": using the palm of the hand to swiftly spin the bar all around, instead of using wrist strokes to kick the ball with a bar-mounted figure. Since the establishment of the International Table Soccer Federation (ITSF), the rules have become standardised in most international competitions. However, since January 2012, the annual World Championships and the World Cup have permitted two full 360-degree rotations.
The winner is determined when one team scores a predetermined number of goals, typically five, ten or eleven in competition. When playing Bonzini competitions, the target number of goals is seven and players must win by at least two clear goals.
|
per this description of the game of fooseball, how many full rotations are allowed in the game at the championship level?
|
2 360-degree rotations are allowed since January 2012.
|
null | false
| null |
Name the place where Queen Elizabeth II is buried
|
Windsor Castle
|
|
null | false
| null |
Where and when did Van Halen start as a band?
|
Van Halen (/væn ˈheɪlɛn/ van HAY-len) was an American rock band formed in Pasadena, California, in 1973. Credited with "restoring hard rock to the forefront of the music scene", Van Halen was known for its energetic live shows and for the virtuosity of its lead guitarist, Eddie Van Halen. The band was inducted into the Rock and Roll Hall of Fame in 2007.
|
formed in Pasadena, California, in 1973
|
null | false
| null |
PayPal was originally established by Max Levchin, Peter Thiel, and Luke Nosek in December 1998 as Confinity, a company that developed security software for hand-held devices. Having had no success with that business model, however, it switched its focus to a digital wallet. The first version of the PayPal electronic payments system was launched in 1999.
In March 2000, Confinity merged with x.com, an online financial services company founded in March 1999 by Elon Musk, Harris Fricker, Christopher Payne, and Ed Ho. Musk was optimistic about the future success of the money transfer business Confinity was developing. Musk and Bill Harris, then-president and CEO of X.com, disagreed about the potential future success of the money transfer business and Harris left the company in May 2000. In October of that year, Musk decided that X.com would terminate its other internet banking operations and focus on payments. That same month, Elon Musk was replaced by Peter Thiel as CEO of X.com, which was renamed PayPal in June 2001 and went public in 2002. PayPal's IPO listed under the ticker PYPL at $13 per share and generated over $61 million.
|
From the passage identify the founders of Confinity. Display the results in comma separated format.
|
Max Levchin, Peter Thiel, Luke Nosek
|
null | false
| null |
Is it safe to invest in stocks?
|
There are no guarantees of profits when you buy stock, which makes stock a risky investments. If a company doesn't do well or falls out of favor with investors, its stock can fall in price, and investors could lose money. With the other hand, Stocks offer investors the greatest potential for growth over the long haul. Investors willing to stick with stocks over long periods of time, say 15 years, generally have been rewarded with strong, positive returns.
|
|
null | false
| 506
|
We analyse the generalization error of CGPFL in this section. Before starting the analysis, we first introduce two important definitions as follows.
Definition 2 (Complexity) Let H be a hypothesis class (correspanding to ω ∈ R d in neural network), and |D| be the size of dataset D, the complexity of H can be expressed by the maximum disagreement between two hypotheses on a dataset D:
Definition 3 (Label-discrepancy) Consider a hypothesis class H, the label-discrepancy between two data distributions D 1 and D 2 is given by: disc
where
Theorem 2 (The generalization error of CGPFL) When Assumption 1 is satisfied, with probability at least 1 − δ, the following holds:
where B is a positive constant with
Remark 2 Theorem 2 gives the generalization error bound of CGPFL. When K = 1, it yields the error bound of PFL with single global model. As the number of clusters increases, the second terms become larger, while the last term get smaller. Hence, our CGPFL can alwalys reach better personalizationgeneralization trade-off by adjusting the number of clusters K, and further achieve higher accuracy than the existing PFL methods. The detailed proof of generalization error is given in the Appendix of this paper.
(Complexity) Let H be a hypothesis class (correspanding to ω ∈ Rd in neural network), and |D| be the size of dataset D, the complexity of H can be expressed by the maximum disagreement between two hypotheses on a dataset D
|
Notation changes completely in section 4.3. What do the statements mean? The notation starts with classic FL notation but then switches to the notation used in [MMRS20].
|
Actually, we explain that the hypothesis h has a direct relationship with the model parameter Ω in Definition 2 in Section 4.3.
|
null | false
| 405
|
To evaluate the effectiveness and interpretability of the proposed model, we conduct a series of experiments, including link prediction and community detection, on several real-world graphs.
Ablation study for the components of our method, including the community membership, node random factors and the HLSM decoder architecture are also conducted.
On all datasets, our model achieves better or comparable performances to the non-generative method SEAL. However, since it requires to materialize a subgraph for each link, SEAL is much more unscalable than our DLSM and takes about 20 multiples longer for running one epoch.
|
How is the proposed method comparing with state-of-the-art non-VAE-based GNN methods?
|
We consider your suggestion and have added SEAL [1] in the link prediction experiments. The results are presented in Table 1, 2 and 6 of our current draft. It seems that SEAL performs close to our proposed method on some datasets. However, compared to our method, SEAL is more unscalable with high time complexity as it requires to materialize a subgraph for each link. (Actually, it takes about 20 multiples longer than our method for running one epoch.)
More importantly, SEAL and other non-generative models for link prediction cannot generate representations as node embeddings for other downstream tasks such as community detection and node classification. There are some gaps between these methods and ours (as well as the baselines selected in our original draft). The main point and contribution of our work is to propose a novel graph representation learning method and generate interpretable latent variables to better represent the common properties of directed graphs, including link reciprocity, degree heterogeneity and community structure. These learned node representations are readily to be applied for multiple downstream tasks, including but not limited to link prediction and community detection. The experiments in our paper are conducted for the purpose of testing whether these learned representations are able to capture the interpreted graph properties. For example, the link prediction task verifies the ability to represent link reciprocity of directed graphs, and the community detection task verifies the ability for modeling clustering effects.
We sincerely hope the reviewer to recosider on this point since our work is not acturally aiming for link prediction. Besides, by comparing our method with other graph representation learning methods, we conclude that our method achieves the state-of-the-art performances on downstream tasks.
|
null | false
| null |
What are monsoons in India?
|
Monsoons in India typically refer to the rainy season, which is generally between June and September. It follows the summer season. Monsoons also refer to the seasonal weather pattern that causes rains in India due to the changing direction of winds. The hot summers cause the air over the land to heat up and rise causing low pressure, which the moisture-filled air from the ocean rushes to fill. As this air rises over the mountains, it cools and brings rain to many parts of the Indian subcontinent.
|
|
null | false
| 86
|
Accurate language identification (LID) is the first step in many natural language processing and machine comprehension pipelines. If the language of a piece of text is known then the appropriate downstream models like parts of speech taggers and language models can be applied as required.
LID is further also an important step in harvesting scarce language resources. Harvested data can be used to bootstrap more accurate LID models and in doing so continually improve the quality of the harvested data. Availability of data is still one of the big roadblocks for applying data driven approaches like supervised machine learning in developing countries.
Having 11 official languages of South Africa has lead to initiatives (discussed in the next section) that have had positive effect on the availability of language resources for research. However, many of the South African languages are still under resourced from the point of view of building data driven models for machine comprehension and process automation.
Table TABREF2 shows the percentages of first language speakers for each of the official languages of South Africa. These are four conjunctively written Nguni languages (zul, xho, nbl, ssw), Afrikaans (afr) and English (eng), three disjunctively written Sotho languages (nso, sot, tsn), as well as tshiVenda (ven) and Xitsonga (tso). The Nguni languages are similar to each other and harder to distinguish. The same is true of the Sotho languages.
This paper presents a hierarchical naive Bayesian and lexicon based classifier for LID of short pieces of text of 15-20 characters long. The algorithm is evaluated against recent approaches using existing test sets from previous works on South African languages as well as the Discriminating between Similar Languages (DSL) 2015 and 2017 shared tasks.
Section SECREF2 reviews existing works on the topic and summarises the remaining research problems. Section SECREF3 of the paper discusses the proposed algorithm and Section SECREF4 presents comparative results.
The algorithm is evaluated against recent approaches using existing test sets from previous works on South African languages as well as the Discriminating between Similar Languages (DSL) 2015 and 2017 shared tasks.
|
What is the algorithm evaluated against?
|
Recent approaches using existing test sets from previous works on South African languages as well as the Discriminating between Similar Languages (DSL) 2015 and 2017 shared tasks.
|
null | false
| 43
|
GluonCV/NLP provide access to modular APIs to allow users to customize their model design, training, and inference by re-using efficient components across different models. Such common components include (but are not limited to) data processing utilities, models with individual components, initialization methods, and loss functions.
To elucidate how the modular API facilitates efficient implementation, let us take the data API of GluonCV/NLP as an example, which is used to build efficient data pipelines with popular benchmark data sets or those supplied by users. In computer vision and natural language processing tasks, inputs or labels often come in with different shapes, such as images with a varying number of objects and sentences of different lengths. Thus, the data API provides a collection of utilities to sample inputs or labels then transform them into mini-batches to be efficiently computed. Besides, users can access a wide range of popular data sets via the data API, including (but are not limited to) ImageNet of image classification, VOC of object detection, COCO of instance segmentation, SQuAD of question answering, and SST of sentiment analysis.
