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 |
Classify the following countries by area as large, medium, small: russia, china, switzerland, egypt, indonesia, argentina, netherlands, south africa, iran, san marino, togo, canada
|
Large: russia, china, canada
Medium: egypt, indonesia, argentina, south africa, iran
Small: switzerland, san marino, togo
|
|
null | false
| null |
Identify which instrument is string or percussion: Atabaque, Ajaeng
|
Ajaeng is string, Atabague is percussion.
|
|
null | false
| null |
The Porce III Dam is an embankment dam on the Porce River 90 kilometres (56 mi) northeast of Medellín in Antioquia Department, Colombia. The dam was constructed between 2004 and 2011 for the primary purpose of hydroelectric power generation.
Background
Between 1974 and 1976, hydrological studies were carried out on the Porce River and between 1982 and 1983, studies regarding the river's hydroelectric development were completed. The study recommended the five projects, Porce I, Porce II, Porce III, Porce IV and Ermitaño. In December 1984, the feasibility report for Porce III was submitted and complementary studies were carried out between 1986 and 1996. In 2002, the design and consultancy contracts were awarded along with the environmental license issued. In 2004, construction on the dam began and the river was diverted by 2007. By 2010, the dam began to impound the reservoir and was complete by 2010. Between 2011, all four generators were commissioned.
|
Where is the Porce III Dam
|
The Porce III Dam is an embankment dam on the Porce River 90 kilometres (56 mi) northeast of Medellín in Antioquia Department, Colombia. The dam was constructed between 2004 and 2011 for the primary purpose of hydroelectric power generation.
|
null | false
| null |
The leading oil-producing area in the United States in 2019 was Texas (5.07 million barrels (806,000 m3) per day), followed by the offshore federal zone of the Gulf of Mexico (1.90 million barrels (302,000 m3) per day), North Dakota (1.42 million barrels (226,000 m3) per day) and New Mexico (0.90 million barrels (143,000 m3) per day). In 2020, the top five U.S. oil producing states were Texas (43%), North Dakota (10.4%), New Mexico (9.2%), Oklahoma (4.1%), and Colorado (4.0%).
|
Which states produced more than 10% of total US oil production in 2020?
|
Texas and North Dakota produced 43% and 10.4% of US oil in 2020, respectively.
|
null | false
| null |
Classify each of the following as either a typical cat name or dog name: Spot, Whiskers, Mittens, Buddy, Fido, Rover, Lassie, Rex, Felix, Socks, Tiger
|
Typical cat names: Whiskers, Mittens, Felix, Socks, Tiger
Typical dog names: Spot, Buddy, Fido, Rover, Lassie, Rex
|
|
null | false
| null |
What is the holographic principle?
|
The holographic principle is an axiom in string theories and a supposed property of quantum gravity that states that the description of a volume of space can be thought of as encoded on a lower-dimensional boundary to the region — such as a light-like boundary like a gravitational horizon.
The holographic principle states that the entropy of ordinary mass (not just black holes) is also proportional to surface area and not volume; that volume itself is illusory and the universe is really a hologram which is isomorphic to the information "inscribed" on the surface of its boundary.
First proposed by Gerard 't Hooft, it was given a precise string-theory interpretation by Leonard Susskind, who combined his ideas with previous ones of 't Hooft and Charles Thorn. Leonard Susskind said, “The three-dimensional world of ordinary experience––the universe filled with galaxies, stars, planets, houses, boulders, and people––is a hologram, an image of reality coded on a distant two-dimensional surface." As pointed out by Raphael Bousso, Thorn observed in 1978 that string theory admits a lower-dimensional description in which gravity emerges from it in what would now be called a holographic way. The prime example of holography is the AdS/CFT correspondence.
|
|
null | false
| null |
Alia Bhatt (/ˈɑːliə ˈbʌt/; born 15 March 1993) is a British actress of Indian descent who predominantly works in Hindi films. She has received several accolades including four Filmfare Awards. One of India's highest-paid actresses, she has appeared in Forbes India's Celebrity 100 list since 2014 and was awarded the TIME100 Impact Award in 2022.
Born into the Bhatt family, she is the daughter of filmmaker Mahesh Bhatt and actress Soni Razdan. After making her acting debut as a child in the 1999 thriller Sangharsh, she played her first leading role in Karan Johar's teen film Student of the Year (2012). She won the Filmfare Critics Award for Best Actress for playing a kidnapping victim in the road drama Highway (2014), and went on to establish herself with starring roles in several romantic films produced by Johar's studio Dharma Productions as well as the coming-of-age drama Dear Zindagi (2016).
|
When did Alia Bhatt get her first leading role?
|
Alia Bhatt received her first leading role in 2012, in the teen film Student of the Year.
|
null | false
| null |
Marvin's Room is a 1996 American drama film directed by Jerry Zaks. The script was written by John Guare and based on the play of the same name by Scott McPherson, who died in 1992. McPherson had completed a screenplay for a film version before he died; however Guare was hired to update it when the film eventually started production years later.
It stars Meryl Streep, Leonardo DiCaprio, Diane Keaton, Robert De Niro, Hume Cronyn, Gwen Verdon, Hal Scardino and Dan Hedaya. Original music for the film was composed by Rachel Portman. Carly Simon wrote and performed the theme song "Two Little Sisters", with Meryl Streep adding background vocals.
|
Who composed the theme song for the movie Marvin's Room?
|
The musician that composed the original music for the Marvin's room was Rachel Portoman, and Carly Simon wrote and performed the song.
|
1810.06743
| false
| null |
FLOAT SELECTED: Table 3: Token-level recall when converting Universal Dependencies tags to UniMorph tags. CSV refers to the lookup-based system. Post-editing refers to the proposed method.
FLOAT SELECTED: Table 3: Token-level recall when converting Universal Dependencies tags to UniMorph tags. CSV refers to the lookup-based system. Post-editing refers to the proposed method.
|
Which languages do they validate on?
|
Ar, Bg, Ca, Cs, Da, De, En, Es, Eu, Fa, Fi, Fr, Ga, He, Hi, Hu, It, La, Lt, Lv, Nb, Nl, Nn, PL, Pt, Ro, Ru, Sl, Sv, Tr, Uk, Ur
|
null | false
| null |
Tell me what sport these players play: Kris Bryant, Zach Lavine, Justin Fields
|
Kris Bryant is a baseball player for the Colorado Rockies, Zach Lavine is a basketball player for the Chicago Bulls, and Justin Fields plays football for the NFL, Chicago Bears.
|
|
1910.10408
| false
| null |
Our experiments are run using the English$\rightarrow $Italian/German portions of the MuST-C corpus BIBREF25, which is extracted from TED talks, using the same train/validation/test split as provided with the corpus (see Table TABREF18). As additional data, we use a mix of public and proprietary data for about 16 million sentence pairs for English-Italian (En-It) and $4.4$ million WMT14 sentence pairs for the English-German (En-De). While our main goal is to verify our hypotheses on a large data condition, thus the need to include proprietary data, for the sake of reproducibility in both languages we also provide results with systems only trained on TED Talks (small data condition). When training on large scale data we use Transformer with layer size of 1024, hidden size of 4096 on feed forward layers, 16 heads in the multi-head attention, and 6 layers in both encoder and decoder. When training only on TED talks, we set layer size of 512, hidden size of 2048 for the feed forward layers, multi-head attention with 8 heads and again 6 layers in both encoder and decoder.
Our experiments are run using the English$\rightarrow $Italian/German portions of the MuST-C corpus BIBREF25, which is extracted from TED talks, using the same train/validation/test split as provided with the corpus (see Table TABREF18). As additional data, we use a mix of public and proprietary data for about 16 million sentence pairs for English-Italian (En-It) and $4.4$ million WMT14 sentence pairs for the English-German (En-De).
|
What dataset do they use for experiments?
|
The answers are shown as follows:
* English$\rightarrow $Italian/German portions of the MuST-C corpus
* As additional data, we use a mix of public and proprietary data for about 16 million sentence pairs for English-Italian (En-It) and $4.4$ million WMT14 sentence pairs for the English-German (En-De)
|
null | false
| null |
Give me a list of the 10 closest airports to San Francisco ?
|
- San Francisco International Airport (SFO)
- Oakland Internaltional Airport
- San Carlos Airport
- Hayward Executive Airport
- Half Moon Bay Airport
- San Rafael Airport
- Paolo Alto Airport
- San Jose Mineta International Airport
- Buchanan Field Airport
- Reid-Hilview County Airport
|
|
null | false
| null |
Categorize each of the following fundamental particles as either interacting via the strong force, the electromagnetic force, the weak force, or the gravitation force: a neutrino, an electron, a proton, a neutron.
|
A neutrino interacts through the weak force.
An electron interacts through the electromagnetic force and the gravitational force.
A proton interacts via the strong force , the electromagnetic force and the gravitational force.
A neutron interacts via the strong force and the gravitational force.
|
|
null | false
| null |
How can I compose electronic music in 3 simple ways ?
|
1 - Use an artificial intelligence backed music application (i.e. Amper Music, MuseNet, AIVA)
2 - Use a software application to compose instruments partitions (i.e. FruityLoops Studio)
3 - Play and record single instruments at home or studio and the mix them using an application
|
|
null | false
| null |
Where was the 7th edition of Slinex conducted?
|
Trincomalee, Sri Lanka
|
|
null | false
| null |
What are the different ways to work with Subject Matter Experts to build a presentation?
|
Clarify the goal of the presentation.
Determine a clear call to action for the audience.
Answer the question, "why will this call to action benefit the audience?"
Focus on message before building slides.
Focus on the flow of the presentation when reviewing and editing.
Pay attention to transitions.
Help the subject matter expert build slides.
Develop a point of view on the content and make suggestions based on that point of view
and your knowledge of what the audience needs.
Develop a script.
Practice the presentation. Edit the script after practicing when you see improvements.
|
|
null | false
| null |
Come up with a marriage proposal in a holiday at Sicily.
|
In the summer breeze of southern Italy waiting for sunset at the beach. Sipping through Italian wine warmed by the yellow sun, the guy brings out a ring along with a bouquet of red roses. This takes the heart out of the girl, she feels like she is melting and accepts the offer.
|
|
null | false
| null |
Explain the use of screens in basketball
|
In basketball, when a team has the ball, one person may set a screen by blocking a defender with his body but making sure his hands are not extended. This allows the person with the ball to pass, shoot, etc. without issuing any fouls. If the person setting a screen moves while screening, the referee will call a moving screen and the ball will be inbounded by the person who moved while screening from the half court line. In any case, screens do not foul the screener for contact unless they are to do physical contact with their hands or feet.
|
|
null | false
| null |
In general, Rolex has three watch lines: Oyster Perpetual, Professional and Cellini (the Cellini line is Rolex's line of "dress" watches). The primary bracelets for the Oyster line are named Jubilee, Oyster, President, and Pearlmaster. The watch straps on the models are usually either stainless steel, yellow gold, white gold, or rose gold. In the United Kingdom, the retail price for the stainless steel 'Pilots' range (such as the GMT Master II) starts from GBP 5,600. Diamond inlay watches are more expensive. The book Vintage Wristwatches by Antiques Roadshow's Reyne Haines listed a price estimate of vintage Rolex watches that ranged between US$650 and US$75,000, while listing vintage Tudors between US$250 and US$9,000.
|
Extract the three names of Rolex watch lines and the typical watch strap types. Arrange them alphabetically and separate them with a comma.
|
The three Rolex watch lines are Cellini, Oyster Perpetual, and Professional. The watch strap types are rose gold, stainless steel, white gold, and yellow gold.
|
null | false
| null |
Who was Charles Guillaume Loys de Bochat
|
Charles Guillaume Loys de Bochat (born 11 December 1695 in Lausanne, died 4 April 1754, also in Lausanne) was an 18th-century Swiss jurist and antiquarian (Lausanne at the time was a subject territory controlled by Berne).
