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 |
Why is it important to hug your child?
|
Hugs can help your child grow a healthy brain and body.
Infants, toddlers, and preschoolers need many different types of sensory stimulation for development. Skin contact or physical touch, like hugging and holding, are great ways to help your child grow a healthy brain and a strong body.
|
|
null | false
| null |
Identify which animal species is alive or extinct: Vegavis, Continental Tiger
|
Continental Tiger is alive, Vegavis is extinct.
|
|
1606.01433
| false
| null |
Phase 2: In order to predict the relationship between an event and the creation time of its parent document, we assign a DocRelTime random variable to every Timex3 and Event mention. For Events, these values are provided by the training data, for Timex3s we have to compute class labels. Around 42% of Timex3 mentions are simple dates (“12/29/08", “October 16", etc.) and can be naively canonicalized to a universal timestamp. This is done using regular expressions to identify common date patterns and heuristics to deal with incomplete dates. The missing year in “October 16", for example, can be filled in using the nearest preceding date mention; if that isn't available we use the document creation year. These mentions are then assigned a class using the parent document's DocTime value and any revision timestamps. Other Timex3 mentions are more ambiguous so we use a distant supervision approach. Phrases like “currently" and “today's" tend to occur near Events that overlap the current document creation time, while “ago" or “ INLINEFORM0 -years" indicate past events. These dominant temporal associations can be learned from training data and then used to label Timex3s. Finally, we define a logistic regression rule to predict entity DocRelTime values as well as specify a linear skip-chain factor over Event mentions and their nearest Timex3 neighbor, encoding the baseline system heuristic directly as an inference rule.
Other Timex3 mentions are more ambiguous so we use a distant supervision approach. Phrases like “currently" and “today's" tend to occur near Events that overlap the current document creation time, while “ago" or “ INLINEFORM0 -years" indicate past events. These dominant temporal associations can be learned from training data and then used to label Timex3s. Finally, we define a logistic regression rule to predict entity DocRelTime values as well as specify a linear skip-chain factor over Event mentions and their nearest Timex3 neighbor, encoding the baseline system heuristic directly as an inference rule.
|
How do they obtain distant supervision rules for predicting relations?
|
The answers are shown as follows:
* dominant temporal associations can be learned from training data
|
null | false
| null |
How should I plan my week as a working mom ?
|
As a working mom you are juggling between your professional commitments, with your personal passion projects, running the house with daily activities like grocery shopping and meal preparation and if you have school going kids you have lessons to teach too. You and your partner are a team and know that he/she should be contributing too. You can do this buy planning out which activities your partner owns, shares and which ones you will have to handle. In addition having some clear activities or goals/tasks that you want to acheive will help you to handle few important things and few long term activities one week at a time.
|
|
null | false
| null |
Ralph continues trying to cross the border. The Spade army begins crossing the border in a mechanized citadel called Candelo powered by siphoning magic from Spade commoners. Asta, who has grown even stronger and can now fly astride his demon slayer sword, flies to Candelo where he is enraged to see Spade Knights siphoning prisoners to death. Asta confronts Candelo’s commander who reveals he is also the vessel for a devil. He poisons Asta with poison magic but Asta easily absorbs the poison into the demon destroyer sword and furiously cuts the citadel in half, defeating the captain in a single blow. The Bulls arrive and Mimosa helps the prisoners recover. Noelle reveals Luck and Leopold were sent to destroy the Spade base near to the prisoner’s village. Realising they get to return home safely the prisoners reveal the Spade kingdom is ruled by the Dark Triad, each possessed by a powerful devil, and their followers the Dark Disciples who draw demonic power from the Triad. The Triad, siblings Dante, Vanica and Zenon Zogratis, consider the loss of Candelo trivial when they are close to their goal for which they only need capture two Arcane Stage mages. Ralph finally crosses the border and collapses by Hage Church where Sister Lily finds him calling out for “Prince Yuno”.
|
In the summary below of episode 158 of the series Black Clover, how did the Spade army lose?
|
Asta countered the Spade Kingdom's commandor's poison attack with the demon destroyer sword, then cut the attacking citadel in half, defeating the captain in a single blow.
|
null | false
| 249
|
The goal of the current research was to investigate the automatic detection of cyberbullying-related posts on social media. Given the information overload on the web, manual monitoring for cyberbullying has become unfeasible. Automatic detection of signals of cyberbullying would enhance moderation and allow to respond quickly when necessary.
Cyberbullying research has often focused on detecting cyberbullying `attacks', hence overlooking posts written by victims and bystanders. However, these posts could just as well indicate that cyberbullying is going on. The main contribution of this paper is that it presents a system for detecting signals of cyberbullying on social media, including posts from bullies, victims and bystanders. A manually annotated cyberbullying dataset was created for two languages, which will be made available for public scientific use. Moreover, while a fair amount of research has been done on cyberbullying detection for English, we believe this is one of the first papers that focus on Dutch as well.
A set of binary classification experiments were conducted to explore the feasibility of automatic cyberbullying detection on social media. In addition, we sought to determine which information sources contribute to this task. Two classifiers were trained on English and Dutch ASKfm data and evaluated on a holdout test of the same genre. Our experiments reveal that the current approach is a promising strategy for detecting signals of cyberbullying in social media data automatically. After feature selection and hyperparameter optimisation, the classifiers achieved an F INLINEFORM0 -score of 64.32% and 58.72% for English and Dutch, respectively. The systems hereby significantly outperformed a keyword and an (unoptimised) INLINEFORM1 -gram baseline. Analysis of the results revealed that false positives often include implicit cyberbullying or offenses through irony, the challenge of which will constitute an important area for future work.
Another interesting direction for future work would be the detection of fine-grained cyberbullying-related categories such as threats, curses and expressions of racism and hate. When applied in a cascaded model, the system could find severe cases of cyberbullying with high precision. This would be particularly interesting for monitoring purposes, since it would allow to prioritise signals of bullying that are in urgent need for manual inspection and follow-up.