To elucidate how the modular API facilitates efficient implementation, let us take the data API of GluonCV/NLP as an example, which is used to build efficient data pipelines with popular benchmark data sets or those supplied by users.
|
What do they use the data API of GluonCV/NLP to do?
|
To build efficient data pipelines with popular benchmark data sets or those supplied by users.
|
null | false
| null |
The Tampa Bay Buccaneers are a professional American football team based in Tampa, Florida. The Buccaneers compete in the National Football League (NFL) as a member club of the league's National Football Conference (NFC) South division. The club joined the NFL in 1976 as an expansion team, along with the Seattle Seahawks, and played its first season in the American Football Conference (AFC) West division. Prior to the 1977 season, Tampa Bay switched conferences and divisions with Seattle, becoming a member of the NFC Central division. As a result of the league's realignment prior to the 2002 season, the Buccaneers joined three former NFC West teams to form the NFC South. The club is owned by the Glazer family and plays its home games at Raymond James Stadium in Tampa.
The Buccaneers have won two Super Bowl championships and, along with the Baltimore Ravens, are the only two NFL franchises who are undefeated in multiple Super Bowl appearances. They were regarded as a perennial losing franchise for most of their first two decades due to suffering 26 consecutive losses in their first two seasons (including a winless inaugural season) and 14 consecutive losing seasons from 1983 to 1996. Despite these early struggles, Tampa Bay is the first post-merger expansion team to clinch a division title, win a playoff game, and host a conference championship; feats they accomplished by their fourth season in 1979. The team's image improved by the time of their first championship in 2002, also the first for a franchise built after the merger, but they would not win another playoff game until their second Super Bowl championship season in 2020. Concurrent to the lack of success outside of their Super Bowl wins, the Buccaneers hold the NFL's lowest winning percentage at .404.
|
Given a reference text about the Tampa Bay Buccaneers, how many Super Bowls have they won.
|
The Buccaneers have won two Super Bowl championships.
|
null | false
| 91
|
Neural machine translation (NMT) has achieved the state-of-the-art results on a mass of language pairs with varying structural differences, such as English-French BIBREF0, BIBREF1 and Chinese-English BIBREF2. However, so far not much is known about how and why NMT works, which pose great challenges for debugging NMT models and designing optimal architectures.
The understanding of NMT models has been approached primarily from two complementary perspectives. The first thread of work aims to understand the importance of representations by analyzing the linguistic information embedded in representation vectors BIBREF3, BIBREF4 or hidden units BIBREF5, BIBREF6. Another direction focuses on understanding the importance of input words by interpreting the input-output behavior of NMT models. Previous work BIBREF7 treats NMT models as black-boxes and provides explanations that closely resemble the attention scores in NMT models. However, recent studies reveal that attention does not provide meaningful explanations since the relationship between attention scores and model output is unclear BIBREF8.
In this paper, we focus on the second thread and try to open the black-box by exploiting the gradients in NMT generation, which aims to estimate the word importance better. Specifically, we employ the integrated gradients method BIBREF9 to attribute the output to the input words with the integration of first-order derivatives. We justify the gradient-based approach via quantitative comparison with black-box methods on a couple of perturbation operations, several language pairs, and two representative model architectures, demonstrating its superiority on estimating word importance.
We analyze the linguistic behaviors of words with the importance and show its potential to improve NMT models. First, we leverage the word importance to identify input words that are under-translated by NMT models. Experimental results show that the gradient-based approach outperforms both the best black-box method and other comparative methods. Second, we analyze the linguistic roles of identified important words, and find that words of certain syntactic categories have higher importance while the categories vary across language. For example, nouns are more important for Chinese$\Rightarrow $English translation, while prepositions are more important for English-French and -Japanese translation. This finding can inspire better design principles of NMT architectures for different language pairs. For instance, a better architecture for a given language pair should consider its own language characteristics.
We justify the gradient-based approach via quantitative comparison with black-box methods on a couple of perturbation operations, several language pairs, and two representative model architectures, demonstrating its superiority on estimating word importance.
|
What does the quantitative comparison demonstrate?
|
Demonstrate the gradient-based approach's superiority on estimating word importance.
|
null | false
| null |
I need to plan a wedding and I have no idea where to start. Will you help me brainstorm a few things I should do?
|
Your wedding a celebration of you and partners love. Here are a couple ideas of where to start.
1. Sit down with your partner and each share you respective vision, hopes and dreams for your wedding. Listen intently and take notes to what your partner says.
2. Make a list of the people you would want at your wedding. Tier each potential invite as Tier 1,2 and 3 and who's side they represent (you, your partner, or both).
3. Identify a preliminary budget based on what you can afford to spend.
4. Make a list of the most important things to you and your partner, in priority order (venue, band, flowers, dress, etc.)
5. Evaluate hiring a wedding planner to assist you in your planning efforts.
Good luck as you begin this exciting process!
|
|
null | false
| null |
What is Gettysburg College?
|
Gettysburg College is a private liberal arts college located in Gettysburg, Pennsylvania. It was founded in 1832 and is noted as being a key fixture in the Battle of Gettysburg fought during the Civil War.Gettysburg College has about 2,600 students and is 90 miles north of Washington D.C.
|
|
null | false
| 340
|
* Equal contribution. Listing order is random.
Lemmatization is a core NLP task that involves a string-to-string transduction from an inflected word form to its citation form, known as the lemma. More concretely, consider the English sentence: The bulls are running in Pamplona. A lemmatizer will seek to map each word to a form you may find in a dictionary—for instance, mapping running to run. This linguistic normalization is important in several downstream NLP applications, especially for highly inflected languages. Lemmatization has previously been shown to improve recall for information retrieval BIBREF0 , BIBREF1 , to aid machine translation BIBREF2 , BIBREF3 and is a core part of modern parsing systems BIBREF4 , BIBREF5 .
However, the task is quite nuanced as the proper choice of the lemma is context dependent. For instance, in the sentence A running of the bulls took place in Pamplona, the word running is its own lemma, since, here, running is a noun rather than an inflected verb. Several counter-examples exist to this trend, as discussed in depth in haspelmath2013understanding. Thus, a good lemmatizer must make use of some representation of each word's sentential context. The research question in this work is, then, how do we design a lemmatization model that best extracts the morpho-syntax from the sentential context?
Recent work BIBREF7 has presented a system that directly summarizes the sentential context using a recurrent neural network to decide how to lemmatize. As N18-1126's system currently achieves state-of-the-art results, it must implicitly learn a contextual representation that encodes the necessary morpho-syntax, as such knowledge is requisite for the task. We contend, however, that rather than expecting the network to implicitly learn some notion of morpho-syntax, it is better to explicitly train a joint model to morphologically disambiguate and lemmatize. Indeed, to this end, we introduce a joint model for the introduction of morphology into a neural lemmatizer. A key feature of our model is its simplicity: Our contribution is to show how to stitch existing models together into a joint model, explaining how to train and decode the model. However, despite the model's simplicity, it still achieves a significant improvement over the state of the art on our target task: lemmatization.
Experimentally, our contributions are threefold. First, we show that our joint model achieves state-of-the-art results, outperforming (on average) all competing approaches on a 20-language subset of the Universal Dependencies (UD) corpora BIBREF8 . Second, by providing the joint model with gold morphological tags, we demonstrate that we are far from achieving the upper bound on performance—improvements on morphological tagging could lead to substantially better lemmatization. Finally, we provide a detailed error analysis indicating when and why morphological analysis helps lemmatization. We offer two tangible recommendations: one is better off using a joint model (i) for languages with fewer training data available and (ii) languages that have richer morphology.
Our system and pre-trained models on all languages in the latest version of the UD corpora are released at https://sigmorphon.github.io/sharedtasks/2019/task2/.
We offer two tangible recommendations: one is better off using a joint model (i) for languages with fewer training data available and (ii) languages that have richer morphology.
|
What are the two tangible recommendations offered by the authors?
|
One is better off using a joint model (i) for languages with fewer training data available and (ii) languages that have richer morphology.
|
1911.03562
| false
| null |
FLOAT SELECTED: Figure 11 A treemap of the 122 languages arranged alphabetically and shaded such that languages that appear more often in AA paper titles have a darker shade of green.
FLOAT SELECTED: Figure 11 A treemap of the 122 languages arranged alphabetically and shaded such that languages that appear more often in AA paper titles have a darker shade of green.
|
Which 5 languages appear most frequently in AA paper titles?
|
English, Chinese, French, Japanese and Arabic
|
null | false
| null |
The Supreme Court of Iran is the highest juridical authority in Iran, established to supervise the correct implementation of laws by courts of justice and consisting of the most prominent judges of the country. The head of the judiciary assigns criteria to ensure uniformity of judicial procedure and to carry out all legal responsibilities.
|
From the given text, provide a couple of points on Supreme Court of Iran
|
1. The Supreme Court of Iran, which is composed of the most eminent judges in the nation and was founded to oversee the proper application of laws by courts of justice, is the highest legal authority in Iran.
2. To ensure judicial procedure is uniform and to fulfil all legal obligations, the head of the court assigns standards.
|
1910.03042
| false
| null |
We assessed the degree of conversational depth by measuring users' mean word count. Prior work has found that an increase in word count has been linked to improved user engagement (e.g., in a social dialog system BIBREF13). For each user conversation, we extracted the overall rating, the number of turns of the interaction, and the user's per-utterance word count (averaged across all utterances). We modeled the relationship between word count and the two metrics of user engagement (overall rating, mean number of turns) in separate linear regressions.
Results showed that users who, on average, produced utterances with more words gave significantly higher ratings ($\beta $=0.01, SE=0.002, t=4.79, p$<$0.001)(see Figure 2) and engaged with Gunrock for significantly greater number of turns ($\beta $=1.85, SE=0.05, t=35.58, p$<$0.001) (see Figure 2). These results can be interpreted as evidence for Gunrock's ability to handle complex sentences, where users are not constrained to simple responses to be understood and feel engaged in the conversation – and evidence that individuals are more satisfied with the conversation when they take a more active role, rather than the system dominating the dialog. On the other hand, another interpretation is that users who are more talkative may enjoy talking to the bot in general, and thus give higher ratings in tandem with higher average word counts.