He studied theology in Basle, which he interrupted for health reasons, and later changed his subject to law, in which he graduated in 1717. He became professor of law in Lausanne in 1718, but he was granted leave to travel for three years, which he spent in Halle, in Leyden and in France.
From 1721, he taught law at the Lausanne Academy, where he acted as rector from 1727 to 1730. In 1738, he proposed to transform the academy into a full university, without success.
Loys de Bochat is best known for his major work, Mémoires critiques pour servir d'Eclaircissemens sur divers points de l'Histoire ancienne de la Suisse, which appeared in three volumes in 1747-1749. This work is dedicated to examining the early history of Switzerland, especially the Gaulish Helvetii, and their legacy in Swiss toponymy.
|
|
null | false
| null |
What are different channels of communication to push information out to a sales organization?
|
Email (single topic)
Weekly bulletin with multiple topics. Email or print.
Slack channel
Short video
Regular cadence of webinars (e.g. weekly, monthly, etc.)
Team meetings
Structured agendas for team meetings
Conference calls (phone only)
Phone trees
Group text
Websites
Website with notifications
Carrier pigeon
Ravens (if you're in Game of Thrones)
|
|
null | false
| null |
Franz Kafka (3 July 1883 – 3 June 1924) was a German-speaking Bohemian novelist and short-story writer based in Prague, who is widely regarded as one of the major figures of 20th-century literature. His work fuses elements of realism and the fantastic. It typically features isolated protagonists facing bizarre or surrealistic predicaments and incomprehensible socio-bureaucratic powers. It has been interpreted as exploring themes of alienation, existential anxiety, guilt, and absurdity. His best known works include the short story "The Metamorphosis" and novels The Trial and The Castle. The term Kafkaesque has entered English to describe absurd situations, like those depicted in his writing.
|
Who is Franz Kafka?
|
Franz Kafka was a German-speaking author who has had a major impact in literature. Common themes in his writing include realism and the fantastic. HIs most famous works are "The Metamorphosis", "The Trial", and "The Castle". The term "Kafkaesque" has been coined based on his writings.
|
null | false
| null |
What's the difference between a knot, a hitch and a bend?
|
A knot stands by itself and will not come undone, an example is the figure 8 knot. A hitch, such as the clove hitch, requires something else such as a rope, a tree, or a carabiner to stay intact. A bend joins two ropes together.
|
|
null | false
| null |
Franz Kafka (3 July 1883 – 3 June 1924) was a German-speaking Bohemian novelist and short-story writer based in Prague, who is widely regarded as one of the major figures of 20th-century literature. His work fuses elements of realism and the fantastic. It typically features isolated protagonists facing bizarre or surrealistic predicaments and incomprehensible socio-bureaucratic powers. It has been interpreted as exploring themes of alienation, existential anxiety, guilt, and absurdity. His best known works include the short story "The Metamorphosis" and novels The Trial and The Castle. The term Kafkaesque has entered English to describe absurd situations, like those depicted in his writing.
|
When was Franz Kafka born and who was he?
|
Franz Kafka was born on 3 July 1883, and was a German-speaking Bohemian novelist and short-story writer based in Prague.
|
null | false
| 144
|
To verify the proposed approach, we conducted a pilot study on Amazon Mechanical Turk using data from TAC2008 BIBREF36 . We collected importance estimates for 474 propositions extracted from the first three clusters using both task designs. Each Likert-scale task was assigned to 5 different workers and awarded $0.06. For comparison tasks, we also collected 5 labels each, paid $0.05 and sampled around 7% of all possible pairs. We submitted them in batches of 100 pairs and selected pairs for subsequent batches based on the confidence of the TrueSkill model.
Following the observations of Lloret.2013, we established several measures for quality control. First, we restricted our tasks to workers from the US with an approval rate of at least 95%. Second, we identified low quality workers by measuring the correlation of each worker's Likert-scores with the average of the other four scores. The worst workers (at most 5% of all labels) were removed.
In addition, we included trap sentences, similar as in BIBREF13 , in around 80 of the tasks. In contrast to Lloret et al.'s findings, both an obvious trap sentence (This sentence is not important) and a less obvious but unimportant one (Barack Obama graduated from Harvard Law) were consistently labeled as unimportant (1.08 and 1.14), indicating that the workers did the task properly.
For Likert-scale tasks, we follow Snow.2008 and calculate agreement as the average Pearson correlation of a worker's Likert-score with the average score of the remaining workers. This measure is less strict than exact label agreement and can account for close labels and high- or low-scoring workers. We observe a correlation of 0.81, indicating substantial agreement. For comparisons, the majority agreement is 0.73. To further examine the reliability of the collected data, we followed the approach of Kiritchenko.2016 and simply repeated the crowdsourcing for one of the three topics. Between the importance estimates calculated from the first and second run, we found a Pearson correlation of 0.82 (Spearman 0.78) for Likert-scale tasks and 0.69 (Spearman 0.66) for comparison tasks. This shows that the approach, despite the subjectiveness of the task, allows us to collect reliable annotations.
In addition to the reliability studies, we extrinsically evaluated the annotations in the task of summary evaluation. For each of the 58 peer summaries in TAC2008, we calculated a score as the sum of the importance estimates of the propositions it contains. Table TABREF13 shows how these peer scores, averaged over the three topics, correlate with the manual responsiveness scores assigned during TAC in comparison to ROUGE-2 and Pyramid scores. The results demonstrate that with both task designs, we obtain importance annotations that are similarly useful for summary evaluation as pyramid annotations or gold-standard summaries (used for ROUGE).
Based on the pilot study, we conclude that the proposed crowdsourcing scheme allows us to obtain proper importance annotations for propositions. As workers are not required to read all documents, the annotation is much more efficient and scalable as with traditional methods.
To verify the proposed approach, we conducted a pilot study on Amazon Mechanical Turk using data from TAC2008 (Dang and Owczarzak, 2008).
|
What platform do they use to conduct their pilot study?
|
Amazon Mechanical Turk.
|
null | false
| null |
Give me a list of the Western States in the United States of America.
|
The eleven Western United States are Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming.
|
|
null | false
| null |
Fembe, or Agala, is a Trans–New Guinea language of New Guinea, spoken in the plains east of the Strickland River.
|
From the passage provided, extract where the Fembe language is spoken.
|
Strickland River
|
null | false
| null |
Slater is also a fundraiser and spokesperson for suicide prevention awareness. He has surfed in celebrity events for Surfers Against Suicide, telling sports website 'Athletes Talk': "I've lost a couple of friends myself to suicide and it's just a horrible thing that can be prevented. People get in this dark place and they don't know what to do so it's always nice to see a non-profit that isn't turning into anything else other than just trying to help people."
Slater is passionate about preserving oceans globally and protecting temperate reefs in California through his relationship with Reef Check. Slater is on the Board of Advisors (the Ocean Advocacy Advisory Board) of ocean conservation organization Sea Shepherd Conservation Society.
|
Based on the paragraph below, please list out the 2 major causes supported by Kelly Slater.
|
Kelly Slater's main activism has been directed towards suicide prevention and ocean conservation.
|
null | false
| 388
|
The current coverage of the political landscape in the press and in social media has led to an unprecedented situation. Like never before, a statement in an interview, a press release, a blog note, or a tweet can spread almost instantaneously and reach the public in no time. This proliferation speed has left little time for double-checking claims against the facts, which has proven critical in politics, e.g., during the 2016 presidential campaign in the USA, which was arguably impacted by fake news in social media and by false claims.
Investigative journalists and volunteers have been working hard trying to get to the root of a claim and to present solid evidence in favor or against it. Manual fact-checking has proven very time-consuming, and thus automatic methods have been proposed as a way to speed-up the process. For instance, there has been work on checking the factuality/credibility of a claim, of a news article, or of an information source BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7. However, less attention has been paid to other steps of the fact-checking pipeline, which is shown in Figure FIGREF1.
The process starts when a document is made public. First, an intrinsic analysis is carried out in which check-worthy text fragments are identified. Then, other documents that might support or rebut a claim in the document are retrieved from various sources. Finally, by comparing a claim against the retrieved evidence, a system can determine whether the claim is likely true or likely false. For instance, BIBREF8 do this on the basis of a knowledge graph derived from Wikipedia. The outcome could then be presented to a human expert for final judgment.
In this paper, we focus on the first step: predicting check-worthiness of claims. Our contributions can be summarized as follows:
New dataset: We build a new dataset of manually-annotated claims, extracted from the 2016 US presidential and vice-presidential debates, which we gathered from nine reputable sources such as CNN, NPR, and PolitiFact, and which we release to the research community.
Modeling the context: We develop a novel approach for automatically predicting which claims should be prioritized for fact-checking, based on a rich input representation. In particular, we model not only the textual content, but also the context: how the target claim relates to the current segment, to neighboring segments and sentences, and to the debate as a whole, and also how the opponents and the public react to it.
State-of-the-art results: We achieve state-of-the-art results, outperforming a strong rivaling system by a margin, while also demonstrating that this improvement is due primarily to our modeling of the context.
We model the problem as a ranking task, and we train both Support Vector Machines (SVM) and Feed-forward Neural Networks (FNN) obtaining state-of-the-art results. We also analyze the relevance of the specific feature groups and we show that modeling the context yields a significant boost in performance. Finally, we also analyze whether we can learn to predict which facts are check-worthy with respect to each of the individual media sources, thus capturing their biases. It is worth noting that while trained on political debates, many features of our model can be potentially applied to other kinds of information sources, e.g., interviews and news.
The rest of the paper is organized as follows: Section SECREF2 discusses related work. Section SECREF3 describes the process of gathering and annotating our political debates dataset. Section SECREF4 presents our supervised approach to predicting fact-checking worthiness, including the explanation of the model and the information sources we use. Section SECREF5 presents the evaluation setup and discusses the results. Section SECREF6 provides further analysis. Finally, Section SECREF7 presents the conclusions and outlines some possible directions for future research.
We develop a novel approach for automatically predicting which claims should be prioritized for factchecking, based on a rich input representation.
|
What did the novel approach automatically predict?
|
It predicts which claims should be prioritized for fact-checking.
|
null | false
| null |
Classify each of the following as "Vegetarian Friendly" or "Not Vegetarian Friendly": chicken, peanuts, eggs, bacon, sausage, lettuce, ketchup
|
Chicken: Not Vegetarian Friendly
Peanuts: Vegetarian Friendly
Eggs: Vegetarian Friendly
Bacon: Not Vegetarian Friendly
Sausage: Not Vegetarian Friendly
Lettuce: Vegetarian Friendly
Ketchup: Vegetarian Friendly
|
|
null | false
| null |
Classify the cities based on whether they are located in Belarus or Cayman Islands.
Vitsyebsk, George Town, Minsk, Homyel’
|
Belarus - Minsk, Homyel’, Vitsyebsk
Cayman Islands - George Town
|
|
null | false
| null |
St. John the Baptist, Blackrock is a Roman Catholic church in the parish of Blackrock, Ireland. The church is still in use and named after the Saint John the Baptist. It is located on Temple Road, Blackrock, County Dublin.
The church was designed by the architect Patrick Byrne, who was educated at the Dublin Society Schools. It is one of the finest examples of the Gothic revival style in Ireland and was the first to be built in the Dublin Archdiocese. It is said to be inspired by the ideas of Augustus Welby Pugin (1812-1852).
The initial building consisted of the nave, chancel and bell tower. To the rear of the church, two carved heads can be seen up high on either side of the rose window. The head on the left depicts St. John the Baptist, while the head on the right depicts the Archbishop Daniel Murray.
|
Who designed the St. John the Baptist, Blackrock church?
|
Patrick Byrne
|
null | false
| 135
|
Ambiguity and implicitness are inherent properties of natural language that cause challenges for computational models of language understanding. In everyday communication, people assume a shared common ground which forms a basis for efficiently resolving ambiguities and for inferring implicit information. Thus, recoverable information is often left unmentioned or underspecified. Such information may include encyclopedic and commonsense knowledge. This work focuses on commonsense knowledge about everyday activities, so-called scripts.