Finally, future work will focus on the detection of participants (or roles) typically involved in cyberbullying. This would allow to analyse the context of a cyberbullying incident and hence evaluate its severity. When applied as moderation support on online platforms, such a system would allow to provide feedback in function of the recipient (i.e., a bully, victim, or bystander).
Analysis of the combined and single feature type sets reveals that word n-grams, character n-grams, and subjectivity exicons prove to be strong features for this task.
|
What is the main contribution of this paper?
|
It presents a system for detecting signals of cyberbullying on social media.
|
null | false
| null |
Where is Jordan Spieth from?
|
Dallas, Texas
|
|
null | false
| null |
How do I create knit stitches?
|
In patterns a knit stitch is often abbreviated as k. A knit stitch is the first basic stitch to learn. Hold the needle with the cast-on stitches in your left hand and the second empty needle in your right hand. The yarn that is coming from your skein or ball of yarn should be held toward the back of your work and this yard strand is considered your working yard. Push the empty needle into the first loop between the first and second stitch from left to right. Grab the working yarn with your right hand and loop it right to left behind the needle that you just pushed into the first stitch. Keeping that loop on the right needle, slide the needle tips toward each other so the stitch on the left needle can fall off of the needle , keeping the loop on the right needle that you created. This is your first knit stitch. Continue the pattern to the end of the row.
|
|
null | false
| null |
What is functional genomics?
|
Functional genomics is a field of molecular biology that attempts to describe gene (and protein) functions and interactions. Functional genomics make use of the vast data generated by genomic and transcriptomic projects (such as genome sequencing projects and RNA sequencing). Functional genomics focuses on the dynamic aspects such as gene transcription, translation, regulation of gene expression and protein–protein interactions, as opposed to the static aspects of the genomic information such as DNA sequence or structures. A key characteristic of functional genomics studies is their genome-wide approach to these questions, generally involving high-throughput methods rather than a more traditional "candidate-gene" approach.
|
|
null | false
| null |
Give me the steps to purchase a new pair of shoes.
|
First, determine the style of shoes you desire. (E.g. sneakers, sandals, tuxedo shoes)
Second, check your shoe size.
Third, browse online or in a store and filter your choices according to your budget, style and availability. Try the shoes on before purchase if possible to ensure a good fit and comfort.
Fourth, select your shoes and proceed to purchase.
Fifth, purchase the shoes either online or in the store.
|
|
null | false
| 390
|
Natural Language Processing (NLP) systems rely on large-scale training data BIBREF0 for supervised training. However, manual annotation can be time-consuming and expensive. Despite detailed annotation standards and rules, inter-annotator disagreement is inevitable because of human mistakes, language phenomena which are not covered by the annotation rules and the ambiguity of language itself BIBREF1 .
Existing annotation tools BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 mainly focus on providing a visual interface for user annotation process but rarely consider the post-annotation quality analysis, which is necessary due to the inter-annotator disagreement. In addition to the annotation quality, efficiency is also critical in large-scale annotation task, while being relatively less addressed in existing annotation tools BIBREF6 , BIBREF7 . Besides, many tools BIBREF6 , BIBREF4 require a complex system configuration on either local device or server, which is not friendly to new users.
To address the challenges above, we propose Yedda , a lightweight and efficient annotation tool for text span annotation. A snapshot is shown in Figure FIGREF4 . Here text span boundaries are selected and assigned with a label, which can be useful for Named Entity Recognition (NER) BIBREF8 , word segmentation BIBREF9 , chunking BIBREF10 ,etc. To keep annotation efficient and accurate, Yedda provides systematic solutions across the whole annotation process, which includes the shortcut annotation, batch annotation with a command line, intelligent recommendation, format exporting and administrator evaluation/analysis.
Figure FIGREF1 shows the general framework of Yedda. It offers annotators with a simple and efficient Graphical User Interface (GUI) to annotate raw text. For the administrator, it provides two useful toolkits to evaluate multi-annotated text and generate detailed comparison report for annotator pair. Yedda has the advantages of being:
• INLINEFORM0 Convenient: it is lightweight with an intuitive interface and does not rely on specific operating systems or pre-installed packages.
• INLINEFORM0 Efficient: it supports both shortcut and command line annotation models to accelerate the annotating process.
• INLINEFORM0 Intelligent: it offers user with real-time system suggestions to avoid duplicated annotation.
• INLINEFORM0 Comprehensive: it integrates useful toolkits to give the statistical index of analyzing multi-user annotation results and generate detailed content comparison for annotation pairs.
This paper is organized as follows: Section 2 gives an overview of previous text annotation tools and the comparison with ours. Section 3 describes the architecture of Yedda and its detail functions. Section 4 shows the efficiency comparison results of different annotation tools. Finally, Section 5 concludes this paper and give the future plans.
To keep annotation efficient and accurate, YEDDA provides systematic solutions across the whole annotation process, which includes the shortcut annotation, batch annotation with a command line, intelligent recommendation, format exporting and administrator evaluation/analysis.
|
What systematic solutions were provided by Yedda across the whole annotation process?
|
Includes the shortcut annotation, batch annotation with a command line, intelligent recommendation, format exporting and administrator evaluation/analysis.
|
null | false
| null |
Why is Cricket popular in India?
|
Cricket is the most popular sport in India. The game was introduced by the Britishers when they were ruling India and since then Indians have started taking to the game. Till the early 1980's the game was mostly followed by the educated and the people living in the large cities like Madras, Bombay, Delhi and Calcutta. India winning the One Day world cup in 1983 made the game very popular and the game started spreading to the entire country. The advent of T20s in the early 2000s further made the game more popular. India won the T20 world cup in 2007 and following that India Premier League (IPL) was launched. IPL is the most lucrative cricket league in the world and this caught the imagination of the people of India. IPL further contributed to making cricket popular in India.