We modeled the relationship between word count and the two metrics of user engagement (overall rating, mean number of turns) in separate linear regressions.
Results showed that users who, on average, produced utterances with more words gave significantly higher ratings ($\beta $=0.01, SE=0.002, t=4.79, p$<$0.001)(see Figure 2) and engaged with Gunrock for significantly greater number of turns ($\beta $=1.85, SE=0.05, t=35.58, p$<$0.001) (see Figure 2). These results can be interpreted as evidence for Gunrock's ability to handle complex sentences, where users are not constrained to simple responses to be understood and feel engaged in the conversation – and evidence that individuals are more satisfied with the conversation when they take a more active role, rather than the system dominating the dialog. On the other hand, another interpretation is that users who are more talkative may enjoy talking to the bot in general, and thus give higher ratings in tandem with higher average word counts.
|
How do they correlate user backstory queries to user satisfaction?
|
The answers are shown as follows:
* modeled the relationship between word count and the two metrics of user engagement (overall rating, mean number of turns) in separate linear regressions
|
null | false
| null |
Tell me which of these US Presidents was Republican or Democrat: Dwight Eisenhower, John Kennedy, Lyndon Johnson, Richard Nixon, Gerald Ford, Jimmy Carter
|
Dwight Eisenhower - Republican
John Kennedy - Democrat
Lyndon Johnson - Democrat
Richard Nixon - Republican
Gerald Ford - Republican
Jimmy Carter - Democrat
|
|
null | false
| null |
Which Italian tractor maker tried making cars in 1960s
|
Ferruchio Lamborghini
|
|
null | false
| null |
What is it like being alone in the world after an apocolypse?
|
I had always imagined the world would be in chaos or destruction. However, it's the same, just that there are no people. It's eerily quiet. Books are now the only source of "human connection" I have. I've never been much of a reader, but I anticipate much of my "free time" will be spent reading. I'm also quickly learning how much I relied on others. I knew I relied on others to do many things like farming, electricity, etc., but it goes much deeper and further than that.
|
|
null | false
| null |
Shane Fenton became famous as who
|
Alvin Stardust
|
|
null | false
| 109
|
We use the CNN structure as described in Section SECREF5 for emotional feature extraction. As a dataset for extracting emotion-related features, we use the corpus developed by BIBREF35 . This dataset consists of blog posts labeled by their corresponding emotion categories. As emotion taxonomy, the authors used six basic emotions, i.e., Anger, Disgust, Surprise, Sadness, Joy and Fear. In particular, the blog posts were split into sentences and each sentence was labeled. The dataset contains 5,205 sentences labeled by one of the emotion labels. After employing this model on the sarcasm dataset, we obtained a 150-dimensional feature vector from the fully-connected layer. As the aim of training is to classify each sentence into one of the six emotion classes, we used six neurons in the softmax layer.
The dataset contains 5,205 sentences labeled by one of the emotion labels.
|
How many sentences does the dataset use in the emotion feature extraction model?
|
5,205 sentences.
|
1904.04055
| false
| null |
The analysis of results from Tables TABREF17 , TABREF18 and TABREF19 show that 12 of 15 best results were obtained using new word embeddings. The evaluation results presented in Table TABREF20 (the chosen best embeddings models from Table TABREF19 ) prove that the best group of word embeddings is EC. The highest type F1-score was obtained for EC1 model, built using binary FastText Skip-gram method utilising subword information, with vector dimension equal to 300 and negative sampling equal to 10. The ability of the model to provide vector representation for the unknown words seems to be the most important. Also, previous models built using KGR10 (EP) are probably less accurate due to an incorrect tokenisation of the corpus. We used WCRFT tagger BIBREF22 , which utilises Toki BIBREF21 to tokenise the input text before the creation of the embeddings model. The comparison of EC1 with previous results obtained using only CRF BIBREF38 show the significant improvement across all the tested metrics: 3.6pp increase in strict F1-score, 1.36pp increase in relaxed precision, 5.61pp increase in relaxed recall and 3.51pp increase in relaxed F1-score.
The analysis of results from Tables TABREF17 , TABREF18 and TABREF19 show that 12 of 15 best results were obtained using new word embeddings. The evaluation results presented in Table TABREF20 (the chosen best embeddings models from Table TABREF19 ) prove that the best group of word embeddings is EC. The highest type F1-score was obtained for EC1 model, built using binary FastText Skip-gram method utilising subword information, with vector dimension equal to 300 and negative sampling equal to 10. The ability of the model to provide vector representation for the unknown words seems to be the most important.
|
What conclusions are drawn from these experiments?
|
The answers are shown as follows:
* best results were obtained using new word embeddings
* best group of word embeddings is EC
* The highest type F1-score was obtained for EC1 model, built using binary FastText Skip-gram method utilising subword information
* ability of the model to provide vector representation for the unknown words seems to be the most important
|
2002.05829
| false
| null |
FLOAT SELECTED: Table 1: Pretraining costs of baseline models. Hardware and pretraining time are collected from original papers, with which costs are estimated with current TPU price at $8 per hour with 4 core TPU v3 chips and V100 GPU at $3.06 per hour. DistilBERT model is trained upon a pretrained BERT model. Parameter numbers are estimated using the pretrained models implemented in the Transformers (https://github.com/huggingface/ transformers) library (Wolf et al., 2019), shown in million.
FLOAT SELECTED: Table 1: Pretraining costs of baseline models. Hardware and pretraining time are collected from original papers, with which costs are estimated with current TPU price at $8 per hour with 4 core TPU v3 chips and V100 GPU at $3.06 per hour. DistilBERT model is trained upon a pretrained BERT model. Parameter numbers are estimated using the pretrained models implemented in the Transformers (https://github.com/huggingface/ transformers) library (Wolf et al., 2019), shown in million.
|
What models are included in baseline benchmarking results?
|
BERT, XLNET RoBERTa, ALBERT, DistilBERT
|
null | false
| null |
Name some famous basketball players from the 1960's and 1970's
|
Oscar Robertson, George Mikan, John Havlicek, Bill Russell, Wilt Chamberlain, and Bill Bradley are all examples of outstanding basketball players from the early days of the National Basketball Association.
|
|
null | false
| null |
When is an object considered a planet?
|
To be considered a planet the object must have an orbit around a sun, have enough mass to form a mostly round shape, and must be the dominant gravitational force in the area with no other similarly sized objects that are not orbiting the object in question.
|
|
null | false
| null |
How do you make Crawfish bread?
|
Ingredients
1 pound crawfish tail meat, peeled, cleaned and deveined
1/2 stick butter
1 clove garlic, minced
3/4 cup chopped and drained ripe tomatoes
1 tablespoon Cajun seasoning
3/4 cup Monterey Jack Cheese
1/4 cup grated Parmesan
1/2 cup mayonnaise
3 tablespoons chopped fresh parsley
1 loaf French bread
Directions :
1. Preheat the oven to 350 degrees F.
2. Rinse the crawfish. In a large skillet, melt the butter over medium heat. Add the garlic and cook for a few minutes. Add the tomatoes, Cajun seasoning and crawfish. Saute for 1 minute. Remove from the heat.
3. In a bowl, mix together the cheeses, mayonnaise and parsley. Add the tomato and crawfish mixture to the mayo mixture.
4. Cut the loaf of bread in half and then cut the bread lengthwise. Spoon the mixture on the bread and bake for 10 to 15 minutes. Then broil for an additional 3 to 5 minutes. Watch closely, as not to burn.
5. Cut the bread into individual portions and serve.
|
|
null | false
| null |
What are some of the languages spoken in Spain?
|
Most languages spoken in Spain belong to the Romance language family, of which Spanish is the only language which has the "official" status for the whole country. There are other languages that have recognized status in specific territories such as: Basque in the Basque Country, Catalan spoken in Catalonia, Valencian in the Valencian community and Gallician the adjancent western parts of the principality of Asturias.
|
|
1706.07179
| false
| null |
We describe the RelNet model in this section. Figure 1 provides a high-level view of the model. The model is sequential in nature, consisting of the following steps: read text, process it into a dynamic relational memory and then attention conditioned on the question generates the answer. We model the dynamic memory in a fashion similar to Recurrent Entity Networks BIBREF17 and then equip it with an additional relational memory.
The model is sequential in nature, consisting of the following steps: read text, process it into a dynamic relational memory and then attention conditioned on the question generates the answer. We model the dynamic memory in a fashion similar to Recurrent Entity Networks BIBREF17 and then equip it with an additional relational memory.
|
What is the architecture of the neural network?
|
The answers are shown as follows:
* extends memory-augmented neural networks with a relational memory to reason about relationships between multiple entities present within the text.
* The model is sequential in nature, consisting of the following steps: read text, process it into a dynamic relational memory and then attention conditioned on the question generates the answer. We model the dynamic memory in a fashion similar to Recurrent Entity Networks BIBREF17 and then equip it with an additional relational memory.
|
null | false
| null |
What is an LCR circuit?
|
An LCR circuit is a resonant circuit that is also often called an RLC circuit. It consist of an inductor (L), a capacitor (C), and a resistor (R). The three components may be connected in series or in parallel.