This paper introduces a dataset to evaluate natural language understanding approaches with a focus on interpretation processes requiring inference based on commonsense knowledge. In particular, we present MCScript, a dataset for assessing the contribution of script knowledge to machine comprehension. Scripts are sequences of events describing stereotypical human activities (also called scenarios), for example baking a cake or taking a bus BIBREF0 . To illustrate the importance of script knowledge, consider Example ( SECREF1 ):
Without using commonsense knowledge, it may be difficult to tell who ate the food: Rachel or the waitress. In contrast, if we utilize commonsense knowledge, in particular, script knowledge about the eating in a restaurant scenario, we can make the following inferences: Rachel is most likely a customer, since she received an order. It is usually the customer, and not the waitress, who eats the ordered food. So She most likely refers to Rachel.
Various approaches for script knowledge extraction and processing have been proposed in recent years. However, systems have been evaluated for specific aspects of script knowledge only, such as event ordering BIBREF1 , BIBREF2 , event paraphrasing BIBREF3 , BIBREF4 or event prediction (namely, the narrative cloze task BIBREF5 , BIBREF6 , BIBREF7 , BIBREF8 , BIBREF9 ). These evaluation methods lack a clear connection to real-world tasks. Our MCScript dataset provides an extrinsic evaluation framework, based on text comprehension involving commonsense knowledge. This framework makes it possible to assess system performance in a multiple-choice question answering setting, without imposing any specific structural or methodical requirements.
MCScript is a collection of (1) narrative texts, (2) questions of various types referring to these texts, and (3) pairs of answer candidates for each question. It comprises approx. 2,100 texts and a total of approx. 14,000 questions. Answering a substantial subset of questions requires knowledge beyond the facts mentioned in the text, i.e. it requires inference using commonsense knowledge about everyday activities. An example is given in Figure FIGREF2 . For both questions, the correct choice for an answer requires commonsense knowledge about the activity of planting a tree, which goes beyond what is mentioned in the text. Texts, questions, and answers were obtained through crowdsourcing. In order to ensure high quality, we manually validated and filtered the dataset. Due to our design of the data acquisition process, we ended up with a substantial subset of questions that require commonsense inference (27.4%).
However, systems have been evaluated for specific aspects of script knowledge only, such as event ordering (Modi and Titov, 2014a; Modi and Titov, 2014b), event paraphrasing (Reg- neri et al., 2010; Wanzare et al., 2017) or event prediction (namely, the narrative cloze task (Chambers and Jurafsky,2008; Chambers and Jurafsky, 2009; Pichotta and Mooney, 2014; Pichotta and Mooney, 2016; Modi, 2016)). These evaluation methods lack a clear connection to real-world tasks.
|
What are the disadvantages of the previous approaches?
|
However, systems have been evaluated for specific aspects of script knowledge only, these evaluation methods lack a clear connection to real-world tasks.
|
null | false
| null |
What towns make up Cape Ann?
|
Cape Ann is a rocky peninsula in northeastern Massachusetts on the Atlantic Ocean. It is about 30 miles (48 km) northeast of Boston and marks the northern limit of Massachusetts Bay. Cape Ann includes the city of Gloucester and the towns of Essex, Manchester-by-the-Sea and Rockport.
|
|
null | false
| 85
|
We compare our approach to two other, the first one uses almost the same tweet data as we use for training, and the second one is the CrowdFlower dataset annotated for emotions.
In the first one Wang et al. BIBREF21 downloaded over 5M tweets which included one of 131 emotional hashtags based on Parrott's three-level categorization of emotions in seven categories: joy, sadness, anger, love, fear, thankfulness, surprise. To assess the quality of using hashtags as labels, the sampled 400 tweets randomly and after comparing human annotations by hashtag labels they came up with simple heuristics to increase the quality of labeling by ignoring tweets with quotations and URLs and only keeping tweets with 5 terms or more that have the emotional hashtags at the end of the tweets. Using these rules they extracted around 2.5M tweets. After sampling another 400 random tweets and comparing it to human annotation the saw that hashtags can classify the tweets with 95% precision. They did some pre-processing by making all words lower-case, replaced user mentions with @user, replaced letters/punctuation that is repeated more than twice with the same two letters/punctuation (e.g., ooooh INLINEFORM0 ooh, !!!!! INLINEFORM1 !!); normalized some frequently used informal expressions (e.g., ll → will, dnt INLINEFORM2 do not); and stripped hash symbols. They used a sub-sample of their dataset to figure out the best approaches for classification, and after trying two different classifiers (multinomial Naive Bayes and LIBLINEAR) and 12 different feature sets, they got their best results using logistic regression branch for LIBLINEAR classifier and a feature set consist of n-gram(n=1,2), LIWC and MPQA lexicons, WordNet-Affect and POS tags.
In the second one, the reported results are from a paper by BIBREF33 in which they used maximum entropy classifier with bag of words model to classify various emotional datasets. Here we only report part of their result for CrowdFlower dataset that can be mapped to one of our seven labels.
We compare our approach to two other, the first one uses almost the same tweet data as we use for training, and the second one is the CrowdFlower dataset annotated for emotions.
|
How does the author compare his method with the other two?
|
The first one uses almost the same tweet data as they use for training, and the second one is the CrowdFlower dataset annotated for emotions.
|
null | false
| null |
What should I do if my laptop has been hacked?
|
If your laptop has been hacked, you can be in a big trouble. Attackers may have stolen all your confidential data and passwords. It is better to ask for helps from security experts. It is also important to reset all your passwords. Do not turn on your laptop and if it is a personal laptop make sure to re-install your operating system. Check your recent bank account activities to ensure that you have not lost any money.
|
|
null | false
| 122
|
Most of the existing works on propaganda detection focus on identifying propaganda at the news article level, or even at the news outlet level with the assumption that each of the articles of the suspected propagandistic outlet are propaganda BIBREF5, BIBREF6.
Here we study two tasks that are more fine-grained, specifically propaganda detection at the sentence and phrase (fragment) levels BIBREF4. This fine-grained setup aims to train models that identify linguistic propaganda techniques rather than distinguishing between the article source styles.
BIBREF4 EMNLP19DaSanMartino were the first to propose this problem setup and release it as a shared task. Along with the released dataset, BIBREF4 proposed a multi-granularity neural network, which uses the deep bidirectional transformer architecture known as BERT, which features pre-trained context-dependent embeddings BIBREF0. Their system takes a joint learning approach to the sentence- and phrase-level tasks, concatenating the output representation of the less granular (sentence-level) task with the more fine-grained task using learned weights.
In this work we also take the BERT model as the basis of our approach and focus on the class imbalance as well as the lack of similarity between training and test data inherent to the task.
It is worth reflecting on the nature of the shared task dataset (PTC corpus) and its structural correspondence (or lack thereof) to some of the definitions of propaganda mentioned in the introduction. First, propaganda is a social phenomenon and takes place as an act of communication BIBREF19, and so it is more than a simple information-theoretic message of zeros and ones—it also incorporates an addresser and addressee(s), each in phatic contact (typically via broadcast media), ideally with a shared denotational code and contextual surround(s) BIBREF20.
As such, a dataset of decontextualised documents with labelled sentences, devoid of authorial or publisher metadata, has taken us at some remove from even a simple everyday definition of propaganda. Our models for this shared task cannot easily incorporate information about the addresser or addressee; are left to assume a shared denotational code between author and reader (one perhaps simulated with the use of pre-trained word embeddings); and they are unaware of when or where the act(s) of propagandistic communication took place. This slipperiness is illustrated in our example document (Fig. FIGREF13): note that while Sentences 3 and 7, labelled as propaganda, reflect a propagandistic attitude on the part of the journalist and/or publisher, Sentence 4—also labelled as propaganda in the training data—instead reflects a “flag-waving" propagandistic attitude on the part of U.S. congressman Jeff Flake, via the conventions of reported speech BIBREF21. While reported speech often is signaled by specific morphosyntactic patterns (e.g. the use of double-quotes and “Flake said") BIBREF22, we argue that human readers routinely distinguish propagandistic reportage from the propagandastic speech acts of its subjects, and to conflate these categories in a propaganda detection corpus may contribute to the occurrence of false positives/negatives.
Most of the existing works on propaganda detection focus on identifying propaganda at the news article level, or even at the news outlet level with the assumption that each of the articles of the suspected propagandistic outlet are propaganda (Rashkin etal., 2017; Barron-Cedeno etal., 2019). Here we study two tasks that are more fine-grained, specifically propaganda detection at the sentence and phrase (fragment) levels (Da San Martino et al., 2019). This fine-grained setup aims to train models that identify linguistic propaganda techniques rather than distinguishing between the article source styles. Da San Martino et al. (2019) were the first to propose this problem setup and release it as a shared task. Along with the released dataset, Da San Martino et al. (2019) proposed a multi-granularity neural network, which uses the deep bidirectional transformer architecture known as BERT, which features pre-trained contextdependent embeddings (Devlin et al., 2018). Their system takes a joint learning approach to the sentence- and phrase-level tasks, concatenating the output representation of the less granular (sentence-level) task with the more fine-grained task using learned weights.
|
What is the recent study of propaganda detection prior to this paper?
|
Previously most of the existing works on propaganda detection focus on identifying propaganda at the news article level, or even at the news outlet level with the assumption that each of the articles of the suspected propagandistic outlet are propaganda (Rashkin etal., 2017; Barron-Cedeno etal., 2019). Later two tasks that are more fine-grained came up, at the sentence and phrase (fragment) levels (Da San Martino et al., 2019). Also Da San Martino et al. (2019) were the first to propose this problem setup and release it as a shared task.
|
null | false
| null |
Give me a bulleted list of the last 10 winners of The Masters in order starting from 2022
|
- Scottie Scheffler
- Hideki Matsuyama
- Dustin Johnson
- Tiger Woods
- Patrick Reed
- Sergio Garcia
- Jordan Spieth
- Bubba Watson
- Adam Scott
|
|
null | false
| null |
Are the following colours considered primary colours or not: red, orange, yellow, green, blue, indigo, violet
|
red - primary
orange - not primary
yellow - primary
green - not primary
blue - primary
indigo - not primary
violet - not primary
|
|
null | false
| null |
According to Greek mythology, who was the first woman on earth?
|
Pandora
|
|
null | false
| null |
The 8086 (also called iAPX 86) is a 16-bit microprocessor chip designed by Intel between early 1976 and June 8, 1978, when it was released. The Intel 8088, released July 1, 1979, is a slightly modified chip with an external 8-bit data bus (allowing the use of cheaper and fewer supporting ICs),[note 1] and is notable as the processor used in the original IBM PC design.
|
When was the 8088 processor released?
|
The Intel 8088 processor was released July 1, 1979.
|
null | false
| null |
Name a few famous Dutch artists, and the respective movements for which they are known.
|
1. Rembrandt - Baroque, or the Dutch Golden Age. Known for his masterful use of light and shadow, and his expressive portraits.
2. Vermeer - Baroque, or the Dutch Golden Age. Known for the interplay of colors and light, and his depictions of ordinary domestic scenes.
3. Van Gogh - Post-Impression. Known for his vibrant and bold colors, and emotional portraits of daily life.
4. Mondrian - De Stijl. A pioneer of the De Stijl movement, which is known for its strict use of geometry and colors, he is well known for his abstract grid paintings.
|
|
null | false
| null |
Solitary bees, such as leafcutters, do not form colonies. Unlike social insects (ants, yellow jackets, honeybees), leafcutters work alone building isolated nests. Similar to honeybees, female bees perform nearly all essential tasks of brood rearing. These native insects perform essential tasks, pollinating wild plants. The alfalfa leaf cutter bee (Megachile rotundata), native to Europe, has been semi-domesticated for crop pollination. In North America, the species was deliberately imported to assist in the pollination of food crops, but has now become feral and widespread.
|
What are the names of the social insects that are mentioned?
|
ants, yellow jackets, honeybees
|
null | false
| null |
What criteria should I be looking at when buying a new motorcycle?
|
There are a number of criteria you should be looking at:
- Price: How does it fit in your budget?