|
|
null | false
| null |
Belize, formerly known as British Honduras, is a Caribbean country located on the northeastern coast of Central America. Belize is bordered on the northwest by Mexico, on the east by the Caribbean Sea, and on the south and west by Guatemala. It has an area of 22,970 square kilometres (8,867 sq mi) and a population of 408,487 (2019). Its mainland is about 290 km (180 mi) long and 110 km (68 mi) wide. It has the lowest population and population density in Central America.
|
Given this short paragraph about Belize, how many people live there and what is notable about its population density?
|
Belize has a population of 408,487 people and has the lowest population density in Central America.
|
1604.00117
| false
| null |
Slot filling models are a useful method for simple natural language understanding tasks, where information can be extracted from a sentence and used to perform some structured action. For example, dates, departure cities and destinations represent slots to fill in a flight booking task. This information is extracted from natural language queries leveraging typical context associated with each slot type. Researchers have been exploring data-driven approaches to learning models for automatic identification of slot information since the 90's, and significant advances have been made BIBREF0 . Our paper builds on recent work on slot-filling using recurrent neural networks (RNNs) with a focus on the problem of training from minimal annotated data, taking an approach of sharing data from multiple tasks to reduce the amount of data for developing a new task.
As candidate tasks, we consider the actions that a user might perform via apps on their phone. Typically, a separate slot-filling model would be trained for each app. For example, one model understands queries about classified ads for cars BIBREF1 and another model handles queries about the weather BIBREF2 . As the number of apps increases, this approach becomes impractical due to the burden of collecting and labeling the training data for each model. In addition, using independent models for each task has high storage costs for mobile devices.
Crowd-sourced data was collected simulating common use cases for four different apps: United Airlines, Airbnb, Greyhound bus service and OpenTable. The corresponding actions are booking a flight, renting a home, buying bus tickets, and making a reservation at a restaurant. In order to elicit natural language, crowd workers were instructed to simulate a conversation with a friend planning an activity as opposed to giving a command to the computer. Workers were prompted with a slot type/value pair and asked to form a reply to their friend using that information. The instructions were to not include any other potential slots in the sentence but this instruction was not always followed by the workers.
Slot filling models are a useful method for simple natural language understanding tasks, where information can be extracted from a sentence and used to perform some structured action
As candidate tasks, we consider the actions that a user might perform via apps on their phone.
Crowd-sourced data was collected simulating common use cases for four different apps: United Airlines, Airbnb, Greyhound bus service and OpenTable. The corresponding actions are booking a flight, renting a home, buying bus tickets, and making a reservation at a restaurant.
|
What tasks are they experimenting with in this paper?
|
The answers are shown as follows:
* Slot filling
* we consider the actions that a user might perform via apps on their phone
* The corresponding actions are booking a flight, renting a home, buying bus tickets, and making a reservation at a restaurant
|
null | false
| null |
Which of the following technology categorized as Infrastructure as code (IAC) or CICD tool? Terraform, AWS Cloud Formation , Github actions, Azure Devops pipeline
|
Terraform and AWS Cloud Formation is categorized as Infrastructure as code (IAC) and Github actions, Azure Devops pipeline is categorized as CICD tool.
|
|
null | false
| null |
Classify each of the following as either a planet, a star, or a galaxy: Mercury, Venus, Polaris, Earth, Sun, Mars, Jupiter, Milky Way, Andromeda, Saturn, Virgo A, Maffei 1, Uranus, Neptune, Sirius, Vega.
|
Stars: Sun, Polaris, Sirius, Vega
Planets: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, Neptune
Galaxies: Milky Way, Andromeda, Virgo A, Maffei 1
|
|
null | false
| 102
|
To make a qualitative assessment of the generated sentences, we conducted a human study on a sample of 500 Infoboxes which were sampled from English dataset. The annotators for this task were undergraduate and graduate students. For each of these infoboxes, we generated summaries using the basic seq2seq model and our final model with bifocal attention and gated orthogonalization. For each description and for each model, we asked three annotators to rank the output of the systems based on i) adequacy (i.e. does it capture relevant information from the infobox), (ii) fluency (i.e. grammar) and (iii) relative preference (i.e., which of the two outputs would be preferred). Overall the average fluency/adequacy (on a scale of 5) for basic seq2seq model was INLINEFORM0 and INLINEFORM1 for our model respectively.
The results from Table TABREF17 suggest that in general gated orthogonalization model performs better than the basic seq2seq model. Additionally, annotators were asked to verify if the generated summaries look natural (i.e, as if they were generated by humans). In 423 out of 500 cases, the annotators said “Yes” suggesting that gated orthogonalization model indeed produces good descriptions.
Overall the average fluency/adequacy (on a scale of 5) for basic seq2seq model was 4.04/3.6 and 4.19/3.9 for our model respectively.
|
How about human evaluation results on the seq2seq model ?
|
Overall the average fluency/adequacy (on a scale of 5) for basic seq2seq model was 4.04/3.6 and 4.19/3.9 for their model respectively.
|
null | false
| 414
|
Figure 7: The figure illustrates two sets of curves. The solid curves represent the median query count required to successfully attack the three architectures as a function of the step length, while the dashed curves show the success rate.
The simple method we propose in Algorithm 1 accepts a single hyperparameter: the step length ϵ(line 14), which indicates the length of the perturbations to be attempted at each iterate along its candidate direction. We chose 2.0 as the default value for our experiments by performing a small grid search over a held-out set (disjoint from the 2000 examples used in the experiments of the main paper).
Results for the three architectures used as victim (and ResNet-152 as the single surrogate) are shown in Fig. 7. Note that the figure has two y-axes and two sets of curves: solid and dashed curves should be considered looking at each axis on the left and right hand side of the plot, respectively. Clearly, better performance can be obtained by separately tuning this hyperparameter for each architecture.