Radio and television receivers contain oscillators that are based on the LCR circuit. The circuit is used to tune into transmitting stations.
|
|
null | false
| null |
What are questions you can ask to get to know someone?
|
Questions you can ask to get to know someone are the following:
1. What is your name?
2. What is your age?
3. What are your hobbies?
4. Did you go to college? What college?
5. What do you do for work?
6. Do you have any brothers or sister?
7. What is most important to you in a relationship?
8. Do you like to read? What was the last book you read?
9. Have you ever done any traveling? Where?
10. What is your favorite restaurant? Where?
|
|
null | false
| 123
|
Language models can be optimized to recognize syntax and semantics with great accuracy BIBREF0. However, the output generated can be repetitive and generic leading to monotonous or uninteresting responses (e.g “I don't know”) regardless of the input BIBREF1. While application of attention BIBREF2, BIBREF3 and advanced decoding mechanisms like beam search and variation sampling BIBREF4 have shown improvements, it does not solve the underlying problem. In creative text generation, the objective is not strongly bound to the ground truth—instead the objective is to generate diverse, unique or original samples. We attempt to do this through a discriminator which can give feedback to the generative model through a cost function that encourages sampling of creative tokens. The contributions of this paper are in the usage of a GAN framework to generate creative pieces of writing. Our experiments suggest that generative text models, while very good at encapsulating semantic, syntactic and domain information, perform better with external feedback from a discriminator for fine-tuning objectiveless decoding tasks like that of creative text. We show this by evaluating our model on three very different creative datasets containing poetry, metaphors and lyrics.
Previous work on handling the shortcomings of MLE include length-normalizing sentence probability BIBREF5, future cost estimation BIBREF6, diversity-boosting objective function BIBREF7, BIBREF1 or penalizing repeating tokens BIBREF8. When it comes to poetry generation using generative text models, Zhang and Lapata BIBREF9, Yi et al. BIBREF10 and Wang et al. BIBREF11 use language modeling to generate Chinese poems. However, none of these methods provide feedback on the quality of the generated sample and hence, do not address the qualitative objective required for creative decoding. For the task of text generation, MaskGAN BIBREF12 uses a Reinforcement Learning signal from the discriminator, FMD-GAN BIBREF13 uses an optimal transport mechanism as an objective function. GumbelGAN BIBREF14 uses Gumbel-Softmax distribution that replaces the non-differentiable sample from a categorical distribution with a differentiable sample to propagate stronger gradients. Li et al. BIBREF1 use a discriminator for a diversity promoting objective. Yu et al. BIBREF15 use SeqGAN to generate poetry and comment on the performance of SeqGAN over MLE in human evaluations, encouraging our study of GANs for creative text generation. However, these studies do not focus solely on creative text.
Our experiments suggest that generative text models, while very good at encapsulating semantic, syntactic and domain information, perform better with external feedback from a discriminator for fine-tuning objectiveless decoding tasks like that of creative text.
|
In which aspect do generative text models perform better?
|
They perform better with external feedback from a discriminator for fine-tuning objectiveless decoding tasks like that of creative text.
|
null | false
| null |
Treatment of ACO is based on expert opinion as there are no universally accepted clinical guidelines. Treatment is usually based on whether clinical features of asthma or COPD predominate. Inhaled corticosteroids are the primary treatment in those with ACOS. Inhaled corticosteroids (ICS) should be continued in those with asthma who develop decreased airway responsiveness to bronchodilators consistent with ACO. Therapy can be escalated to include a long acting beta-agonist (LABA) and inhaled steroid combination (ICS-LABA) or by adding on a long-acting anti-muscarinic inhaler (LAMA), known as triple therapy, in those with more severe or resistant disease.
Monoclonal antibodies targeting type 2 inflammation (which is predominant in asthma) have been used to treat severe asthma, and may also be used in severe cases of ACO. These monoclonal antibodies include omalizumab (an Anti-IgE antibody), mepolizumab (an anti-IL-5 antibody) and benralizumab (an anti-IL-5 receptor α antibody). People with ACOS and eosinophilia have a better response to ICS; with fewer exacerbations and hospitalizations seen in ACOS treated with long term ICS. Systemic corticosteroids (intravenous or oral steroids) may be used during exacerbations of ACOS.
https://en.wikipedia.org/wiki/Asthma-COPD_overlap
|
From this passage, extract the treatments for ACO in a bulleted list.
|
-Inhaled corticosteroids
-long acting beta-agonist (LABA) and inhaled steroid combination (ICS-LABA)
-triple therapy (in other words, adding on a long-acting anti-muscarinic inhaler (LAMA)
-monoclonal antibodies targeting type 2 inflammation
-systemic corticosteroids (intravenous or oral steroids)
|
null | false
| 41
|
Performance appraisal (PA) is an important HR process, particularly for modern organizations that crucially depend on the skills and expertise of their workforce. The PA process enables an organization to periodically measure and evaluate every employee's performance. It also provides a mechanism to link the goals established by the organization to its each employee's day-to-day activities and performance. Design and analysis of PA processes is a lively area of research within the HR community BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 .
The PA process in any modern organization is nowadays implemented and tracked through an IT system (the PA system) that records the interactions that happen in various steps. Availability of this data in a computer-readable database opens up opportunities to analyze it using automated statistical, data-mining and text-mining techniques, to generate novel and actionable insights / patterns and to help in improving the quality and effectiveness of the PA process BIBREF4 , BIBREF5 , BIBREF6 . Automated analysis of large-scale PA data is now facilitated by technological and algorithmic advances, and is becoming essential for large organizations containing thousands of geographically distributed employees handling a wide variety of roles and tasks.
A typical PA process involves purposeful multi-step multi-modal communication between employees, their supervisors and their peers. In most PA processes, the communication includes the following steps: (i) in self-appraisal, an employee records his/her achievements, activities, tasks handled etc.; (ii) in supervisor assessment, the supervisor provides the criticism, evaluation and suggestions for improvement of performance etc.; and (iii) in peer feedback (aka INLINEFORM0 view), the peers of the employee provide their feedback. There are several business questions that managers are interested in. Examples:
In this paper, we develop text mining techniques that can automatically produce answers to these questions. Since the intended users are HR executives, ideally, the techniques should work with minimum training data and experimentation with parameter setting. These techniques have been implemented and are being used in a PA system in a large multi-national IT company.
The rest of the paper is organized as follows. Section SECREF2 summarizes related work. Section SECREF3 summarizes the PA dataset used in this paper. Section SECREF4 applies sentence classification algorithms to automatically discover three important classes of sentences in the PA corpus viz., sentences that discuss strengths, weaknesses of employees and contain suggestions for improving her performance. Section SECREF5 considers the problem of mapping the actual targets mentioned in strengths, weaknesses and suggestions to a fixed set of attributes. In Section SECREF6 , we discuss how the feedback from peers for a particular employee can be summarized. In Section SECREF7 we draw conclusions and identify some further work.
Performance appraisal (PA) is an important HR process, particularly for modern organizations that crucially depend on the skills and expertise of their workforce. The PA process enables an organization to periodically measure and evaluate every employee’s performance. It also provides a mechanism to link the goals established by the organization to its each employee’s day-to-day activities and performance.
|
What is PA?
|
Performance appraisal (PA) is an important HR process, particularly for modern organizations that crucially depend on the skills and expertise of their workforce.
|
null | false
| 135
|
As a starting point for our pilots, we made use of texts from the InScript corpus BIBREF10 , which provides stories centered around everyday situations (see Section SECREF7 ). We conducted three different pilot studies to determine the best way of collecting questions that require inference over commonsense knowledge:
The most intuitive way of collecting reading comprehension questions is to show texts to workers and let them formulate questions and answers on the texts, which is what we tried internally in a first pilot. Since our focus is to provide an evaluation framework for inference over commonsense knowledge, we manually assessed the number of questions that indeed require common sense knowledge. We found too many questions and answers collected in this manner to be lexically close to the text.
In a second pilot, we investigated the option to take the questions collected for one text and show them as questions for another text of the same scenario. While this method resulted in a larger number of questions that required inference, we found the majority of questions to not make sense at all when paired with another text. Many questions were specific to a text (and not to a scenario), requiring details that could not be answered from other texts. Since the two previous pilot setups resulted in questions that centered around the texts themselves, we decided for a third pilot to not show workers any specific texts at all. Instead, we asked for questions that centered around a specific script scenario (e.g. eating in a restaurant). We found this mode of collection to result in questions that have the right level of specificity for our purposes: namely, questions that are related to a scenario and that can be answered from different texts (about that scenario), but for which a text does not need to provide the answer explicitly.
The next section will describe the mode of collection chosen for the final dataset, based on the third pilot, in more detail.
As a starting point for our pilots, we made use of texts from the InScript corpus (Modi et al., 2016), which provides stories centered around everyday situations (see Section 2.2.2.). We conducted three different pilot studies to determine the best way of collecting questions that require inference over commonsense knowledge:
|
How is the best way of collecting questions decided?
|
They conducted three different pilot studies to determine the best way of collecting questions that require inference over commonsense knowledge.
|
null | false
| null |
Why do I have blonde hair?
|
You have blonde hair because of your parent's genes. If your mom and dad both have blonde hair your hair is blonde through a dominant gene. If either your mom or your dad has brown or any other color hair then you got blonde hair through a recessive gene. However, it would be more common for you to have brown hair if both your parents had brown hair.
|
|
null | false
| null |
Why Italy is known for
|
Italy is known for island vacation. There are so many island (amalfi, Capri etc ) and their food and wine
|
|
null | false
| null |
Write a short story about a woman named Mary who is a barista making a latte.