- Engine type: Do you want a traditional combustion engine, or go for an electric one?
- Number of cylinders: If you go for combustion engine, how many cylinders do you choose? Less cylinders are usually cheaper, make the motorcycle more lightweight and thus easier to maneuver. More cylinders make the engine feel smoother, also have more power, but are heavier and more expensive.
- Usage: What do you use it for? Just for fun, commuting to work, general transportation? Travel? Offroad? Racing?
- Assistance systems: Modern motorcycles have a range of assistance systems like ABS, Cruise Control, Distance sensors, electronic driving modes for different weather conditions etc. They can make driving safer , but also add weight and make the bike more expensive. Which ones do you need?
- Style: There are different styles like naked bike, cruiser, chopper, dual sport, race bike , scrambler etc. Different styles can be more or less adequate depending on your usage, or you can simply pick the one that appeals to you.
- Size: The two most important aspects in terms of size are seat height and weight. Both have a huge impact on how well you are in control when not driving, for example when you need to stop at a traffic light or going slowly in stop & go traffic.
Seat height is important , so your feet can reach the ground when stopping. However the width of the seat and general width dimensions of the bike also play a role here. So two bikes with same seat height might feel very differently. Best is to try it out
|
|
null | false
| null |
Identify which instrument is string or percussion: Metallophone, Doshpuluur
|
Doshpuluur is string, Metallophone is percussion.
|
|
null | false
| 162
|
BIBREF13 proposed a MP framework under which many of the recently introduced GNNs can be reformulated. MP consists in an aggregation phase followed by a combination phase BIBREF14. More precisely, let $G(V,E)$ be a graph, and let us consider $v \in V$. At time $t+1$, a message vector $\mathbf {m}_v^{t+1}$ is computed from the representations of the neighbors $\mathcal {N}(v)$ of $v$:
The new representation $\mathbf {h}^{t+1}_v$ of $v$ is then computed by combining its current feature vector $\mathbf {h}^{t}_v$ with the message vector $\mathbf {m}_v^{t+1}$:
Messages are passed for $T$ time steps. Each step is implemented by a different layer of the MP network. Hence, iterations correspond to network depth. The final feature vector $\mathbf {h}_v^T$ of $v$ is based on messages propagated from all the nodes in the subtree of height $T$ rooted at $v$. It captures both the topology of the neighborhood of $v$ and the distribution of the vertex representations in it.
If a graph-level feature vector is needed, e.g., for classification or regression, a READOUT pooling function, that must be invariant to permutations, is applied:
Next, we present the MP network we developed for document understanding.
We represent a document as a statistical word co-occurrence network BIBREF18, BIBREF19 with a sliding window of size 2 overspanning sentences. Let us denote that graph $G(V,E)$. Each unique word in the preprocessed document is represented by a node in $G$, and an edge is added between two nodes if they are found together in at least one instantiation of the window. $G$ is directed and weighted: edge directions and weights respectively capture text flow and co-occurrence counts.
$G$ is a compact representation of its document. In $G$, immediate neighbors are consecutive words in the same sentence. That is, paths of length 2 correspond to bigrams. Paths of length more than 2 can correspond either to traditional $n$-grams or to relaxed $n$-grams, that is, words that never appear in the same sentence but co-occur with the same word(s). Such nodes are linked through common neighbors.
Master node. Inspired by BIBREF3, our $G$ also includes a special document node, linked to all other nodes via unit weight bi-directional edges. In what follows, let us denote by $n$ the number of nodes in $G$, including the master node.
We formulate our AGGREGATE function as:
where $\mathbf {H}^t \in \mathbb {R}^{n \times d}$ contains node features ($d$ is a hyperparameter), and $\mathbf {A} \in \mathbb {R}^{n \times n}$ is the adjacency matrix of $G$. Since $G$ is directed, $\mathbf {A}$ is asymmetric. Also, $\mathbf {A}$ has zero diagonal as we choose not to consider the feature of the node itself, only that of its incoming neighbors, when updating its representation. Since $G$ is weighted, the $i^{th}$ row of $A$ contains the weights of the edges incoming on node $v_i$. $\mathbf {D} \in \mathbb {R}^{n \times n}$ is the diagonal in-degree matrix of $G$. MLP denotes a multi-layer perceptron, and $\mathbf {M}^{t+1} \in \mathbb {R}^{n \times d}$ is the message matrix.
The use of a MLP was motivated by the observation that for graph classification, MP neural nets with 1-layer perceptrons are inferior to their MLP counterparts BIBREF14. Indeed, 1-layer perceptrons are not universal approximators of multiset functions. Note that like in BIBREF14, we use a different MLP at each layer.
Renormalization. The rows of $\mathbf {D}^{-1}\mathbf {A}$ sum to 1. This is equivalent to the renormalization trick of BIBREF9, but using only the in-degrees. That is, instead of computing a weighted sum of the incoming neighbors' feature vectors, we compute a weighted average of them. The coefficients are proportional to the strength of co-occurrence between words. One should note that by averaging, we lose the ability to distinguish between different neighborhood structures in some special cases, that is, we lose injectivity. Such cases include neighborhoods in which all nodes have the same representations, and neighborhoods of different sizes containing various representations in equal proportions BIBREF14. As suggested by the results of an ablation experiment, averaging is better than summing in our application (see subsection SECREF30). Note that instead of simply summing/averaging, we also tried using GAT-like attention BIBREF11 in early experiments, without obtaining better results.
As far as our COMBINE function, we use the Gated Recurrent Unit BIBREF20, BIBREF21:
Omitting biases for readability, we have:
where the $\mathbf {W}$ and $\mathbf {U}$ matrices are trainable weight matrices not shared across time steps, $\sigma (\mathbf {x}) = 1/(1+\exp (-\mathbf {x}))$ is the sigmoid function, and $\mathbf {R}$ and $\mathbf {Z}$ are the parameters of the reset and update gates. The reset gate controls the amount of information from the previous time step (in $\mathbf {H}^t$) that should propagate to the candidate representations, $\tilde{\mathbf {H}}^{t+1}$. The new representations $\mathbf {H}^{t+1}$ are finally obtained by linearly interpolating between the previous and the candidate ones, using the coefficients returned by the update gate.
Interpretation. Updating node representations through a GRU should in principle allow nodes to encode a combination of local and global signals (low and high values of $t$, resp.), by allowing them to remember about past iterations. In addition, we also explicitly consider node representations at all iterations when reading out (see Eq. DISPLAY_FORM18).
After passing messages and performing updates for $T$ iterations, we obtain a matrix $\mathbf {H}^T \in \mathbb {R}^{n \times d}$ containing the final vertex representations. Let $\hat{G}$ be graph $G$ without the special document node, and matrix $\mathbf {\hat{H}}^T \in \mathbb {R}^{(n-1) \times d}$ be the corresponding representation matrix (i.e., $\mathbf {H}^T$ without the row of the document node).
We use as our READOUT function the concatenation of self-attention applied to $\mathbf {\hat{H}}^T$ with the final document node representation. More precisely, we apply a global self-attention mechanism BIBREF22 to the rows of $\mathbf {\hat{H}}^T$. As shown in Eq. DISPLAY_FORM17, $\mathbf {\hat{H}}^T$ is first passed to a dense layer parameterized by matrix $\mathbf {W}_A^T \in \mathbb {R}^{d \times d}$. An alignment vector $\mathbf {a}$ is then derived by comparing, via dot products, the rows of the output of the dense layer $\mathbf {Y}^T \in \mathbb {R}^{(n-1) \times d}$ with a trainable vector $\mathbf {v}^T \in \mathbb {R}^d$ (initialized randomly) and normalizing with a softmax. The normalized alignment coefficients are finally used to compute the attentional vector $\mathbf {u}^T \in \mathbb {R}^d$ as a weighted sum of the final representations $\mathbf {\hat{H}}^T$.
Note that we tried with multiple context vectors, i.e., with a matrix $\mathbf {V}^T$ instead of a vector $\mathbf {v}^T$, like in BIBREF22, but results were not convincing, even when adding a regularization term to the loss to favor diversity among the rows of $\mathbf {V}^T$.
Master node skip connection. $\mathbf {h}_G^T \in \mathbb {R}^{2d}$ is obtained by concatenating $\mathbf {u}^T$ and the final master node representation. That is, the master node vector bypasses the attention mechanism. This is equivalent to a skip or shortcut connection BIBREF23. The reason behind this choice is that we expect the special document node to learn a high-level summary about the document, such as its size, vocabulary, etc. (more details are given in subsection SECREF30). Therefore, by making the master node bypass the attention layer, we directly inject global information about the document into its final representation.
Multi-readout. BIBREF14, inspired by Jumping Knowledge Networks BIBREF12, recommend to not only use the final representations when performing readout, but also that of the earlier steps. Indeed, as one iterates, node features capture more and more global information. However, retaining more local, intermediary information might be useful too. Thus, instead of applying the readout function only to $t=T$, we apply it to all time steps and concatenate the results, finally obtaining $\mathbf {h}_G \in \mathbb {R}^{T \times 2d}$ :
In effect, with this modification, we take into account features based on information aggregated from subtrees of different heights (from 1 to $T$), corresponding to local and global features.
Through the successive MP iterations, it could be argued that MPAD implicitly captures some soft notion of the hierarchical structure of documents (words $\rightarrow $ bigrams $\rightarrow $ compositions of bigrams, etc.). However, it might be beneficial to explicitly capture document hierarchy. Hierarchical architectures have brought significant improvements to many NLP tasks, such as language modeling and generation BIBREF24, BIBREF25, sentiment and topic classification BIBREF26, BIBREF27, and spoken language understanding BIBREF28, BIBREF29. Inspired by this line of research, we propose several hierarchical variants of MPAD, detailed in what follows. In all of them, we represent each sentence in the document as a word co-occurrence network, and obtain an embedding for it by applying MPAD as previously described.
MPAD-sentence-att. Here, the sentence embeddings are simply combined through self-attention.
MPAD-clique. In this variant, we build a complete graph where each node represents a sentence. We then feed that graph to MPAD, where the feature vectors of the nodes are initialized with the sentence embeddings previously obtained.
MPAD-path. This variant is similar to the clique one, except that instead of a complete graph, we build a path according to the natural flow of the text. That is, two nodes are linked by a directed edge if the two sentences they represent follow each other in the document.
Note that instead of simply summing/averaging, we also tried using GAT-like attention (Velickovic et al. 2017) in early experiments, without obtaining better results.
|
Will GAT-like attention work well with the experiments?
|
No, they tried using GAT-like attention (Velickovic et al. 2017) in early experiments, but they did not obtain better results.
|
null | false
| null |
The Żuk (pl. beetle) was a van and light truck produced in Lublin, Poland, between 1958 and 1998 by FSC. It was based on FSO Warszawa, which in turn was licensed from the Soviet passenger car GAZ-M20 Pobeda. The chassis, suspension and engine from FSO Warszawa formed the basis of the Żuk and the Nysa light vans designed in the late 1950s. About 587818 were manufactured.
The Żuk was mainly sold to state organizations, but also to individuals. After 1989, with the liberalization of the Polish economy, the Żuk was able to maintain sales to the traditional markets and expand the number sold to individual consumers. The final few years of production was in parallel to its successor, the Lublin van, as a cheaper alternative.
|
What kind of car was The Zuk
|
The Zuk (pl. beetle) was a van and light truck produced in Lublin, Poland, between 1958 and 1998 by FSC. It was based on FSO Warszawa, which in turn was licensed from the Soviet passenger car GAZ-M20 Pobeda. The chassis, suspension and engine from FSO Warszawa formed the basis of the Zuk and the Nysa light vans designed in the late 1950s. About 587818 were manufactured.