Instead, we simply chose a fixed value that is reasonable for all models, balancing the query count and success rate. Note also the fact that it is possible to achieve an even lower query count if one is willing to sacrifice about 1% of the success rate.
We observed that this choice of ϵ is also significantly better than the default one made by SimBA_x005f_x0002_ODS (Tashiro et al., 2020) (i.e. 0.2), and have reported results for both in our experiments.
|
How is the value of epsilon been chosen as it significantly affects the performance of Simba-ODS?
|
Regarding the specific case of SimBA-ODS, we have presented two different epsilon tunings: 0.2 and 2.0. The original value of 0.2 is taken from the original paper/code, and is most likely inherited from the original SimBA work. As it is a fairly inappropriate tuning for this problem, we have also chosen the alternative step length of 2.0 based on the search described in Section A.4 and depicted in Fig. 7 of the appendix. We fix the same value of 2.0 for both GFCS and SimBA-ODS, as it has an identical meaning in both contexts, and is a fair like-for-like comparison.
We typically assume that authors have already presented their own papers in the best light possible, with parameters that they themselves optimised for the evaluation, and we always present results using their recommended settings as one of the (possibly multiple) options. However, as noted in Section A.2 of the appendix, we do sometimes take measures to fix competitors’ implementation issues, improving their performance in doing so (as with LeBA’s failure to interpolate between input domains of different sizes). Additionally, as ODS-RGF did not standardise its parameters against P-RGF in its own evaluation (in the manner we have done against SimBA-ODS), we have presented P-RGF using both its own settings and those of ODS-RGF, as discussed in the paper. This is itself a very interesting ablation, given its result.
|
null | false
| null |
How to make a traditional cup of English tea?
|
Add loose leaf tea to a china teapot and add near boiling water. Allow the tea to brew for 4-5 minutes. Following this time pour the tea through a tea strainer to catch the loose leaves and for the brewed tea to flow into a tea cup. Fill the cup three quarters high and add milk and sugar to your taste. Enjoy your traditional cup of English tea.
|
|
null | false
| null |
How do you score three points in basketball?
|
You score three points by making a basket from beyond the three point line, which is an arc which is drawn around each basket. By sinking the ball from far away from the basket, you are rewarded with an extra point.
|
|
null | false
| null |
Classify each of the following plates as either food or a cocktail: sushi, mojito, quiche, tacos, moscow mule, old fashioned, pizza, pasta, sandwich, pho, margarita, martini, ramen
|
From the list that has been provided, the following items to belong to the food category: sushi quiche, tacos, pizza, pasta, sandwich, pho and ramen. The rest of the items can be categorized as cocktails: mojito, moscow mule, old fashioned, margarita and martini.
|
|
null | false
| 363
|
Understanding what people are talking about and how they feel about it is valuable especially for industries which need to know the customers' opinions on their products. Aspect-Based Sentiment Analysis (ABSA) is a branch of sentiment analysis which deals with extracting the opinion targets (aspects) as well as the sentiment expressed towards them. For instance, in the sentence The spaghetti was out of this world., a positive sentiment is mentioned towards the target which is spaghetti. Performing these tasks requires a deep understanding of the language. Traditional machine learning methods such as SVM BIBREF2, Naive Bayes BIBREF3, Decision Trees BIBREF4, Maximum Entropy BIBREF5 have long been practiced to acquire such knowledge. However, in recent years due to the abundance of available data and computational power, deep learning methods such as CNNs BIBREF6, BIBREF7, BIBREF8, RNNs BIBREF9, BIBREF10, BIBREF11, and the Transformer BIBREF12 have outperformed the traditional machine learning techniques in various tasks of sentiment analysis. Bidirectional Encoder Representations from Transformers (BERT) BIBREF13 is a deep and powerful language model which uses the encoder of the Transformer in a self-supervised manner to learn the language model. It has been shown to result in state-of-the-art performances on the GLUE benchmark BIBREF14 including text classification. BIBREF1 show that adding domain-specific information to this model can enhance its performance in ABSA. Using their post-trained BERT (BERT-PT), we add adversarial examples to further improve BERT's performance on Aspect Extraction (AE) and Aspect Sentiment Classification (ASC) which are two major tasks in ABSA. A brief overview of these two sub-tasks is given in Section SECREF3.
Adversarial examples are a way of fooling a neural network to behave incorrectly BIBREF15. They are created by applying small perturbations to the original inputs. In the case of images, the perturbations can be invisible to human eye, but can cause neural networks to output a completely different response from the true one. Since neural nets make mistakes on these examples, introducing them to the network during the training can improve their performance. This is called adversarial training which acts as a regularizer to help the network generalize better BIBREF0. Due to the discrete nature of text, it is not feasible to produce perturbed examples from the original inputs. As a workaround, BIBREF16 apply this technique to the word embedding space for text classification. Inspired by them and building on the work of BIBREF1, we experiment with adversarial training for ABSA.
Our contributions are twofold. First, by carrying out an ablation study on the number of training epochs and the values for dropout in the classification layer, we show that there are values that outperform the specified ones for BERT-PT. Second, we introduce the application of adversarial training in ABSA by proposing a novel architecture which combines adversarial training with the BERT language model for AE and ASC tasks. Our experiments show that the proposed model outperforms the best performance of BERT-PT in both tasks.