|
Mary was a barista at the local coffee shop. She has worked at this specific cafe for over 5 years and has learned to love the job, and has grown close to her coworkers and regular customers. While Mary is a specialist in preparing many types of coffee and tea drinks, her favorite of all time was to make the latte. One morning around 8:00 am, a regular customer named Eric came into the cafe. Eric usually orders a latte every morning around the same time and Mary really enjoys preparing his drink. When Mary saw Eric walk in, she started preparing his drink without even asking what he would order. First Mary greeted Eric and engaged in some small talk, conversing about the seasonally warm weather we were having. While simultaneously having this conversation, Mary was in process of making the coffee drink. Mary first ground 18g of espresso roast coffee on a fine grind setting. Once the coffee was ground, she filled the portafilter with the ground coffee. Once the portafilter was full of coffee, Mary tamped the coffee with a tamper, using about 30 pounds of force. When the portafilter was ready, she fixed it upon the espresso machine. Next Mary filled a milk pitcher with 12 ounces of whole milk. She then took the pitcher, held it to the steam wand, and started steaming the milk. For the first 5 seconds, she held the milk just below the surface of the milk to aerate the milk, and then moved the pitcher up so the steam wand was fully emerged. While the milk was steaming, Mary pressed the espresso button on the machine to start the pouring of the espresso into the two shot glasses. Mary was an expert in doing multiple latte making tasks simultaneously. As the milk finished steaming, she removed the pitcher, wiped down the steam wand, and set the milk aside. At that time, the espresso shots were done, and Mary poured the espresso from the shot glasses into the 16 oz paper cup. She then took the steamed milk and poured it over the espresso shots in the cup. She poured the milk in a specific way that would create latte art, this time she made a little heart out of the milk and foam. Once all the milk was poured, she put a cup on the coffee and handed the coffee over to Eric, who she was still having a conversation with. Once he received the coffee, Eric scanned his gift card on the computer which Mary had prepared for payment for the latte. After Eric was done paying, he took a sip of the drink and complimented Mary for another excellently prepared latte. After taking his sip, Eric then left the store, and got into this car, and drove away. It was moments like this every morning that reminded Mary why she loves this job so much, and looked forward to helping more customers for the remainder of her shift that morning.
|
|
null | false
| null |
In mathematics, gradient descent (also often called steepest descent) is a first-order iterative optimization algorithm for finding a local minimum of a differentiable function. The idea is to take repeated steps in the opposite direction of the gradient (or approximate gradient) of the function at the current point, because this is the direction of steepest descent. Conversely, stepping in the direction of the gradient will lead to a local maximum of that function; the procedure is then known as gradient ascent.
Gradient descent is generally attributed to Augustin-Louis Cauchy, who first suggested it in 1847. Jacques Hadamard independently proposed a similar method in 1907. Its convergence properties for non-linear optimization problems were first studied by Haskell Curry in 1944, with the method becoming increasingly well-studied and used in the following decades.
|
Given this text about gradient descent, what is gradient descent, and who invented it?
|
In mathematics, gradient descent (also often called steepest descent) is a first-order iterative optimization algorithm for finding a local minimum of a differentiable function. Gradient descent is generally attributed to Augustin-Louis Cauchy, who first suggested it in 1847
|
null | false
| 308
|
Given the large collections of unstructured and semi-structured data available on the web, there is a crucial need to enable quick and efficient access to the knowledge content within them. Traditionally, the field of information extraction has focused on extracting such knowledge from unstructured text documents, such as job postings, scientific papers, news articles, and emails. However, the content on the web increasingly contains more varied types of data, including semi-structured web pages, tables that do not adhere to any schema, photographs, videos, and audio. Given a query by a user, the appropriate information may appear in any of these different modes, and thus there's a crucial need for methods to construct knowledge bases from different types of data, and more importantly, combine the evidence in order to extract the correct answer.
Motivated by this goal, we introduce the task of multimodal attribute extraction. Provided contextual information about an entity, in the form of any of the modes described above, along with an attribute query, the goal is to extract the corresponding value for that attribute. While attribute extraction on the domain of text has been well-studied BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 , to our knowledge this is the first time attribute extraction using a combination of multiple modes of data has been considered. This introduces additional challenges to the problem, since a multimodal attribute extractor needs to be able to return values provided any kind of evidence, whereas modern attribute extractors treat attribute extraction as a tagging problem and thus only work when attributes occur as a substring of text.
In order to support research on this task, we release the Multimodal Attribute Extraction (MAE) dataset, a large dataset containing mixed-media data for over 2.2 million commercial product items, collected from a large number of e-commerce sites using the Diffbot Product API. The collection of items is diverse and includes categories such as electronic products, jewelry, clothing, vehicles, and real estate. For each item, we provide a textual product description, collection of images, and open-schema table of attribute-value pairs (see Figure 1 for an example). The provided attribute-value pairs only provide a very weak source of supervision; where the value might appear in the context is not known, and further, it is not even guaranteed that the value can be extracted from the provided evidence. In all, there are over 4 million images and 7.6 million attribute-value pairs. By releasing such a large dataset, we hope to drive progress on this task similar to how the Penn Treebank BIBREF5 , SQuAD BIBREF6 , and Imagenet BIBREF7 have driven progress on syntactic parsing, question answering, and object recognition, respectively.
To asses the difficulty of the task and the dataset, we first conduct a human evaluation study using Mechanical Turk that demonstrates that all available modes of information are useful for detecting values. We also train and provide results for a variety of machine learning models on the dataset. We observe that a simple most-common value classifier, which always predicts the most-common value for a given attribute, provides a very difficult baseline for more complicated models to beat (33% accuracy). In our current experiments, we are unable to train an image-only classifier that can outperform this simple model, despite using modern neural architectures such as VGG-16 BIBREF8 and Google's Inception-v3 BIBREF9 . However, we are able to obtain significantly better performance using a text-only classifier (59% accuracy). We hope to improve and obtain more accurate models in further research.
In order to support research on this task, we release the Multimodal Attribute Extraction (MAE) dataset, a large dataset containing mixed-media data for over 2.2 million commercial product items, collected from a large number of e-commerce sites using the Diffbot Product API.
|
What dataset do they introduce in the paper?
|
Multimodal attribute extraction (MAE) dataset.
|
null | false
| null |
What is american football?
|
American football is a an American sports. Other countries use the term "football" to refer to soccer. American football is a team sport and two opposing teams play against each other. The objective of the game is to get a football to the end zone of the opponent. If you can accomplish this you score a touchdown. The ball moves down the field when players catch the ball, and run it opposite of their own end zone. After they score they will get the opportunity to kick a field goal as well.
|
|
null | false
| null |
Why indian Marriage is so long process
|
1. Its an ancient process to get married and go through all rituals for Indians. So it takes time. Because they pray for so long in a Marriage
|
|
null | false
| null |
Name a very famous family of composers & musicians of late baroque
|
Bach family gave to the world few famous composers - Johann Sebastian Bach, Carl Philipp Emanuel Bach, Wilhelm Friedemann Bach, Johann Christian Bach and others
|
|
null | false
| 473
|
Table and Table show the performance comparison on two different imbalanced CIFAR10 datasets. In these two tables, we use bold values to denote the highest accuracy among all methods and use the underline values to indicate our SRAT variants which achieve the highest accuracy among their corresponding baseline methods utilizing the same loss function for making predictions. Due to the limited space, we report the performance comparison on other imbalanced datasets created from CIFAR10, CIFAR100 and SVHN datasets in Appendix A.6.
From Table, we can make the following observations. First, compared to baseline methods, our SRAT method obtains improved performance in terms of both overall standard & robust accuracy under almost all imbalanced settings. More importantly, SRAT makes significant improvements on those under-represented classes, especially under the extremely imbalanced settings. For the CIFAR10 Step-100 dataset, our SRAT-Focal method improves the standard accuracy on under-represented classes from 21.81% achieved by the best baseline method utilizing Focal loss to 51.83% and robust accuracy from 3.24% to 15.89%. These results demonstrate that our SRAT method is able to obtain more robustness under imbalanced settings. Second, the performance gap among three SRAT variants are mainly caused by the gap between the loss functions in these methods. As shown in these two tables, DRCB-LDAM typically performs better than DRCE-CE and DRCB-Focal, and similarly, SRAT-LDAM outperforms SRAT-CE and SRAT-Focal under the same settings.
In these two tables, we use bold values to denote the highest accuracy among all methods and use the underline values to indicate our SRAT variants which achieve the highest accuracy among their
corresponding baseline methods utilizing the same loss function for making predictions.
|
Is the confidence score missing in the paper?
|
In our revised version, We updated all result tables to include mean and variance of models' performance obtained from three times trials with different model initialization. Based on our experimental results, we observed that our SRAT method outperforms all baseline methods under most imbalanced scenarios.
|
null | false
| null |
Why is pickleball the greatest sport on the planet?
|
Pickleball is a wonderful sport, though it is arguable if it is the greatest sport on the planet. Pickeball requires little skill to get started, you simply need to be able to hit a ball with a raquet and move around a small court. Pickeball is a cross between ping pong and tennis but is simpler. It gets you moving and is fast-paced without requiring significant kill or practice. Pickleball is a social sport, it is usually played in groups of two or four and people generally play multiple games in a setting.
|
|
null | false
| null |
In computing, a data warehouse (DW or DWH), also known as an enterprise data warehouse (EDW), is a system used for reporting and data analysis and is considered a core component of business intelligence. DWs are central repositories of integrated data from one or more disparate sources. They store current and historical data in one single place that are used for creating analytical reports for workers throughout the enterprise. This is beneficial for companies as it enables them to interrogate and draw insights from their data and make decisions.
The data stored in the warehouse is uploaded from the operational systems (such as marketing or sales). The data may pass through an operational data store and may require data cleansing for additional operations to ensure data quality before it is used in the DW for reporting.
Extract, transform, load (ETL) and extract, load, transform (ELT) are the two main approaches used to build a data warehouse system.
|
What is data warehousing?
|
A centralized repository of data from multiple source systems organized in subject areas typically in 3NF or dimensional data modeling techniques.
|
1909.13375
| false
| null |
MTMSN BIBREF4 is the first, and only model so far, that specifically tried to tackle the multi-span questions of DROP. Their approach consisted of two parts. The first was to train a dedicated categorical variable to predict the number of spans to extract. The second was to generalize the single-span head method of extracting a span, by utilizing the non-maximum suppression (NMS) algorithm BIBREF7 to find the most probable set of non-overlapping spans. The number of spans to extract was determined by the aforementioned categorical variable.