The Zuk was mainly sold to state organizations, but also to individuals. After 1989, with the liberalization of the Polish economy, the Zuk was able to maintain sales to the traditional markets and expand the number sold to individual consumers. The final few years of production was in parallel to its successor, the Lublin van, as a cheaper alternative.
|
2002.00876
| true
| null |
The library design of Torch-Struct follows the distributions API used by both TensorFlow and PyTorch BIBREF29. For each structured model in the library, we define a conditional random field (CRF) distribution object. From a user's standpoint, this object provides all necessary distributional properties. Given log-potentials (scores) output from a deep network $\ell $, the user can request samples $z \sim \textsc {CRF}(\ell )$, probabilities $\textsc {CRF}(z;\ell )$, modes $\arg \max _z \textsc {CRF}(\ell )$, or other distributional properties such as $\mathbb {H}(\textsc {CRF}(\ell ))$. The library is agnostic to how these are utilized, and when possible, they allow for backpropagation to update the input network. The same distributional object can be used for standard output prediction as for more complex operations like attention or reinforcement learning.
The library design of Torch-Struct follows the distributions API used by both TensorFlow and PyTorch BIBREF29.
|
Does API provide ability to connect to models written in some other deep learning framework?
|
Yes.
|
null | false
| null |
Manis Lamond (born 14 September 1966) is a retired soccer player from Papua New Guinea. An international who played as a striker, Lamond played professionally in Australia for Canberra City, Sydney United, A.P.I.A. Leichhardt Tigers, Wollongong City, Marconi Stallions and Newcastle United and in Singapore for Sembawang Rangers.
|
What position did Manis Lamond play?
|
Manis Lamond was a striker
|
null | false
| 81
|
Writing errors can occur in many different forms – from relatively simple punctuation and determiner errors, to mistakes including word tense and form, incorrect collocations and erroneous idioms. Automatically identifying all of these errors is a challenging task, especially as the amount of available annotated data is very limited. Rei2016 showed that while some error detection algorithms perform better than others, it is additional training data that has the biggest impact on improving performance.
Being able to generate realistic artificial data would allow for any grammatically correct text to be transformed into annotated examples containing writing errors, producing large amounts of additional training examples. Supervised error generation systems would also provide an efficient method for anonymising the source corpus – error statistics from a private corpus can be aggregated and applied to a different target text, obscuring sensitive information in the original examination scripts. However, the task of creating incorrect data is somewhat more difficult than might initially appear – naive methods for error generation can create data that does not resemble natural errors, thereby making downstream systems learn misleading or uninformative patterns.
Previous work on artificial error generation (AEG) has focused on specific error types, such as prepositions and determiners BIBREF0 , BIBREF1 , or noun number errors BIBREF2 . Felice2014a investigated the use of linguistic information when generating artificial data for error correction, but also restricting the approach to only five error types. There has been very limited research on generating artificial data for all types, which is important for general-purpose error detection systems. For example, the error types investigated by Felice2014a cover only 35.74% of all errors present in the CoNLL 2014 training dataset, providing no additional information for the majority of errors.
In this paper, we investigate two supervised approaches for generating all types of artificial errors. We propose a framework for generating errors based on statistical machine translation (SMT), training a model to translate from correct into incorrect sentences. In addition, we describe a method for learning error patterns from an annotated corpus and transplanting them into error-free text. We evaluate the effect of introducing artificial data on two error detection benchmarks. Our results show that each method provides significant improvements over using only the available training set, and a combination of both gives an absolute improvement of 4.3% in INLINEFORM0 , without requiring any additional annotated data.
In addition, we describe a method for learning error patterns from an annotated corpus and transplanting them into error-free text.
|
What is the method that the authors described in this paper?
|
A method for learning error patterns from an annotated corpus and transplanting them into error-free text.
|
null | false
| 80
|
Finally, Figure FIGREF28 shows INLINEFORM0 correlation of different variants of Sera with pyramid based on selection of different cut-off points ( INLINEFORM1 and INLINEFORM2 correlations result in very similar graphs). When the cut-off point increases, more documents are retrieved for the candidate and the gold summaries, and therefore the final Sera score is more fine-grained. A general observation is that as the search cut-off point increases, the correlation with pyramid scores decreases. This is because when the retrieved result list becomes larger, the probability of including less related documents increases which negatively affects correct estimation of the similarity of the candidate and gold summaries. The most accurate estimations are for metrics with cut-off points of 5 and 10 which are included in the reported results of all variants in Table TABREF23 .
Figure 1: ρ correlation of SERA with pyramid based on different cut-off points. The x-axis shows the cut-off point parameter. DIS: Discounted variant of SERA; NP: Query reformulation with Noun Phrases; KW: Query reformulation with Keywords.
|
What parameter does the x-axis represent?
|
The cut-off point parameter.
|
null | false
| null |
Where was the beer Yuengling founded?
|
Pennsylvania
|
|
null | false
| 46
|
In this work, we incorporate a domain discriminator to align the mismatched domains between ASR hypotheses and reference transcriptions. The adversarial learning allows the end-to-end QA model to learn domain-invariant features and improve the robustness to ASR errors. The experiments demonstrate that the proposed model successfully achieves superior performance and outperforms the previous best model by 2% EM score and over 1.5% F1 score.
The experiments demonstrate that the proposed model successfully achieves superior performance and outperforms the previous best model by 2% EM score and over 1.5% F1 score.
|
Does the proposed model outperform the previous best model?
|
Yes, the proposed model outperforms the previous best model by 2% EM score and over 1.5% F1 score.
|
null | false
| null |
How are healthcare organizations thinking about their strategy to improve healthcare delivery?
|
One popular model that many healthcare organizations use to set goals and measure their progress is known as the "Quadruple Aim." This framework looks to achieve four simultaneous and interconnected goals: Improving the patient experience, Improving healthcare outcomes, reducing cost of care, and improving the working experience of healthcare providers.
|
|
null | false
| 158
|
Many machine learning models in question answering tasks often involve matching mechanism. For example, in factoid question answering such as SQuAD BIBREF1 , one needs to match between query and corpus in order to find out the most possible fragment as answer. In multiple choice question answering, such as MC Test BIBREF2 , matching mechanism can also help make the correct decision.
The easiest way of matching is to calculate the cosine similarity between two vectors. It is generally done by two step: First, encode text into word vectors, sentence vectors or paragraph vectors. Second, simply calculate the cosine similarity between target vectors. This method performs well when applied to word-level matching. However, as for matching between sentences or paragraphs, a single vector is not sufficient to encode all the important information. In order to solve this problem, Wang and Jiang proposed a “compare-aggregate” BIBREF3 framework that performs word-level matching using multiple techniques followed by aggregation with convolutional neural network. In their work, they show that compare-aggregate framework can effectively match two sequences through a wide range.
Although "compare-aggregate" matching mechanism performs well on multiple question answering tasks, it has two deficiencies. First, it tends to aggregate passively through the sequence rather than take the importance of each element into account. That is, "compare aggregate" model considers all the sequential contents equally. Second, "compare aggregate" can only take few neighboring elements into account at the same time because of the limitation of CNN kernel size.
In this paper, we propose Query-based Attention CNN (QACNN) to deal with the deficiencies above. First, we add query-based attention mechanism into original "compare aggregate" model. Moreover, We re-design the aggregation mechanism in "compare aggregate" to a two-staged CNN architecture which comprises word-level aggregation and sentence-level aggregation. In this way, QACNN can efficiently extract features cross sentences.
Our model consists of three components: 1) The similarity mapping layer which converts the input passage, query and choice into feature representation and perform a similarity operation to each other. 2) The attention-based CNN matching network composed of a two-staged CNN focusing on word-level and sentence-level matching respectively. 3) The prediction layer which makes the final decision.
The main contributions of this work are three-fold. First, we introduce a two-staged CNN architecture which integrates information from word-level to sentence-level, and then from sentence-level to passage-level. Second, we introduce attention mechanism into this net. We use specially designed CNN structure and attention mechanism to recognize the pattern of similarity map and eventually identify specific syntactic structure of queries. By transforming passage-query feature into attention maps and applying it to passage-choice matching result, we reasonably give weight to every word in the passage. Lastly, our model reaches 79.99% accuracy on the MovieQA dataset which yields top 1 result on this dataset.
In this paper, we propose Query-based Attention CNN (QACNN) to deal with the deficiencies above.
|
What did the authors propose to deal with the deficiencies mentioned in the paper?
|
A Query-based Attention CNN (QACNN) to deal with the deficiencies.
|
null | false
| null |
What are 6 natural resources of Africa?
|
Gold, Oil, Natural Gas,Copper, Silver,Diamonds
|
|
null | false
| null |
Classify each of these vehicles, based on whether they are two wheel vehicles, or four wheels vehicles: car, Gyrocar, van, motorbike, scooter, all-terrain vehicle (ATV), bicycle and Segway.
|
Vehicles with two wheels include: bicycle, motorbike, scooter, Gyrocar, and Segway.
Vehicles with four wheels include: car, van, all-terrain vehicle (ATV)
|
|
null | false
| null |
What is Kotlin?
|
Kotlin is a cross-platform, statically typed, general-purpose high-level programming language with type inference. Kotlin is designed to interoperate fully with Java, and the JVM version of Kotlin's standard library depends on the Java Class Library, but type inference allows its syntax to be more concise.
|
|
null | false
| null |
Ketchup or catsup is a table condiment with a sweet and sour flavor. The unmodified term ("ketchup") now typically refers to tomato ketchup, although early recipes used egg whites, mushrooms, oysters, grapes, mussels, or walnuts, among other ingredients.
Tomato ketchup is made from tomatoes, sugar, and vinegar, with seasonings and spices. The spices and flavors vary, but commonly include onions, allspice, coriander, cloves, cumin, garlic, and mustard, and sometimes include celery, cinnamon, or ginger.[citation needed] The market leader in the United States (60% market share) and the United Kingdom (82%) is Heinz Tomato Ketchup. Tomato ketchup is most often used as a condiment to dishes that are usually served hot and are fried or greasy: french fries and other potato dishes, hamburgers, hot dogs, chicken tenders, hot sandwiches, meat pies, cooked eggs, and grilled or fried meat. Ketchup is sometimes used as the basis for, or as one ingredient in, other sauces and dressings, and the flavor may be replicated as an additive flavoring for snacks, such as potato chips.
|
What foods are commonly eaten with ketchup?
|
Ketchup is often used with french fries and other potato dishes, hamburgers, hot dogs, chicken tenders, hot sandwiches, meat pies, cooked eggs, grilled or fried meat and as an ingredient for salad dressing and sauces.
|
null | false
| null |
Lizzie McGuire is an American comedy television series created by Terri Minsky that premiered on Disney Channel on January 12, 2001. The series stars Hilary Duff as the titular character who learns to navigate the personal and social issues of her teenage years. Duff also voices an animated version of Lizzie that performs soliloquies to express the character's inner thoughts and emotions. The series also stars Lalaine, Adam Lamberg, Jake Thomas, Hallie Todd and Robert Carradine. The series concluded on February 14, 2004, after a total of 65 episodes were produced. A feature film based on the series, The Lizzie McGuire Movie, was released in 2003.
|
What is the show Lizzie McGuire about?
|
Lizzie McGuire is a show premiered on Disney Channel about a character (Lizzie McGuire) navigating the personal and social issues of her teenage years.
|
1711.11221
| false
| null |
FLOAT SELECTED: Table 1: Experiment results on the NIST Chinese-English translation tasks. [+Cd] is the proposed model with the dynamic cache. [+Cd,Ct] is the proposed model with both the dynamic and topic cache. The BLEU scores are case-insensitive. Avg means the average BLEU score on all test sets.