Second, we introduce the application of adversarial training in ABSA by proposing a novel architecture which combines adversarial training with the BERT language model for AE and ASC tasks.
|
What is the novel architecture proposed by authors?
|
It combines adversarial training with the BERT language model for AE and Aspect Sentiment Classification tasks.
|
null | false
| null |
What are the Spice Girls nicknames?
|
The Spice Girls iconic nicknames are Sporty, Ginger, Scary, Baby, and Posh Spice.
|
|
null | false
| null |
Estonia, formally the Republic of Estonia, is a country by the Baltic Sea in Northern Europe. It is bordered to the north by the Gulf of Finland across from Finland, to the west by the sea across from Sweden, to the south by Latvia, and to the east by Lake Peipus and Russia. The territory of Estonia consists of the mainland, the larger islands of Saaremaa and Hiiumaa, and over 2,200 other islands and islets on the eastern coast of the Baltic Sea, covering a total area of 45,339 square kilometres (17,505 sq mi). The capital city Tallinn and Tartu are the two largest urban areas of the country. The Estonian language is the autochthonous and the official language of Estonia; it is the first language of the majority of its population, as well as the world's second most spoken Finnic language. The land of what is now modern Estonia has been inhabited by Homo sapiens since at least 9,000 BC. The medieval indigenous population of Estonia was one of the last pagan civilisations in Europe to adopt Christianity following the Papal-sanctioned Livonian Crusade in the 13th century. After centuries of successive rule by the Teutonic Order, Denmark, Sweden, and the Russian Empire, a distinct Estonian national identity began to emerge in the mid-19th century. This culminated in the 24 February 1918 Estonian Declaration of Independence from the then warring Russian and German Empires. Democratic throughout most of the interwar period, Estonia declared neutrality at the outbreak of World War II, but the country was repeatedly contested, invaded and occupied, first by the Soviet Union in 1940, then by Nazi Germany in 1941, and was ultimately reoccupied in 1944 by, and annexed into, the USSR as an administrative subunit (Estonian SSR). Throughout the 1944–1991 Soviet occupation, Estonia's de jure state continuity was preserved by diplomatic representatives and the government-in-exile. Following the bloodless Estonian "Singing Revolution" of 1988–1990, the nation's de facto independence from the Soviet Union was restored on 20 August 1991. Estonia is a developed country, with a high-income advanced economy. The sovereign state of Estonia is a democratic unitary parliamentary republic, administratively subdivided into 15 maakond (counties). With a population of just over 1.3 million, it is one of the least populous members of the European Union, the Eurozone, the OECD, the Schengen Area, and NATO. Estonia is nowadays often considered to be one of the three "Baltic countries" or "Baltic states".
|
Where is Estonia?
|
Estonia is a small European Union nation located in Northern Europe on the Baltic Sea across from Finland and bordered by Lativa to the south and Russia to the east. With a population of 1.3 million it is among the least populated EU countries. The country was occupied for much of the 20th century by Russia as part of the USSR. Today it is known for its high income advanced economy and as a member of the Eurozone, the OECD, the Schengen Area as well as NATO.
|
null | false
| null |
International Booker Prize 2021 was given to
|
At night all blood is black
|
|
null | false
| null |
Bieh State (previously named Eastern Bieh State) was a state in South Sudan that existed between 2 October 2015 and 22 February 2020. It was located in the Greater Upper Nile region and it bordered Fangak to the west, Central Upper Nile to the north, Latjoor and Akobo to the east, and Jonglei to the south.
History
On 2 October 2015, President Salva Kiir issued a decree establishing 28 states in place of the 10 constitutionally established states. The decree established the new states largely along ethnic lines. A number of opposition parties and civil society groups challenged the constitutionality of the decree. Kiir later resolved to take it to parliament for approval as a constitutional amendment. In November the South Sudanese parliament empowered President Kiir to create new states.
|
what is the Bieh State
|
Bieh State (previously named Eastern Bieh State) was in South Sudan between 2 October 2015 and 22 February 2020. It was located in the Greater Upper Nile region, bordering Fangak to the west, Central Upper Nile to the north, Latjoor and Akobo to the east, and Jonglei to the south. On 2 October 2015, President Salva Kiir issued a decree establishing 28 states instead of the ten constitutionally established states. The legislation established the new states primarily along ethnic lines.
|
1912.03457
| false
| null |
Often, many state-of-the-art tools cannot be applied to low-resource languages due to the lack of data. Table TABREF6 describes the various technologies and their presence concerning languages with different levels of resource availability and the ease of data collection. We can observe that for low resource languages, there is considerable difficulty in adopting these tools. Machine Translation can potentially be used as a fix to bridge the gap. Translation engines can help in translating documents from minority languages to majority languages. This allows the pool of data to be used in a number of NLP tasks like sentiment analysis and summarization. Doing so allows us to leverage the existing body of work in NLP done on resource-rich languages and subsequently apply it to the resource-poor languages, thereby foregoing any attempt to reinvent the wheel for these languages. This ensures a quicker and wider impact.BIBREF16 performs sentiment analysis on Chinese customer reviews by translating them to English. They observe that the quality of machine translation systems are sufficient for sentiment analysis to be performed on the automatically translated texts without a substantial trade-off in accuracy.
FLOAT SELECTED: Table 1: Enabling language technologies, their availability and quality ( ? ? ? - excellent quality technology, ?? - moderately good but usable, ? - rudimentary and not practically useful) for differently resourced languages, and their data/knowledge requirements (? ? ? - very high data/expertise, ?? - moderate, ? - nominal and easily procurable). This information is based on authors’ analysis and personal experience.
Table TABREF6 describes the various technologies and their presence concerning languages with different levels of resource availability and the ease of data collection. We can observe that for low resource languages, there is considerable difficulty in adopting these tools.