MTMSN BIBREF4 is the first, and only model so far, that specifically tried to tackle the multi-span questions of DROP.
|
What is the previous model that attempted to tackle multi-span questions as a part of its design?
|
The answers are shown as follows:
* MTMSN BIBREF4
|
null | false
| null |
Tell me who were the last three monarchs of the United Kingdom. Give it to me in a bulleted list with the format <name> (<start date> - <end date>)
|
The last three monarchs of the United Kingdom are:
- Charles III (8 September 2022 - Present)
- Elizabeth II (6 February 1952 - 8 September 2022)
- George VI (11 December 1936 - 6 February 1952)
|
|
null | false
| null |
How can I change a bad habit?
|
Each habit has a trigger event, an action you subconsciously take after the trigger event happens, and the reward that make you keep following the same habit. When you say a habit is bad, it actually means the action you take after the trigger event is something you do not want to keep doing.
An effective way to make changes is to replace the action to something different. To make this alternative action effective, this action will need to also produce the reward that your brain is seeking for.
In order to find an effective alternative action, you will need to understand what is your trigger and what is the reward.
Figuring out the trigger is the first step. You can try to think about what happened when you started to take actions. One way to find the pattern is to write down what happened. Then, you can look at them to find the pattern.
Once you find out the trigger, you can try to remind you to change your action after experiencing the trigger event. With your changed action, you will be able to gradually figure out the reward hidden in your old habit. Some actions will not give you the satisfying feeling like old time. After some try-and-error, you can likely tell the new action that actually gives you the same satisfying feeling. Then, you can keep following the new action. Essentially, this new action helps you form the new habit, which suppress the old habit.
Finally, please note that when you are having extra stress, you may find it is harder to follow the new habit and start to worry about picking up the old habit again. Please trust yourself to keep the new habit. You faith will help you push through difficult situations.
|
|
null | false
| null |
Esko Olavi Ahonen (born June 13, 1955 in Evijärvi) is a Finnish politician and member of the parliament from the Centre Party. He was elected to Parliament of Finland in 2003 for the constituency of Vaasa. In the 2011 election he was dropped out of the parliament.
|
Extract from the text the year in which Esko Olavi Ahonen was born
|
1955
|
null | false
| null |
Suntory Holdings Limited (サントリーホールディングス株式会社, Santorī Hōrudingusu Kabushiki-gaisha) (commonly referred to as simply Suntory) is a Japanese multinational brewing and distilling company group. Established in 1899, it is one of the oldest companies in the distribution of alcoholic beverages in Japan, and makes Japanese whisky. Its business has expanded to other fields, and the company now also makes soft drinks and operates sandwich chains. With its 2014 acquisition of Beam, Inc., it has diversified internationally and become the third largest maker of distilled beverages in the world. Suntory is headquartered in Dojimahama 2-chome, Kita-ku, Osaka, Osaka Prefecture. Suntory Beverage & Food Limited (サントリー食品インターナショナル株式会社, Santorī shokuhin
|
When was Suntory established?
|
Suntory was established in 1899
|
null | false
| null |
List some of the television gameshows that Bob Stewart created, co-created, and/or produced.
|
The Price is Right, Password, To Tell the Truth, Pyramid shows (such as the $10,000 Pyramid, the $25,000 Pyramid, and so on), Jackpot!, Winning Streak
|
|
null | false
| null |
How should I clean a stainless steel pan?
|
The simplest way is to add soapy water to the pan and bring it to a boil. Scraping off the food bits with a wooden spatula becomes easy then. For tougher stains you should use a combination of baking soda and vinegar. Let the mixture sit in the pan for a while, rinse it out, and then follow the soapy water procedure.
|
|
null | false
| null |
The county is home to several endangered species including the San Francisco garter snake and the San Bruno elfin butterfly, both of which are endemic to San Mateo County. The endangered Ridgway's Rail is also found on the shores of San Francisco Bay, in the cities of Belmont and San Mateo. The endangered wildflower Hickman's potentilla is found near the Pacific Ocean on the lower slopes of Montara Mountain. The endangered wildflowers White-rayed pentachaeta, Pentachaeta bellidiflora, San Mateo Woolly Sunflower, Eriophyllum latilobum, Marin Dwarf Flax, Hesperolinon congestum and the San Mateo Thornmint, Acanthomintha duttonii, are found in the vicinity of the Crystal Springs Reservoir.
In May 2014, a California condor was spotted near Pescadero, a coastal community south of San Francisco—it was the first California condor spotted in San Mateo County since 1904. The condor, tagged with the number "597", and also known as "Lupine", is one of 439 condors living in the wild or captivity in California, Baja California and Arizona. The three-year-old female flew more than 100 miles (160 km) north from Pinnacles National Park, in San Benito County, on May 30, and landed on a private, forested property near Pescadero, on the San Mateo County Coast, where it was photographed by a motion-activated wildlife camera. Harold Heath, Professor Emeritus, of Stanford University was responsible for the 1904 sighting, 1 mile (1.6 km) west of the University campus.
Pumas (Puma concolor), also known as cougars or mountain lions, roam the county.
Tule elk (Cervus canadensis nannodes) were native to San Mateo County and among the "favored foods" of the Ohlone people based on ethnohistoric and archeological evidence there. The discovery of two elk specimens made news in 1962, one a royal elk (royal elk bulls have six tines per antler) from a peat bog excavated in Pacifica's historic Laguna Alta, and now in the Museum of Vertebrate Zoology collection. These may date from the time of Spanish settlement. Laguna Alta lay just south of the Interstate 280 and Skyline Boulevard intersection, east of Mussel Rock. The California Academy of Sciences also has an elk skull fragment collected one mile inland from the mouth of Purisima Creek in 1951. Additional coastal elk remains dating from the Middle and Late Periods in Northern California were found in at least five more late Holocene archeological sites in San Mateo County: SMA-115 (Montara State Beach site), SMA-118 (Bean Hollow State Beach site), SMA-244 (Butano Ridge site), SMA-97 (Año Nuevo Creek site) and SMA-218 (Año Nuevo State Reserve site). On the eastern side of the San Francisco Peninsula, elk remains were also unearthed at multiple archaeological sites along San Francisquito Creek.
|
What are three mammals found within San Mateo County?
|
Pumas and tule elk are two mammal species found in San Mateo County.
|
null | false
| 335
|
The data for this study was gathered from the ted.com website on November 15, 2017. We removed the talks published six months before the crawling date to make sure each talk has enough ratings for a robust analysis. More specifically, we filtered any talk that—
More specifically, we filtered any talk that— 1. was published less than 6 months prior to the crawling date, 2. contained any of the following keywords: live music, dance, music, performance, entertainment, or, 3. contained less than 450 words in the transcript.
|
What talk is filtered by authors?
|
1. was published less than 6 months prior to the crawling date, 2. contained any of the following keywords: live music, dance, music, performance, entertainment, or, 3. contained less than 450 words in the transcript.
|
1904.00648
| false
| null |
In this work, we analyze a set of tweets related to a specific classical music radio channel, BBC Radio 3, interested in detecting two types of musical named entities, Contributor and Musical Work.
In this work, we analyze a set of tweets related to a specific classical music radio channel, BBC Radio 3, interested in detecting two types of musical named entities, Contributor and Musical Work.
|
What language is the Twitter content in?
|
English
|
null | false
| 84
|
Our experiments evaluate the effect of shallow syntax, via contextualization (mSynC, §SECREF2) and features (§SECREF3). We provide comparisons with four baselines—ELMo-transformer BIBREF0, our reimplementation of the same, as well as two cwr-free baselines, with and without shallow syntactic features. Both ELMo-transformer and mSynC are trained on the 1B word benchmark corpus BIBREF19; the latter also employs chunk annotations (§SECREF2). Experimental settings are detailed in Appendix §SECREF22.
We employ four tasks to test the impact of shallow syntax. The first three, namely, coarse and fine-grained named entity recognition (NER), and constituency parsing, are span-based; the fourth is a sentence-level sentiment classification task.
|
How many tasks were employed to test the impact of shallow syntax?
|
Four tasks.
|
null | false
| null |
How would you go about a planning a backpacking trip into the Grand Canyon National Park.
|
Suppose you wish to camp anywhere in the park other than in developed campgrounds on the North or South Rim. In that case, you must obtain a permit from the Backcountry Information Center. Camping sites are limited in the Grand Canyon, and spots are issued four months before a given month. E.g., For January, permit applications should be submitted between Aug 20 and Sep 1 to be considered "earliest requested."
For requests received during the Earliest Consideration period, the day received does not matter; all Earliest Consideration requests received during this span of dates are treated the same. It may take Rangers up to three weeks to process all Earliest Consideration requests.
Subsequent Requests: Once all Earliest Consideration requests are processed, Rangers will begin processing written requests received on subsequent dates. These Subsequent Requests are kept and considered in the order received. Submissions will be accepted as late as two weeks before a hike start date.
If your permit request is selected, you must diligently plan your trip. You will need to carry everything you need for the duration of your journey with you except for water which is few and far between. Know where the water sources are on your route and carry filtration devices, water purification equipment, and backup equipment. The hotter the trip, the more water you will require. Have a plan for where the necessary water will be on your route, and double-check with the park service ahead of time that water is still there. Dehydration is the number one reason for rescue and death in the canyon. Some springs dry up in the summer months.