FLOAT SELECTED: Table 3: The average number of words in translations of beginning sentences of documents that are also in the topic cache. Reference represents the average number of words in four human translations that are also in the topic cache.
FLOAT SELECTED: Table 6: The average cosine similarity of adjacent sentences (coherence) on all test sets.
FLOAT SELECTED: Table 1: Experiment results on the NIST Chinese-English translation tasks. [+Cd] is the proposed model with the dynamic cache. [+Cd,Ct] is the proposed model with both the dynamic and topic cache. The BLEU scores are case-insensitive. Avg means the average BLEU score on all test sets.
FLOAT SELECTED: Table 3: The average number of words in translations of beginning sentences of documents that are also in the topic cache. Reference represents the average number of words in four human translations that are also in the topic cache.
FLOAT SELECTED: Table 6: The average cosine similarity of adjacent sentences (coherence) on all test sets.
|
What evaluations did the authors use on their system?
|
BLEU scores, exact matches of words in both translations and topic cache, and cosine similarities of adjacent sentences for coherence.
|
null | false
| null |
After Jimmy John Liautaud graduated second to last in his class at Elgin Academy in 1982, his father gave him a choice to either join the Army or start a business. Although his father wanted him to enlist, the younger Liautaud chose the latter, and his father agreed to loan him $25,000 in exchange for a 48% stake in the business. Initially Jimmy John wanted to open a hot dog stand, but after visiting numerous such stands throughout the summer of 1982, he realized the $25,000 would not be enough for such a venture. After a chance encounter at a sandwich shop, Jimmy John realized that he could open a sandwich shop within his available budget by purchasing premium meats at a neighborhood market and baking his own bread. With the help of his family as tasters, he decided to put four sandwiches on his original menu. Paying $200 a month in rent, Liautaud could only afford used equipment consisting of a refrigerator, a chest freezer, an oven, and a meat slicer.
On January 13, 1983, Jimmy John's Gourmet Sandwiches opened in Charleston, Illinois. Due to the poor location of his first store, Liautaud decided to include delivery of his sandwiches to boost sales. He began by bringing samples door-to-door to the dorms of nearby Eastern Illinois University, which he had attended for one semester.
|
Given this reference text about Jimmy John's, where was the original Jimmy John's located?
|
The original Jimmy John's was located in Charleston, Illinois.
|
null | false
| null |
Alexander "Greek" Thomson (9 April 1817 – 22 March 1875) was an eminent Scottish architect and architectural theorist who was a pioneer in sustainable building. Although his work was published in the architectural press of his day, it was little appreciated outside Glasgow during his lifetime. It has only been since the 1950s and 1960s that his critical reputation has revived—not least of all in connection with his probable influence on Frank Lloyd Wright.
Henry-Russell Hitchcock wrote of Thomson in 1966: "Glasgow in the last 150 years has had two of the greatest architects of the Western world. C. R. Mackintosh was not highly productive but his influence in central Europe was comparable to such American architects as Louis Sullivan and Frank Lloyd Wright. An even greater and happily more productive architect, though one whose influence can only occasionally be traced in America in Milwaukee and in New York City and not at all as far as I know in Europe, was Alexander Thomson".
|
Who was Alexander Greek Thomson
|
Alexander Greek Thomson (9 April 1817 – 22 March 1875) was an eminent Scottish architect and architectural theorist who was a pioneer in sustainable building. Although his work was published in the architectural press of his day, it was little appreciated outside Glasgow during his lifetime. It has only been since the 1950s and 1960s that his critical reputation has revived—not least of all in connection with his probable influence on Frank Lloyd Wright.
Henry-Russell Hitchcock wrote of Thomson in 1966: "Glasgow in the last 150 years has had two of the greatest architects of the Western world. C. R. Mackintosh was not highly productive but his influence in central Europe was comparable to such American architects as Louis Sullivan and Frank Lloyd Wright. An even greater and happily more productive architect, though one whose influence can only occasionally be traced in America in Milwaukee and in New York City and not at all as far as I know in Europe, was Alexander Thomson".
|
null | false
| null |
Which is a species of fish? Puppy or Pupfish
|
Pupfish
|
|
null | false
| null |
The Spellenspektakel is the largest board game convention in the Netherlands, featuring game demonstrations, tournaments and other game related activities. It focusses on traditional strategy boardgames, collectible card games, pen-and-paper role-playing games and miniatures wargames. Game publishers participate by showing and demonstrating their new releases to the public. Additionally, the Spellenspektakel hosts numerous game stores and player communities. Visitors can try any game on display and buy the ones they prefer. During the 2019 edition at the Jaarbeurs venue in Utrecht, the Spellenspektakel welcomed over 14,000 people
|
Given a reference text about the Spellenspektakel, tell me where it takes place and what games are played.
|
The Spellenspektakel takes place in the Netherlands where board games are played.
|
null | false
| null |
Name five NHL teams in the Eastern Conference
|
Five NHL teams in the Eastern Conference are the Tampa Bay Lightning, Boston Bruins, Toronto Maple Leafs, New York Rangers, and Carolina Hurricanes
|
|
2003.07723
| true
| null |
FLOAT SELECTED: Table 7: Recall and precision scores of the best model (dbmdz) for each emotion on the test set. ‘Support’ signifies the number of labels.
FLOAT SELECTED: Table 7: Recall and precision scores of the best model (dbmdz) for each emotion on the test set. ‘Support’ signifies the number of labels.
|
Does the paper report macro F1?
|
Yes.
|
null | false
| null |
Categorize the below in terms of avocado or apple types.
Hass, Fuerte, Zutano, Fuji, Gala
|
Apple- Fuji, Gala
Avocado- Hass, Fuerte, Zutano
|
|
null | false
| null |
What is the Pacific Ring of Fire?
|
The ring of fire is the area around the pacific with high amounts of tectonic activity such as earthquakes and volcanos. Countries that have coastlines in the ring of fire include the United States and Japan
|
|
null | false
| null |
What is the volume of a rectangular prism
|
length * base * height
|
|
null | false
| null |
Some time after Donkey Kong Country, Donkey Kong is relaxing on the beach, until he is ambushed by the Kremlings. He gets kidnapped and brought to Kaptain K. Rool, King K. Rool's moniker in this game, who then demands the Banana Hoard he unsuccessfully tried to steal in the previous game for a ransom from the Kongs. Instead of complying, Diddy Kong and his friend Dixie resolve to go to the Kremling's home island, Crocodile Isle, to rescue Donkey Kong. Together, they travel through Crocodile Isle and are helped on their way by an assortment of animals to defeat Kaptain K. Rool. Diddy and Dixie eventually battle and defeat K. Rool, releasing Donkey Kong in the process. K. Rool manages to escape though, and shortly after, Diddy and Dixie confront him in the Lost World, a secret area powered by a geyser at the heart of Crocodile Isle. They once again defeat K. Rool, who is hurled into the geyser, causing it to clog up and explode. The explosion causes all of Crocodile Isle to sink, as the Kongs watch K. Rool escape on a small sailboat.
|
Given this article on Donkey Kong Country 2, who is the villain of the plot and where does the story take place?
|
Kaptain K. Rool is the villain and kidnaps Donkey Kong. The story takes places on the Kremling's home island, Crocodile Isle.
|
null | false
| null |
Why is the sky blue?
|
When the sunlight reaches Earth's atmosphere, the gases and particules in the air will scattered it in all directions. Because of its shorter wave length, blue light is scattered more than the other colors, which make it more visible in the sky, most of the time.
|
|
null | false
| 382
|
Paragraphs and other longer texts typically make multiple references to the same entities. Tracking these references and resolving coreference is essential for full machine comprehension of these texts. Significant progress has recently been made in reading comprehension research, due to large crowdsourced datasets BIBREF0, BIBREF1, BIBREF2, BIBREF3. However, these datasets focus largely on understanding local predicate-argument structure, with very few questions requiring long-distance entity tracking. Obtaining such questions is hard for two reasons: (1) teaching crowdworkers about coreference is challenging, with even experts disagreeing on its nuances BIBREF4, BIBREF5, BIBREF6, BIBREF7, and (2) even if we can get crowdworkers to target coreference phenomena in their questions, these questions may contain giveaways that let models arrive at the correct answer without performing the desired reasoning (see §SECREF3 for examples).
We introduce a new dataset, Quoref , that contains questions requiring coreferential reasoning (see examples in Figure FIGREF1). The questions are derived from paragraphs taken from a diverse set of English Wikipedia articles and are collected using an annotation process (§SECREF2) that deals with the aforementioned issues in the following ways: First, we devise a set of instructions that gets workers to find anaphoric expressions and their referents, asking questions that connect two mentions in a paragraph. These questions mostly revolve around traditional notions of coreference (Figure FIGREF1 Q1), but they can also involve referential phenomena that are more nebulous (Figure FIGREF1 Q3). Second, inspired by BIBREF8, we disallow questions that can be answered by an adversary model (uncased base BERT, BIBREF9, trained on SQuAD 1.1, BIBREF0) running in the background as the workers write questions. This adversary is not particularly skilled at answering questions requiring coreference, but can follow obvious lexical cues—it thus helps workers avoid writing questions that shortcut coreferential reasoning.
Quoref contains more than 15K questions whose answers are spans or sets of spans in 3.5K paragraphs from English Wikipedia that can be arrived at by resolving coreference in those paragraphs. We manually analyze a sample of the dataset (§SECREF3) and find that 78% of the questions cannot be answered without resolving coreference. We also show (§SECREF4) that the best system performance is 49.1% $F_1$, while the estimated human performance is 87.2%. These findings indicate that this dataset is an appropriate benchmark for coreference-aware reading comprehension.
We also show (§SECREF4) that the best system performance is 49.1% $F_1$, while the estimated human performance is 87.2%.
|
What is the best system performance?
|
49.1%.
|
null | false
| null |
IKEA pencils are small pencils provided for free in IKEA stores worldwide. They are found in small boxes attached to poles, together with maps, measuring tapes and shopping forms. The IKEA pencil has been known for the wide variety of designs. Through the years the color changed from blue, to yellow to the natural color of wood. Despite the different colors, its dimensions have always been 7×87mm. Their common in-store application is for notetaking, with customers making note of selected items from product tags onto their notepads and visiting the self-service furniture warehouse to collect their showroom products in flat pack form, using their notes to locate their products.
|
What's changed and unchanged for IKEA pencils through the years?
|
The color changed from blue, to yellow to the natural color of wood. The dimensions of the pencil has never changed which is always 7 x 87mm.
|
1909.09524
| true
| null |
Nonetheless, the main caveat of this basic pre-training is that the source encoder is trained to be used by an English decoder, while the target decoder is trained to use the outputs of an English encoder — not of a source encoder. In the following, we propose three techniques to mitigate the inconsistency of source$\rightarrow $pivot and pivot$\rightarrow $target pre-training stages. Note that these techniques are not exclusive and some of them can complement others for a better performance of the final model.