FLOAT SELECTED: Table 1: Enabling language technologies, their availability and quality ( ? ? ? - excellent quality technology, ?? - moderately good but usable, ? - rudimentary and not practically useful) for differently resourced languages, and their data/knowledge requirements (? ? ? - very high data/expertise, ?? - moderate, ? - nominal and easily procurable). This information is based on authors’ analysis and personal experience.
|
What language technologies have been introduced in the past?
|
- Font & Keyboard
- Speech-to-Text
- Text-to-Speech
- Text Prediction
- Spell Checker
- Grammar Checker
- Text Search
- Machine Translation
- Voice to Text Search
- Voice to Speech Search
|
null | false
| 516
|
As far as we know, there are no existing methods designed and evaluated specifically for our sourcefree few-shot DA setting. We apply existing methods in DA and few-shot transfer learning to provide benchmark performance. We compare with AdaBN by replacing source BN statistics with those calculated on target support set, finetuning the source model on BN layers or classifier or feature extractor, finetuning entire model with L 2 or L 2 -SP or DELTA regularization, Late Fusion which averages scores from source and target classifier, and FLUTE which optimizes BN parameters with nearest-centroid classifier. FLUTE assumes the availability of multiple source datasets to train multiple sets of BN parameters for further blending to initialize the finetuning process. Since we only have access to the pre-trained source model in our setting, we reduce FLUTE to the single source dataset case and initialize FLUTE with single source BN parameters. It is conventional to update the classifier when sufficient target samples are available, hence besides implementing with source classifier (SC), we also implement our proposed method with finetuned classifier (FC) and nearestcentroid classifier (NCC). We follow learning settings in for regularized finetuning and use SGD optimizer with momentum and weight decay 0.0004, and L 2 regularization is also added to finetuning on classifier or feature extractor to prevent overfitting. For all other methods that finetune on BN layers, we use Adam optimizer following. We set learning rate 0.001, mini-batch size 32 and epochs 10 for all methods and datasets evaluated.
From Table, we observe that regularized finetuning tends to adapt well at k = 1, but performance can lag behind with larger support sets. AdaBN does not consistently improve adaptation; we show in Appendix Figure that completely replacing source BN statistics degrades performance for VisDA. Overall, finetuning LCCS with source classifier has at least comparable performance with regularized finetuning for small support sets. For larger support sets, finetuning LCCS with NCC has the best performance, with classification accuracy consistently higher than FLUTE which also uses NCC. We also compare our method to L 2 and FLUTE on Office and OfficeHome in Table 16. Though the transfer learning methods produce comparable performance on the easy dataset of Camelyon17 (two classes, two domains), LCCS outperforms on more difficult datasets, which demonstrates the effectiveness of the proposed low-dimensional finetuning strategy.
LCCS Finetuning vs BN Finetuning: We further study our proposed constrained optimization of BN layer versus the unconstrained alternative of optimizing directly on the high-dimensional BN parameters γ and β in Equation 1, with fixed source classifier. We also evaluate a combination of the two strategies by first finetuning LCCS followed by finetuning BN parameters. From Figure, we see that when the support set is small at k ≤ 3, finetuning LCCS is always better than finetuning BN parameters. The results are mixed at larger k, which is expected since the unconstrained optimization is also less prone to overfit with more samples. In addition, we observe that finetuning LCCS first provides good initialization for further finetuning of BN parameters, and consistently achieves better adaptation performance than finetuning BN parameters alone. We make similar observations when using NCC as classifier in Appendix Section C.5. Figure: Finetuning LCCS has best performance when support set is extremely small (k ≤ 3).
With larger support set (k ≥ 5), finetuning LCCS+BN parameters attains better performance than finetuning BN parameters alone.
We additionally compare our proposed method with source-free UDA methods which adapts with entire unlabelled target dataset, including AdaBN, SHOT, SFDA and SDDA, as well as the few-shot method L 2 and FLUTE, in Table on Office and Office-Home. We observe that self-supervision over the entire unlabelled target can produce good adaptation performance. Despite using only 5 samples per class, finetuning LCCS has equal and better adaptation performance than source-free UDA methods in 5 out of 6 domain pairs on Office. SHOT outperforms in the more challenging OfficeHome dataset, which reflects the difficulty of source-free few-shot setting. However, our proposed method performs the best out of the few-shot methods evaluated, which demonstrates its effectiveness in the few-shot setting. We refer readers to Appendix Section C.4 for detailed results and comparisons with nonsource-free few-shot DA methods.
We compare the resources of the three source-free methods from different settings: SHOT, Tent, and our fewshot LCCS finetuning. From Table, the number and storage size of samples for our proposed method is much smaller than those used for SHOT. At test time, our method requires no computation for updating the model, which is at least on par with test-time adaptation methods since they can need additional backward passes for online adaptation. For instance, Tent needs at least an additional backward pass. More importantly, our setting is one way to address test-time adaptation's reliance on streaming conditions which may challenge its usefulness in practice.
We apply existing methods in DA and few-shot transfer learning to provide benchmark performance. We compare with AdaBN (Li et al., 2016) by replacing source BN statistics with those calculated on target support set, finetuning the source model on BN layers or classifier or feature extractor, finetuning entire model with L2 or L2 -SP (Li et al., 2018) or DELTA (Li et al., 2019) regularization, Late Fusion (Hoffman et al., 2013) which averages scores from source and target classifier, and FLUTE (Triantafillou et al., 2021) which optimizes BN parameters with nearest-centroid classifier.****We additionally compare our proposed method with source-free UDA methods which adapts with entire unlabelled target dataset, including AdaBN (Li et al., 2016), SHOT (Liang et al., 2020), SFDA (Kim et al., 2020) and SDDA (Kurmi et al., 2021), as well as the few-shot method L2 and FLUTE, in Table 6 on Office and Office-Home.****Additionally, we compare with state-of-the-art source-free UDA method SHOT (Liang et al., 2020), which adapts on the entire unlabelled target dataset, on OfficeHome. From Table 16, SHOT performs better than finetuning LCCS in most domain pairs.
|
Whether you miss any baselines?