Travel as light as possible; you must pack everything you bring into the canyon, including used toilet paper. Plan each meal, knowing how much calories and water you consume based on your planned exertion level. Calorie and water requirements also change based on temperature (the hotter the day, the more water and food you will need). Also, make sure you have the appropriate gear based on temperatures. It is approximately 20 degrees cooler at the top of the canyon than at the bottom because there is a 5000 feet vertical difference at the South Rim and 6000 at the North Rim. Ensure your footwear is "broken in", and carry first-aid kits with particular attention to foot care, such as blister pads and moleskin.
Start with shorter, more leisurely trips; if you like them, move up to more adventurous multi-day trips. Plan on rest days when you can enjoy the majestic scenery of one of the world's natural wonders. Above all, enjoy the canyon; it's a magical place 6 million years in the making that exposes geology over 4 billion years old.
|
|
null | false
| null |
Classify each of the following as either a flying or aquatic animal: salmon, eagle, dolphin, whale, octopus, bat, goose
|
Among the given list, salmon, dolphin, whale, octopus are aquatic animals whereas eagle, bat and goose are flying animals.
|
|
null | false
| null |
Wood-decay fungi can be classified according to the type of decay that they cause. The best-known types are brown rot, soft rot, and white rot. Each produce different enzymes, can degrade different plant materials, and can colonise different environmental niches. Brown rot and soft rot both digest a tree's cellulose and hemicellulose but not its lignin; white rot digests lignin as well. The residual products of decomposition from fungal action have variable pH, solubility and redox potentials. Over time this residue becomes incorporated in the soil and sediment so can have a noticeable effect on the environment of that area.
|
Extract the best known types of wood-decay fungi described in this paragraph and list them in a bulleted list.
|
The best known types of wood-decay fungi are:
- Brown rot
- Soft rot
- White rot
|
null | false
| null |
Johnny Galecki as Leonard Hofstadter: An experimental physicist with an IQ of 173, who received his Ph.D. when he was 24 years old. Leonard is a nerd who loves video games, comic books, and Dungeons & Dragons. Leonard is the straight man of the series, sharing an apartment in Pasadena, CA, with Sheldon Cooper. Leonard is smitten with his new neighbor Penny when they first meet, and they eventually marry.
Jim Parsons as Sheldon Cooper: Originally from Galveston, Texas, Sheldon was a child prodigy with an eidetic memory who began college at the age of eleven and earned a Ph.D. at age sixteen. He is a theoretical physicist researching quantum mechanics and string theory, and, despite his IQ of 187, he finds many routine aspects of social situations difficult to grasp. He is determined to have his own way, continually boasts of his intelligence, and has an extremely ritualized way of living. Despite these quirks, he begins a relationship with Amy Farrah Fowler, and they eventually marry.
Kaley Cuoco as Penny: An aspiring actress from Omaha, Nebraska. Penny moves in across the hall from Sheldon and Leonard. She waits tables and occasionally tends the bar at The Cheesecake Factory. After giving up hope of becoming a successful actress, Penny becomes a pharmaceutical sales representative. Penny becomes friends with Bernadette and Amy, and they often hang out in each other's apartments. Penny and Leonard form a relationship and eventually marry.
Simon Helberg as Howard Wolowitz: An aerospace engineer who got his master's degree at the Massachusetts Institute of Technology. Howard is Jewish and lived with his mother, Debbie (Carol Ann Susi). Unlike Sheldon, Leonard, Raj, Bernadette, and Amy, Howard does not hold a doctorate. He trains as an astronaut and goes into space as a payload specialist on the International Space Station. Howard initially fancies himself as a ladies man, but he later starts dating Bernadette, and they get engaged and married. Howard also has a tendency to waste money on toys and argues with Bernadette because of his oddly low income as an engineer and her high income as a pharmaceutical biochemist.
Kunal Nayyar as Rajesh Koothrappali: A particle astrophysicist originally from New Delhi, India. Initially, Raj had selective mutism, rendering him unable to talk to or be around women unless under the influence of alcohol. Raj also has very feminine tastes and often takes on a stereotypical female role in his friendship with Howard as well as in the group of four men. Raj later dates Lucy (Kate Micucci), who also suffers from social anxiety, but it eventually ends. He later speaks to Penny without alcohol, overcoming his selective mutism. He begins dating Emily Sweeney, and their relationship later becomes exclusive. In the series' final season, Raj has an on-again, off-again engagement with a fellow Indian, a hotel concierge named Anu (Rati Gupta). He also has a Yorkshire Terrier named Cinnamon.
These actors were first credited as guest stars and later promoted to main cast:
Sara Gilbert as Leslie Winkle (recurring season 1, starring season 2, guest seasons 3, 9): A physicist who works in the same lab as Leonard. In appearance, she is essentially Leonard's female counterpart and has conflicting scientific theories with Sheldon. Leslie has casual sex with Leonard and later Howard. Gilbert was promoted to a main cast member during the second season but resumed guest star status because producers could not come up with enough material for the character. Gilbert returned to The Big Bang Theory for its 200th episode.
Melissa Rauch as Bernadette Rostenkowski-Wolowitz (recurring season 3, starring seasons 4–12): A young woman who initially is a co-worker at The Cheesecake Factory with Penny to pay her way through graduate school, where she is studying microbiology. Bernadette is introduced to Howard by Penny; at first, they do not get along, apparently having nothing in common. They date and later get engaged and married. Although generally a sweet and good-natured person, Bernadette has a short fuse and can be vindictive and lash out when provoked.
Mayim Bialik as Amy Farrah Fowler (guest star season 3, starring seasons 4–12): A woman selected by an online dating site as Sheldon's perfect mate, Amy is from Glendale, CA. While she and Sheldon initially share social cluelessness, after befriending Penny and Bernadette, she eventually becomes more interested in social and romantic interaction. Her relationship with Sheldon slowly progresses to the point at which Sheldon considers her his girlfriend, and eventually, they get married. Amy believes she and Penny are best friends, a sentiment that Penny does not initially share. Amy has a Ph.D. in neurobiology.
Kevin Sussman as Stuart Bloom (recurring seasons 2–5, 7, starring seasons 6, 8–12): A mild-mannered, under-confident owner of a comic book store. A competent artist, Stuart is a graduate of the prestigious Rhode Island School of Design. Though he is socially awkward, he possesses slightly better social skills. Stuart implies he is in financial trouble and that the comic book store now also is his home. He is later invited to join the guys' group while Howard is in space. Stuart gets a new job caring for Howard's mother later. After Mrs. Wolowitz's death, Stuart continues to live in her home, along with Howard and Bernadette, until he finds a place of his own.
Laura Spencer as Emily Sweeney (recurring seasons 7–8, 10, starring season 9): A dermatologist at Huntington Hospital. Emily went to Harvard and delights in the macabre, and she states that she likes her job because she can cut things with knives. Prior to meeting Raj, Emily was set up on a blind date with Howard. After finding Emily's online dating profile, Raj has Amy contact her as his wingman instead. Their relationship becomes exclusive, but Raj later breaks up with Emily when he becomes infatuated with Claire (Alessandra Torresani), a bartender and children's author.
|
Given the list of the Big Bang Theory actress, select only the main cast.
|
Kaley Cuoco as Penny, Melissa Rauch as Bernadette Rostenkowski-Wolowitz, Mayim Bialik as Amy Farrah Fowler
|
null | false
| 41
|
In this paper, we used the supervisor assessment and peer feedback text produced during the performance appraisal of 4528 employees in a large multi-national IT company. The corpus of supervisor assessment has 26972 sentences. The summary statistics about the number of words in a sentence is: min:4 max:217 average:15.5 STDEV:9.2 Q1:9 Q2:14 Q3:19.
In this paper, we used the supervisor assessment and peer feedback text produced during the performance appraisal of 4528 employees in a large multi-national IT company. The corpus of supervisor assessment has 26972 sentences. The summary statistics about the number of words in a sentence is: min:4 max:217 average:15.5 STDEV:9.2 Q1:9 Q2:14 Q3:19.
|
What data was used in the paper?
|
They used the supervisor assessment and peer feedback text produced during the performance appraisal of 4528 employees in a large multi-national IT company.
|
1909.00100
| false
| null |
For model distillation BIBREF6 , we extract sentences from Wikipedia in languages for which public multilingual is pretrained. For each sentence, we use the open-source BERT wordpiece tokenizer BIBREF4 , BIBREF1 and compute cross-entropy loss for each wordpiece: INLINEFORM0
where INLINEFORM0 is the cross-entropy function, INLINEFORM1 is the softmax function, INLINEFORM2 is the BERT model's logit of the current wordpiece, INLINEFORM3 is the small BERT model's logits and INLINEFORM4 is a temperature hyperparameter, explained in Section SECREF11 .
To train the distilled multilingual model mMiniBERT, we first use the distillation loss above to train the student from scratch using the teacher's logits on unlabeled data. Afterwards, we finetune the student model on the labeled data the teacher is trained on.
For model distillation BIBREF6 , we extract sentences from Wikipedia in languages for which public multilingual is pretrained. For each sentence, we use the open-source BERT wordpiece tokenizer BIBREF4 , BIBREF1 and compute cross-entropy loss for each wordpiece: INLINEFORM0
where INLINEFORM0 is the cross-entropy function, INLINEFORM1 is the softmax function, INLINEFORM2 is the BERT model's logit of the current wordpiece, INLINEFORM3 is the small BERT model's logits and INLINEFORM4 is a temperature hyperparameter, explained in Section SECREF11 .