Nonetheless, the main caveat of this basic pre-training is that the source encoder is trained to be used by an English decoder, while the target decoder is trained to use the outputs of an English encoder — not of a source encoder.
|
Is pivot language used in experiments English or some other language?
|
Yes.
|
null | false
| null |
What makes plants grow faster?
|
According to a study by scientists from South Korea, plants grow at a faster pace when they are played classical music. Using 14 different pieces of music, the scientists played music to a rice field and studied the results. Findings were that the music helped the crops grow and even suggested evidence that plants could “hear”. We suggest practicing your instrument in your veggie garden!
|
|
null | false
| null |
As of 2023 the SR-71 holds the world record it set in 1976 as the fastest air-breathing manned aircraft.
|
What is the fastest air-breathing manned aircraft?
|
As of 2023 the SR-71 is the fastest air-breathing manned aircraft. It can reach a maximum speed of Mach 3.3
|
1910.14497
| false
| null |
We evaluate our framework on fastText embeddings trained on Wikipedia (2017), UMBC webbase corpus and statmt.org news dataset (16B tokens) BIBREF11. For simplicity, only the first 22000 words are used in all embeddings, though preliminary results indicate the findings extend to the full corpus. For our novel methods of mitigating bias, a shallow neural network is used to adjust the embedding. The single layer of the model is an embedding layer with weights initialized to those of the original embedding. For the composite method, these weights are initialized to those of the embedding after probabilistic bias mitigation. A batch of word indices is fed into the model, which are then embedded and for which a loss value is calculated, allowing back-propagation to adjust the embeddings. For each of the models, a fixed number of iterations is used to prevent overfitting, which can eventually hurt performance on the embedding benchmarks (See Figure FIGREF12). We evaluated the embedding after 1000 iterations, and stopped training if performance on a benchmark decreased significantly.
We construct a list of candidate words to debias, taken from the words used in the WEAT gender bias statistics. Words in this list should be gender neutral, and are related to the topics of career, arts, science, math, family and professions (see appendix). We note that this list can easily be expanded to include a greater proportion of words in the corpus. For example, BIBREF4 suggested a method for identifying inappropriately gendered words using unsupervised learning.
We compare this method of bias mitigation with the no bias mitigation ("Orig"), geometric bias mitigation ("Geo"), the two pieces of our method alone ("Prob" and "KNN") and the composite method ("KNN+Prob"). We note that the composite method performs reasonably well according the the RIPA metric, and much better than traditional geometric bias mitigation according to the neighborhood metric, without significant performance loss according to the accepted benchmarks. To our knowledge this is the first bias mitigation method to perform reasonably both on both metrics.
We evaluate our framework on fastText embeddings trained on Wikipedia (2017), UMBC webbase corpus and statmt.org news dataset (16B tokens) BIBREF11. For simplicity, only the first 22000 words are used in all embeddings, though preliminary results indicate the findings extend to the full corpus. For our novel methods of mitigating bias, a shallow neural network is used to adjust the embedding. The single layer of the model is an embedding layer with weights initialized to those of the original embedding. For the composite method, these weights are initialized to those of the embedding after probabilistic bias mitigation. A batch of word indices is fed into the model, which are then embedded and for which a loss value is calculated, allowing back-propagation to adjust the embeddings. For each of the models, a fixed number of iterations is used to prevent overfitting, which can eventually hurt performance on the embedding benchmarks (See Figure FIGREF12). We evaluated the embedding after 1000 iterations, and stopped training if performance on a benchmark decreased significantly.
We construct a list of candidate words to debias, taken from the words used in the WEAT gender bias statistics. Words in this list should be gender neutral, and are related to the topics of career, arts, science, math, family and professions (see appendix). We note that this list can easily be expanded to include a greater proportion of words in the corpus. For example, BIBREF4 suggested a method for identifying inappropriately gendered words using unsupervised learning.
We compare this method of bias mitigation with the no bias mitigation ("Orig"), geometric bias mitigation ("Geo"), the two pieces of our method alone ("Prob" and "KNN") and the composite method ("KNN+Prob"). We note that the composite method performs reasonably well according the the RIPA metric, and much better than traditional geometric bias mitigation according to the neighborhood metric, without significant performance loss according to the accepted benchmarks. To our knowledge this is the first bias mitigation method to perform reasonably both on both metrics.
|
How is embedding quality assessed?
|
The answers are shown as follows:
* We compare this method of bias mitigation with the no bias mitigation ("Orig"), geometric bias mitigation ("Geo"), the two pieces of our method alone ("Prob" and "KNN") and the composite method ("KNN+Prob"). We note that the composite method performs reasonably well according the the RIPA metric, and much better than traditional geometric bias mitigation according to the neighborhood metric, without significant performance loss according to the accepted benchmarks. To our knowledge this is the first bias mitigation method to perform reasonably both on both metrics.
|
null | false
| null |
The former Chairman of Tata Sons died in road accident en route from where?
|
Ahmedabad to Mumbai
|
|
null | false
| null |
What causes that feeling of weightlessness when you're on a rollercoaster?
|
The feeling of weightlessness you may get when on a rollercoaster is caused by gravitational force (g-force), and more specifically, negative g-force. Negative g-force occurs when you accelerate downwards faster than freefall, and that downward acceleration produces a force in the upward direction, thus pulling you out of your seat and causing a feeling of weightlessness.
|
|
null | false
| 26
|
In this section, we introduce a novel RvNN architecture, called SATA Tree-LSTM (Structure-Aware Tag Augmented Tree-LSTM). This model is similar to typical Tree-LSTMs, but provides dynamic compositionality by augmenting a separate tag-level tree-LSTM which produces structure-aware tag representations for each node in a tree. In other words, our model has two independent tree-structured modules based on the same constituency tree, one of which (word-level tree-LSTM) is responsible for constructing sentence representations given a sequence of words as usual, while the other (tag-level tree-LSTM) provides supplementary syntactic information to the former.
In section 3.1, we first review tree-LSTM architectures. Then in section 3.2, we introduce a tag-level tree-LSTM and structure-aware tag representations. In section 3.3, we discuss an additional technique to boost the performance of tree-structured models, and in section 3.4, we describe the entire architecture of our model in detail.
In other words, our model has two independent tree-structured modules based on the same constituency tree, one of which (word-level tree-LSTM) is responsible for constructing sentence representations given a sequence of words as usual, while the other (tag-level tree-LSTM) provides supplementary syntactic information to the former.
|
How does the new model introduced by the author work?
|
Their model has two independent tree-structured modules based on the same constituency tree, one of which (word-level tree-LSTM) is responsible for constructing sentence representations given a sequence of words as usual, while the other (tag-level tree-LSTM) provides supplementary syntactic information to the former.
|
null | false
| null |
Classify the musical genres of the following bands: Metallica, AC/DC, Aerosmith, Madonna
|
These artists are one of the most well-known musicians worldwide. As of their genres, Metallica can be categorised as a Heavy Metal band, AC/DC and Aerosmith as Hard Rock and Madonna as a Pop artist
|
|
null | false
| 111
|
In Turkish, people sometimes prefer to spell English characters for the corresponding Turkish characters (e.g. i for ı, c for ç) when writing in electronic format. To normalise such words, we used the Zemberek tool BIBREF15. All punctuation marks except “!" and “?" are removed, since they do not contribute much to the polarity of a document. We took into account emoticons, such as “:))", and idioms, such as “kafayı yemek” (lose one's mind), since two or more words can express a sentiment together, irrespective of the individual words thereof. Since Turkish is an agglutinative language, we used the morphological parser and disambiguator tools BIBREF16, BIBREF17. We also performed negation handling and stop-word elimination. In negation handling, we append an underscore to the end of a word if it is negated. For example, “güzel değil" (not beautiful) is redefined as “güzel_" (beautiful_) in the feature selection stage when supervised scores are being computed.
All punctuation marks except “!" and “?" are removed, since they do not contribute much to the polarity of a document.
|
Why are most punctuation marks removed in preprocessing?
|
They do not contribute much to the polarity of a document except “!" and “?".
|
null | false
| null |
Which characters belong to DC or Marvel Universe? Wasp, Bumblebee
|
Wasp is Marvel, Bumblebee is DC
|
|
null | false
| null |
What is the most popular sport in the UK?
|
Football is the most popular game played in the U.K. and follows a traditional league system which consists of more than one hundred teams. The most popular league is known as the Premier League and consists of the 20 best teams from all over the U.K. The most popular of these teams are Manchester City, Manchester United, Liverpool, Arsenal and Chelsea. Football in the U.K. is governed by the Football Association which is one of the oldest governing bodies in the entire world. The two most famous Football Championships in the U.K. are the FA Cup and the Football League Cup. There are 92 professional football clubs that participate in each of these tournaments each year.
|
|
null | false
| null |
Who is the only US president elected from Pennsylvania?
|
James Buchanan, the 15th President of the United States.
|
|
null | false
| null |
"Virtual reality (VR) is a simulated experience that employs pose tracking and 3D near-eye displays to give the user an immersive feel of a virtual world. Applications of virtual reality include entertainment (particularly video games), education (such as medical or military training) and business (such as virtual meetings). Other distinct types of VR-style technology include augmented reality and mixed reality, sometimes referred to as extended reality or XR, although definitions are currently changing due to the nascence of the industry.
Currently, standard virtual reality systems use either virtual reality headsets or multi-projected environments to generate some realistic images, sounds and other sensations that simulate a user's physical presence in a virtual environment. A person using virtual reality equipment is able to look around the artificial world, move around in it, and interact with virtual features or items. The effect is commonly created by VR headsets consisting of a head-mounted display with a small screen in front of the eyes, but can also be created through specially designed rooms with multiple large screens. Virtual reality typically incorporates auditory a"
|
What can a person do in Virtual Reality?
|
A person using virtual reality equipment is able to look around the artificial world, move around in it, and interact with virtual features or items.
|
1911.09886
| false
| null |
Among the baselines, HRL achieves significantly higher F1 scores on the two datasets. We run their model and our models five times and report the median results in Table TABREF15. Scores of other baselines in Table TABREF15 are taken from previous published papers BIBREF6, BIBREF11, BIBREF14. Our WordDecoding (WDec) model achieves F1 scores that are $3.9\%$ and $4.1\%$ higher than HRL on the NYT29 and NYT24 datasets respectively. Similarly, our PtrNetDecoding (PNDec) model achieves F1 scores that are $3.0\%$ and $1.3\%$ higher than HRL on the NYT29 and NYT24 datasets respectively. We perform a statistical significance test (t-test) under a bootstrap pairing between HRL and our models and see that the higher F1 scores achieved by our models are statistically significant ($p < 0.001$). Next, we combine the outputs of five runs of our models and five runs of HRL to build ensemble models. For a test instance, we include those tuples which are extracted in the majority ($\ge 3$) of the five runs. This ensemble mechanism increases the precision significantly on both datasets with a small improvement in recall as well. In the ensemble scenario, compared to HRL, WDec achieves $4.2\%$ and $3.5\%$ higher F1 scores and PNDec achieves $4.2\%$ and $2.9\%$ higher F1 scores on the NYT29 and NYT24 datasets respectively.
Our WordDecoding (WDec) model achieves F1 scores that are $3.9\%$ and $4.1\%$ higher than HRL on the NYT29 and NYT24 datasets respectively. Similarly, our PtrNetDecoding (PNDec) model achieves F1 scores that are $3.0\%$ and $1.3\%$ higher than HRL on the NYT29 and NYT24 datasets respectively.
|
Which one of two proposed approaches performed better in experiments?
|
The answers are shown as follows:
* WordDecoding (WDec) model
|
2002.06053
| true
| null |
The Word2Vec architecture has inspired a great deal of research in the bio/cheminformatics domains. The Word2Vec algorithm has been successfully applied for determining protein classes BIBREF44 and protein-protein interactions (PPI) BIBREF56. BIBREF44 treated 3-mers as the words of the protein sequence and observed that 3-mers with similar biophysical and biochemical properties clustered together when their embeddings were mapped onto the 2D space. BIBREF56, on the other hand, utilized BPE-based word segmentation (i.e. bio-words) to determine the words. The authors argued that the improved performance for bio-words in the PPI prediction task might be due to the segmentation-based model providing more distinct words than $k$-mers, which include repetitive segments. Another recent study treated multi-domain proteins as sentences in which each domain was recognized as a word BIBREF60. The Word2Vec algorithm was trained on the domains (i.e. PFAM domain identifiers) of eukaryotic protein sequences to learn semantically interpretable representations of them. The domain representations were then investigated in terms of the Gene Ontology (GO) annotations that they inherit. The results indicated that semantically similar domains share similar GO terms.