|
We have added Table 5 in Section 4.2 for source-free few-shot adaptation comparison. We include AdaBN, Late Fusion, finetuning methods (finetune BN/classifier/feature extractor, regularized finetuning), cross-domain few-shot learning (FLUTE), hypothesis transfer (finetune BN/feature extractor). Note that finetuning feature extractor is equivalent to SHOT (Liang et al., 2020) with supervised instead of unsupervised loss. We have also included additional comparisons with source-free UDA which uses entire unlabelled target dataset in Table 6 and non-source-free few-shot DA in Appendix Section C.4. Overall, our proposed method achieves competitive performance against other source-free few-shot methods. Source-free UDA can achieve better performance than our proposed method in some cases, but at a cost of less data-efficiency and increased model update computation.
|
null | false
| null |
What s epigenomics?
|
Epigenomics is the study of the complete set of epigenetic modifications on the genetic material of a cell, known as the epigenome. The field is analogous to genomics and proteomics, which are the study of the genome and proteome of a cell. Epigenetic modifications are reversible modifications on a cell's DNA or histones that affect gene expression without altering the DNA sequence. Epigenomic maintenance is a continuous process and plays an important role in stability of eukaryotic genomes by taking part in crucial biological mechanisms like DNA repair. Plant flavones are said to be inhibiting epigenomic marks that cause cancers. Two of the most characterized epigenetic modifications are DNA methylation and histone modification. Epigenetic modifications play an important role in gene expression and regulation, and are involved in numerous cellular processes such as in differentiation/development and tumorigenesis. The study of epigenetics on a global level has been made possible only recently through the adaptation of genomic high-throughput assays.
|
|
null | false
| 214
|
The growing popularity of online interactions through social media has been shown to have both positive and negative impacts. While social media improves information sharing, it also facilitates the propagation of online harassment, including hate speech. These negative experiences can have a measurable negative impact on users. Recently, the Pew Research Center BIBREF0 reported that “roughly four-in-ten Americans have personally experienced online harassment, and 63% consider it a major problem.”
To address the growing problem of online hate, an extensive body of work has focused on developing automatic hate speech detection models and datasets BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7, BIBREF8. However, simply detecting and blocking hate speech or suspicious users often has limited ability to prevent these users from simply turning to other social media platforms to continue to engage in hate speech as can be seen in the large move of individuals blocked from Twitter to Gab BIBREF9. What's more, such a strategy is often at odds with the concept of free speech. As reported by the Pew Research Center BIBREF0, “Despite this broad concern over online harassment, 45% of Americans say it is more important to let people speak their minds freely online; a slightly larger share (53%) feels that it is more important for people to feel welcome and safe online.” The special rapporteurs representing the Office of the United Nations High Commissioner for Human Rights (OHCHR) have recommended that “The strategic response to hate speech is more speech.” BIBREF10 They encourage to change what people think instead of merely changing what they do, so they advocate more speech that educates about cultural differences, diversity, and minorities as a better strategy to counter hate speech.
Therefore, in order to encourage strategies of countering online hate speech, we propose a novel task of generative hate speech intervention and introduce two new datasets for this task. Figure FIGREF5 illustrates the task. Our datasets consist of 5K conversations retrieved from Reddit and 12k conversations retrieved from Gab. Distinct from existing hate speech datasets, our datasets retain their conversational context and introduce human-written intervention responses. The conversational context and intervention responses are critical in order to build generative models to automatically mitigate the spread of these types of conversations.
To summarize, our contributions are three-fold:
We introduce the generative hate speech intervention task and provide two fully-labeled hate speech datasets with human-written intervention responses.
Our data is collected in the form of conversations, providing better context.
The two data sources, Gab and Reddit, are not well studied for hate speech. Our datasets fill this gap.
Due to our data collecting strategy, all the posts in our datasets are manually labeled as hate or non-hate speech by Mechanical Turk workers, so they can also be used for the hate speech detection task. The performance of commonly-used classifiers on our datasets is shown in Section SECREF6.
The conversational context and intervention responses are critical in order to build generative models to automatically mitigate the spread of these types of conversations.
|
Why do the authors' datasets retain the conversational context and introduce human-written intervention responses?
|
Because the conversational context and intervention responses are critical in order to build generative models to automatically mitigate the spread of these types of conversations.
|
null | false
| null |
Ford Motor Company (commonly known as Ford) is an American multinational automobile manufacturer headquartered in Dearborn, Michigan, United States. It was founded by Henry Ford and incorporated on June 16, 1903. The company sells automobiles and commercial vehicles under the Ford brand, and luxury cars under its Lincoln luxury brand. Ford also owns Brazilian SUV manufacturer Troller, an 8% stake in Aston Martin of the United Kingdom and a 32% stake in China's Jiangling Motors. It also has joint ventures in China (Changan Ford), Taiwan (Ford Lio Ho), Thailand (AutoAlliance Thailand), and Turkey (Ford Otosan). The company is listed on the New York Stock Exchange and is controlled by the Ford family; they have minority ownership but the majority of the voting power.
|
Who founded Ford
|
Henry Ford founded Ford Motor Company (aka Ford) in on June 16, 1903 in Dearborn Michigan.
|
null | false
| null |
What is the best way to approach a new cat?
|
First, it's important to know the background of the cat. If it is a stray, you might want to keep your distance since you don't know its vaccination status. If it is a known, healthy cat, then you can approach it. Do so with care so that you do not scare it away.
There are several techniques to make yourself more approachable. Ideally, the cat will approach you; you will not approach the cat.
Make yourself look smaller. Crouch down, and do not look threatening or overly anxious to approach the cat.
Try the "slow blink." Look at the cat and slowly close and open your eyes. Also try looking away lazily. If the cat respond with its own slow blink, that's a good sign.