To train the distilled multilingual model mMiniBERT, we first use the distillation loss above to train the student from scratch using the teacher's logits on unlabeled data. Afterwards, we finetune the student model on the labeled data the teacher is trained on.
|
How do they compress the model?
|
The answers are shown as follows:
* we extract sentences from Wikipedia in languages for which public multilingual is pretrained. For each sentence, we use the open-source BERT wordpiece tokenizer BIBREF4 , BIBREF1 and compute cross-entropy loss for each wordpiece: INLINEFORM0
|
null | false
| null |
Why is baseball considered such a quintessential US sport?
|
There are many factors that make many Americans think of baseball as "America's pastime." Possibly the biggest factor is that baseball has been played for so long that it's history ties together generations of Americans through their common rooting interest in their favorite teams. Another reason: baseball players became celebrities and figures in pop culture long before such a thing was common in other sports. While baseball has lagged in growth in recent times when compared to leagues such as the NFL and the NBA, it is still a vibrant sport with large revenues and extremely passionate fan bases around the country.
|
|
null | false
| null |
What is the name of the famous park in New York City?
|
Central Park
|
|
null | false
| null |
Should I buy a Shinkansen Rail Pass if I visit Japan?
|
Shinkansen Rail Pass is quite expensive. The cost is slightly cheaper comparing with Shinkansen round trip between Tokyo and Osaka. If you are planning to have a round trip between these 2 cities, then you should definitely consider to get a Shinkansen Rail Pass. If your Shinkansen trip is shorter than that, you probably better estimate the railway cost before get the Shinkansen Rail Pass.
|
|
null | false
| null |
Identify which car manufacturer is French or American: Renault, RAM
|
RAM is American, Renault is French
|
|
null | false
| 12
|
Bidirectional Encoder Representations from Transformers (BERT) is a novel Transformer BIBREF0 model, which recently achieved state-of-the-art performance in several language understanding tasks, such as question answering, natural language inference, semantic similarity, sentiment analysis, and others BIBREF1. While well-suited to dealing with relatively short sequences, Transformers suffer from a major issue that hinders their applicability in classification of long sequences, i.e. they are able to consume only a limited context of symbols as their input BIBREF2.
There are several natural language (NLP) processing tasks that involve such long sequences. Of particular interest are topic identification of spoken conversations BIBREF3, BIBREF4, BIBREF5 and call center customer satisfaction prediction BIBREF6, BIBREF7, BIBREF8, BIBREF9. Call center conversations, while usually quite short and to the point, often involve agents trying to solve very complex issues that the customers experience, resulting in some calls taking even an hour or more. For speech analytics purposes, these calls are typically transcribed using an automatic speech recognition (ASR) system, and processed in textual representations further down the NLP pipeline. These transcripts sometimes exceed the length of 5000 words. Furthermore, temporal information might play an important role in tasks like CSAT. For example, a customer may be angry at the beginning of the call, but after her issue is resolved, she would be very satisfied with the way it was handled. Therefore, simple bag of words models, or any model that does not include temporal dependencies between the inputs, may not be well-suited to handle this category of tasks. This motivates us to employ model such as BERT in this task.
In this paper, we propose a method that builds upon BERT's architecture. We split the input text sequence into shorter segments in order to obtain a representation for each of them using BERT. Then, we use either a recurrent LSTM BIBREF10 network, or another Transformer, to perform the actual classification. We call these techniques Recurrence over BERT (RoBERT) and Transformer over BERT (ToBERT). Given that these models introduce a hierarchy of representations (segment-wise and document-wise), we refer to them as Hierarchical Transformers. To the best of our knowledge, no attempt has been done before to use the Transformer architecture for classification of such long sequences.
Our novel contributions are:
Two extensions - RoBERT and ToBERT - to the BERT model, which enable its application in classification of long texts by performing segmentation and using another layer on top of the segment representations.
State-of-the-art results on the Fisher topic classification task.
Significant improvement on the CSAT prediction task over the MS-CNN model.
Two extensions - RoBERT and ToBERT - to the BERT model
|
What are the two extensions developed based on BERT by the team?
|
RoBERT and ToBERT.
|
null | false
| null |
Identify which car manufacturer is Chinese or American: Apex Motors, Chevrolet
|
Apex Motors is Chinese, Chevrolet is American
|
|
null | false
| null |
Give me a bulleted list of the current members of Pearl Jam.
|
- Jeff Ament
- Stone Gossard
- Mike McCready
- Eddie Vedder
- Matt Cameron
|
|
null | false
| 202
|
Natural language generation (NLG) has been one of the key topics of research in natural language processing, which was highlighted by the huge body of work on NLG surveyed in BIBREF0, BIBREF1. With the advances of several devices capable of understanding spoken language and conducting conversation with human (e.g., Google Home, Amazon Echo) and the shrinking gap created by the digital devices, it is not difficult to foresee that the market and application areas of NLG systems will continue to grow, especially in applications whose users are non-experts. In such application, a user often asks for certain information and waits for the answer and a NLG module would return the answer in spoken language instead of text such as in question-answering systems or recommendation systems. The NLG system in these two applications uses templates to generate the answers in natural language for the users. A more advanced NLG system in this direction is described in BIBREF2, which works with ontologies annotated using the Attempto language and can generate a natural language description for workflows created by the systems built in the Phylotastic project. The applications targeted by these systems are significantly different from NLG systems, whose main purpose is to generate high-quality natural language description of objects or reports, such as those reported in the recent AAAI conference BIBREF3, BIBREF4, BIBREF5.
The present paper is motivated by the need to generate natural language description of computational results to non-expert users such as those developed in the Phylotastic project. In this project, the users are experts in evolutionary biology but are none experts in ontologies and web services. When a user places a request, he/she will receive a workflow consisting of web services, whose inputs and outputs are specified by instances of classes in the ontologies working with web services, as well as the ordering and relationships between the services. To assist the user in understanding the workflow, a natural language description of the workflow is generated. In order to accomplish the task, the NLG system in the Phylotastic project proposes to annotate elements of the ontologies using Attempto, a simple subset of English with precisely defined syntax and semantics.
In this paper, we propose a system that addresses the limitation of the system discussed in the Phylotastic project BIBREF2. Specifically, we assume that the annotations given in an ontology are natural language sentences. This is a reasonable assumption given that the developers of an ontology are usually those who have intimate knowledge about entities described in the ontology and often have some sort of comments about classes, objects, and instances of the ontology. We then show that the system is very flexible and can be used for the same purpose with new ontologies.
The rest of the paper is organized as follows. Section SECREF2 briefly reviews the basics of Grammatical Framework (GF)BIBREF6. Section SECREF3 describes the main modules of the system. Section SECREF4 includes two use cases of the system using an available ontologies against in the context of reasoning about ontologies. Specifically, it compares with the system used in the Phylotastic project and an ontology about people. This section also contains a use case that highlights the versatility of the proposed system by addressing a challenge to create an abstract Wikipedia BIBREF7. Related works are discussed in Section SECREF5. Section SECREF6 concludes the paper.
Specifically, we assume that the annotations given in an ontology are natural language sentences.
|
What is their assumption?
|
The annotations given in an ontology are natural language sentences.
|
null | false
| null |
what does it mean to have brunette hair?
|
A person with brunette hair, or a brunette, is a person with brown or black hair color. People typically fall into the category of blonde, brunette, and red hair. It is sometimes misconceived to be only brown hair but it signifies all darker hair including black hair color.
|
|
1909.05478
| true
| null |
Feature selection is considered an indispensable task in text classification as it removes redundant and irrelevant features of the corpus BIBREF18. Broadly, feature selection approaches can be divided into three classes namely wrapper, embedded, and filter BIBREF7, BIBREF8. In recent years, researchers have proposed various filter based feature selection methods to raise the performance of document text classification BIBREF19.
To develop the vocabulary of most discriminative features, we remove all punctuation symbols and non-significant words (stop words) as a part of the preprocessing step. Furthermore, in order to rank the terms based on their discriminative power among the classes, we use filter based feature selection method named as Normalized Difference Measure (NDM)BIBREF5. Considering the features contour plot, Rehman et al. BIBREF5 suggested that all those features which exist in top left, and bottom right corners of the contour are extremely significant as compared to those features which exist around diagonals. State-of-the-art filter based feature selection algorithms such as ACC2 treat all those features in the same fashion which exist around the diagonals BIBREF5. For instance, ACC2 assigns same rank to those features which has equal difference ($|t_{pr} -f_{pr}|$) value but different $t_{pr}$ and $f_{pr}$ values. Whereas NDM normalizes the difference ($|t_{pr} -f_{pr}|$) with the minimum of $t_{pr}$ and $f_{pr}$ (min($t_{pr}$, $f_{pr}$)) and assign different rank to those terms which have same difference value. Normalized Difference Measure (NDM) considers those features highly significant which have the following properties:
In recent years, researchers have proposed various filter based feature selection methods to raise the performance of document text classification BIBREF19.
Furthermore, in order to rank the terms based on their discriminative power among the classes, we use filter based feature selection method named as Normalized Difference Measure (NDM)BIBREF5. Considering the features contour plot, Rehman et al. BIBREF5 suggested that all those features which exist in top left, and bottom right corners of the contour are extremely significant as compared to those features which exist around diagonals. State-of-the-art filter based feature selection algorithms such as ACC2 treat all those features in the same fashion which exist around the diagonals BIBREF5.
|
Is the filter based feature selection (FSE) a form of regularization?
|
No.
|
null | false
| null |
Does money make you happy?
|
The short answer is no. It makes it more convenient but does not make people inherently happier. Also, the more you try to acquire money and focus on it, the less happy you become because there is never enough. On average, people want about 15-20% more money than they currently have. Additionally, folks with less money but who live in close-knit communities are among the happiest and most satisfied with life.
|
|
null | false
| null |
What is the difference between instant oatmeal and regular oatmeal?
|
Instant oatmeal is ready to eat immediately after sitting in hot water, whereas regular oatmeal has to be boiled in hot water for a longer period of time (the specific duration depends on the type of oats).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.