The Word2Vec architecture has inspired a great deal of research in the bio/cheminformatics domains. The Word2Vec algorithm has been successfully applied for determining protein classes BIBREF44 and protein-protein interactions (PPI) BIBREF56.
|
Is there any concrete example in the paper that shows that this approach had huge impact on drug discovery?
|
Yes.
|
null | false
| null |
Why do people have children?
|
There are many reasons people have children. A common reason is simply tradition and expectations. Some people have grown up expecting to have children, their parents expect them to have children, and all of their friends have children, so they have children. Another reason is economic. In agricultural societies children can be an economic benefit, providing labor to help the family manage the farm. Another common reason is that children can provide support when the parents become elderly. Finally, the parents may simply be looking for the meaning, affection, and love that children bring to their lives.
|
|
1811.08048
| false
| null |
We present and evaluate a model that we have developed for this, called QuaSP+Zero, that modifies the QuaSP+ parser as follows: During decoding, at points where the parser is selecting which property to include in the LF (e.g., Figure FIGREF31 ), it does not just consider the question tokens, but also the relationship between those tokens and the properties INLINEFORM0 used in the qualitative model. For example, a question token such as “longer” can act as a cue for (the property) length, even if unseen in the training data, because “longer” and a lexical form of length (e.g.,“length”) are similar. This approach follows the entity-linking approach used by BIBREF11 Krishnamurthy2017NeuralSP, where the similarity between question tokens and (words associated with) entities - called the entity linking score - help decide which entities to include in the LF during parsing. Here, we modify their entity linking score INLINEFORM1 , linking question tokens INLINEFORM2 and property “entities” INLINEFORM3 , to be: INLINEFORM4
We present and evaluate a model that we have developed for this, called QuaSP+Zero, that modifies the QuaSP+ parser as follows: During decoding, at points where the parser is selecting which property to include in the LF (e.g., Figure FIGREF31 ), it does not just consider the question tokens, but also the relationship between those tokens and the properties INLINEFORM0 used in the qualitative model.
This approach follows the entity-linking approach used by BIBREF11 Krishnamurthy2017NeuralSP, where the similarity between question tokens and (words associated with) entities - called the entity linking score - help decide which entities to include in the LF during parsing.
|
How does the QuaSP+Zero model work?
|
The answers are shown as follows:
* does not just consider the question tokens, but also the relationship between those tokens and the properties
|
null | false
| null |
Classify each of the following as a liquid or solid: water, mailbox, soda, coffee, bush, grass, milk
|
Water: Liquid
Mailbox: Solid
Soda: Liquid
Coffee: Liquid
Bush: Solid
Grass: Solid
Milk: Liquid
|
|
null | false
| 401
|
Medical providers across the United States are required to document clinical visits in the Electronic Health Records. This need for documentation takes up a disproportionate amount of their time and attention, resulting in provider burnout BIBREF0, BIBREF1. One study found that full-time primary care physicians spent about 4.5 hours of an 11-hour workday interacting with the clinical documentation systems, yet were still unable to finish their documentation and had to spend an additional 1.4 hours after normal clinical hours BIBREF2.
Speech and natural language processing are now sufficiently mature that there is considerable interest, both in academia and industry, to investigate how these technologies can be exploited to simplify the task of documentation, and to allow providers to dedicate more time to patients. While domain-specific automatic speech recognition (ASR) systems that allow providers to dictate notes have been around for a while, recent work has begun to address the challenges associated with generating clinical notes directly from speech recordings. This includes inducing topic structure from conversation data, extracting relevant information, and clinical summary generation BIBREF3. In one recent work, authors outlined an end-to-end system; however, the details were scant without empirical evaluations of their building blocks BIBREF4. One of the simplistic approaches uses a hand crafted finite state machine based grammar to locate clinical entities in the ASR transcripts and map them to canonical clinical terms BIBREF5. This seems to perform well in a narrowly scoped task. A more ambitious approach mapped ASR transcripts to clinical notes by adopting a machine translation approach BIBREF6. However this performed poorly. To address the difficulty in accessing clinical data, researchers have experimented with synthetic data to develop a system for documenting nurse-initiated telephone conversations for congestive heart failure patients who are undergoing telemonitoring after they have been discharged from the hospital BIBREF7. In their task, a question-answer based model achieved an F-score of 0.80. This naturally raises the question of how well state-of-art techniques will perform in helping the broader population of clinicians such as primary care providers.
One might expect that the task of extracting clinical concepts from audio faces challenges similar to the domain of unstructured clinical texts. In that domain, one of the earliest public-domain tasks is the i2b2 relations challenge, defined on a small corpus of written discharge summaries consisting of 394 reports for training, 477 for test, and 877 for evaluation BIBREF8. Given the small amount of training data, not surprisingly, a disproportionately large number of teams fielded rule-based systems. Conditional random field-based (CRF) systems BIBREF9 however did better even with the limited amount of training data BIBREF10. Other i2b2/n2c2 challenges focused on coreference resolution BIBREF11, temporal relation extraction BIBREF12, drug event extraction BIBREF13 on medical records, and extracting family history BIBREF14. Even though the text was largely unstructured, they benefited from punctuation and capitalization, section headings and other cues in written domain which are unavailable in audio to the same extent.
With the goal of creating an automated medical scribe, we broke down the task into modular components, including ASR and speaker diarization which are described elsewhere BIBREF15. In this work, we investigate the task of extracting relevant clinical concepts from transcripts. Our key contributions include: (i) defining three tasks – the Medications Task, the Symptoms Task, and the Conditions Task along with principles employed in developing the annotation guidelines for them (Section SECREF2); (ii) measuring the label quality using inter-labeler agreements and refining the quality iteratively (Section SECREF3), (iii) evaluating the performance of the state-of-the-art models on these tasks (Section SECREF4), and (iv) a comprehensive analysis of the performance of the models including manual error categorization (Section SECREF5). The corpus we have created in this work is based on private, proprietary data that cannot be publicly shared. Instead, we are sharing the learnings from our experience that might be useful for the wider community as well as the detailed labeling guidelines as supplementary material in the extended version of this paper on arxiv.org.
Our key contributions include: (i) defining three tasks – the Medications Task, the Symptoms Task, and the Conditions Task along with principles employed in developing the annotation guidelines for them (Section 2.);
|
What are the three tasks they identify?
|
The Medications Task, the Symptoms Task, and the Conditions Task.
|
null | false
| 104
|
We propose a Bayesian model of semantic role induction (SRI) that uses crosslingual latent variables to capture role alignments in parallel corpora. The crosslingual latent variables capture correlations between roles in different languages, and regularize the parameter estimates of the monolingual models. Because this is a joint Bayesian model of multilingual SRI, we can apply the same model to a variety of training scenarios just by changing the inference procedure appropriately. We evaluate monolingual SRI with a large unlabeled dataset, bilingual SRI with a parallel corpus, bilingual SRI with annotations available for the source language, and monolingual SRI with a small labeled dataset. Increasing the amount of monolingual unlabeled data significantly improves SRI in German but not in English. Adding word alignments in parallel sentences results in small, non significant improvements, even if there is some labeled data available in the source language. This difficulty in showing the usefulness of parallel corpora for SRI may be due to the current assumptions about role alignments, which mean that only a small percentage of roles are aligned. Further analyses reveals that annotating small amounts of data can easily outperform the performance gains obtained by adding large unlabeled dataset as well as adding parallel corpora.
Future work includes training on different language pairs, on more than two languages, and with more inclusive models of role alignment.
The crosslingual latent variables capture correlations between roles in different languages, and regularize the parameter estimates of the monolingual models.
|
What are the functions of the crosslingual latenfut variables in the model proposed?
|
The crosslingual latent variables capture correlations between roles in different languages, and regularize the parameter estimates of the monolingual models.
|
null | false
| 388
|
Table TABREF21 shows the performance of our models when using all features described in Section SECREF4: see the SVM$_{All}$ and the FNN$_{All}$ rows. In order to put the numbers in perspective, we also show the results for five increasingly competitive baselines.
First, there is a random baseline, followed by an SVM classifier based on a bag-of-words representation with TF.IDF weights learned on the training data. Then come three versions of the ClaimBuster system: CB-Platform uses scores from the online demo, which we accessed on December 20, 2016, and SVM$_{CBfeat}$ and FNN$_{CBfeat}$ are our re-implementations, trained on our dataset.
We can see that all systems perform well above the random baseline. The three versions of ClaimBuster also outperform the TF.IDF baseline on most measures. Moreover, our reimplementation of ClaimBuster performs better than the online platform in terms of MAP. This is expected as their system is trained on a different dataset and it may suffer from testing on slightly out-of-domain data. At the same time, this is reassuring for our implementation of the features, and allows for a more realistic comparison to the ClaimBuster system.
More importantly, both the SVM and the FNN versions of our system consistently outperform all three versions of ClaimBuster on all measures. This means that the extra information coded in our model, mainly more linguistic, structural, and contextual features, has an important contribution to the overall performance.
We can further see that the neural network model, FNN$_{All}$, clearly outperforms the SVM model: consistently on all measures. As an example, with the precision values achieved by FNN$_{All}$, the system would rank on average 4 positive examples in the list of its top-5 choices, and also 14-15 in the top-20 list. Considering the recall at the first $R$ sentences, we will be able to encounter 43% of the total number of check-worthy sentences. This is quite remarkable given the difficulty of the task.
More importantly, both the SVM and the FNN versions of our system consistently outperform all three versions of ClaimBuster on all measures.
|
Do the SVM and the FNN versions of the system consistently outperform all three versions of ClaimBuster on all measures?
|
Yes, they do.
|
null | false
| null |
What is the current recommendation for a healthy diet?
|
The current guidance on eating a healthy diet consists of eating a balance of healthy foods. Most of your meal should consist of a mix of fruits and vegetables, taking into account that potatoes are not considered to count as a vegetable as they have a negative impact on blood sugar. A quarter of your meal should consist of whole grains, things like whole wheat, quinoa, oats, brown rice and things made from them such as whole wheat bread and pasta. The other quarter of your plate should consist of proteins such as fish, poultry, beans and nuts. You should limit red meat and avoid processed meat such as bacon and sausages. You should choose to cook or dress your food with healthy plant oils, such as olive oil or sunflower oil and avoid partially hydrogenated oils which contain trans fats. You should drink water, coffee or tea, avoid sugary drinks and limit fruit juice to only one small glass a day.
|
|
null | false
| null |
Classify each as either a vodka drink or a gin drink: Screwdriver, Bloody Mary, French 75, White Russian, Gimlet, Cosmopolitan, Tom Collins, Negroni
|
Screwdrivers, Bloody Mary's, White Russians, and Cosmopolitans are traditionally vodka cocktails, whereas French 75's, Gimlets, Tom Collins', and Negronis are traditionally gin cocktails.
|
|
null | false
| null |
What streaming shows are worth binging right now?
|
First series that comes to mind is the HBO show Succession. This series dives into the dynamic of a family business on a big scale and the power shifts and crazy situations the children of the CEO get themselves into. On a total different viewing experience is Love is Blind on Netflix. This reality show involves an experiment that is trying to determine if Love is Blind. Contestants date in pods, with a with in between them and only can talk to one another and not physically see the other person. Love does come about and so does heart break.
|
|
null | false
| null |
Geberit (Alemannic German pronunciation: [ˈɡeberɪ:t] is a Swiss multinational group specialized in manufacturing and supplying sanitary parts and related systems. It is a leader in its field in Europe with a global presence through its subsidiaries.
|
Who is the company Geberit
|
Geberit is a Swiss multinational group specialized in manufacturing and supplying sanitary parts and related systems. It is a leader in its field in Europe with a global presence through its subsidiaries.
|
null | false
| null |
Who is the Aston Martin formula 1 team principal?
|
The Aston Martin Aramco Team Principal is Mike Krack
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.