If the cat comes near, try offering your hand down low, with the palm turned down. Hopefully the cat will come to you and rub against the back of your hand. If so, then you can try petting the cat more actively.
|
|
1604.05781
| false
| null |
Causal documents were chosen to contain one occurrence only of the exact unigrams: `caused', `causing', or `causes'. The word `cause' was not included due to its use as a popular contraction for `because'. One `cause-word' per document restricted the analysis to single relationships between two relata. Documents that contain bidirectional words (`associate', `relate', `connect', `correlate', and any of their stems) were also not selected for analysis. This is because our focus is on causality, an inherently one-sided relationship between two objects. We also did not consider additional synonyms of these cause words, although that could be pursued for future work. Control documents were also selected. These documents did not contain any of `caused', `causing', or `causes', nor any bidirectional words, and are further matched temporally to obtain the same number of control documents as causal documents in each fifteen-minute period during 2013. Control documents were otherwise selected randomly; causal synonyms may be present. The end result of this procedure identified 965,560 causal and 965,560 control documents. Each of the three “cause-words”, `caused', `causes', and `causing' appeared in 38.2%, 35.0%, and 26.8% of causal documents, respectively.
Causal documents were chosen to contain one occurrence only of the exact unigrams: `caused', `causing', or `causes'.
|
What are the selection criteria for "causal statements"?
|
Presence of only the exact unigrams 'caused', 'causing', or 'causes'
|
null | false
| null |
What type of jacket do grizzlies usually wear?
|
Grizzlies are bears typically found in North America. Bears are animals that are covered in fur and usually live in the woods. Because they are covered in fur and live in the woods, it is unnecessary for bears to wear jackets. To date, bears have not shown the capacity to be able to create cloth or sew jackets, so they would have to get them from humans. Humans are usually afraid of bears and are not likely to approach a bear with a jacket. Therefore bears usually do not wear jackets.
|
|
null | false
| null |
Who were the last 5 Republican presidents?
|
Gerald Ford – 1974 to 1977 (38th President)
Ronald Reagan – 1981 to 1989 (40th President)
George H.W. Bush – 1989 to 1993 (41st President)
George W. Bush – 2001 to 2009 (43rd President)
Donald J. Trump – 2017 to 2021 (45th President)
|
|
null | false
| null |
Identify which instrument is string or percussion: Bongo drum, Kingri
|
Kingri is string, Bongo drum is percussion.
|
|
null | false
| null |
The New JEWEL Movement (NJM) was formally established on 11 March 1973 as an alliance of the Joint Endeavor for Welfare, Education, and Liberation (JEWEL), Organization for Revolutionary Education and Liberation, and the Movement for Assemblies of the People (MAP), led by young lawyer Maurice Bishop. The NJM's initial manifesto was largely drafted by MAP's major intellectual, Franklyn Harvey, who had been heavily influenced by the writings of C.L.R. James. From 1973 to 1979, the NJM was an opposition political party active in Grenada. During the 1970s, the political situation in Grenada became increasingly polarized and violent. For the 1976 general elections the organisation formed an electoral coalition known as the People's Alliance with the Grenada National Party and the United People's Party. However, the alliance lost to the ruling Grenada United Labour Party in elections which were branded fraudulent by international observers. In the late 1970s, the NJM formed the National Liberation Army (NLA), also known as "the 12 Apostles".
|
Based on this information given, what groups combined to create the New JEWEL Movement, and on what date were they organized by which leader?
|
On March 11th 1973, Maurice Bishop brought together the Joint Endeavor for Welfare, Education, and Liberation (JEWEL), Organization for Revolutionary Education and Liberation, and the Movement for Assemblies of the People to form the New JEWEL Movement.
|
null | false
| null |
In 1988, due to the relationship he had built up with Honda throughout the 1987 season with Lotus, and with the approval of McLaren's number-one driver and then-double world champion, Alain Prost, Senna joined the McLaren team. The foundation for a fierce competition between Senna and Prost was laid, culminating in a number of dramatic race incidents between the two over the next five years. However, the experienced pair also quickly realized, despite their personal rivalry, they had to work together, especially in testing, to keep ahead of their main opposition from Ferrari, Williams, Benetton and Lotus.
|
Given this text, who was Mclaren's number-one driver in 1988
|
In 1988, according to this text Alain Prost was McLaren's number one driver
|
null | false
| null |
Stanley Franklin Browne, DFC & Bar (29 October 1919 – 13 January 2011) was a New Zealand flying ace of the Royal New Zealand Air Force (RNZAF) during the Second World War. He was officially credited with at least five aerial victories.
Born in Wellington, Browne was a university student when he joined the RNZAF in March 1941. Once his flight training in New Zealand was completed, he proceeded to the United Kingdom to serve with the Royal Air Force. He was posted to No. 485 Squadron in early 1942 but was shot down later in the year over Germany-occupied France. With the help of the French Resistance he avoided the Germans and made his way to the south of France. He was caught when crossing the border into Vichy France and was interned for several weeks. He escaped captivity and eventually back to the United Kingdom. He was transferred to the Middle East where he ferried aircraft before being posted to No. 93 Squadron. He flew Supermarine Spitfires in the Tunisian campaign and in the Allied invasion of Sicily, shooting down a number of German aircraft before being rested in December 1943. After a spell on instructing duties he returned to operational flying with No. 485 Squadron during the final stages of the war in Europe. He briefly commanded the squadron before it was disbanded in August 1945.
Returning to civilian life after the war, he returned to his university studies, becoming a chemist. He briefly served in the New Zealand Territorial Air Force and later became a farmer and then a teacher. He died in 2011, aged 91, at Tauranga.
|
Given a reference text about Stanley Franklin Browne, how many aerial victories during World War II did he have and what did he do when he returned from the war.
|
Stanley Franklin Browne was credited with five aerial victories during World War II. After the war, he returned to university and became a chemist.
|
null | false
| null |
Isa Melikov (Azerbaijani: Məlikov İsa Fazil oğlu, born 21 October 1980, in Baku, Azerbaijan SSR) is a famous Azerbaijani composer and producer. He is the author of numerous musical projects, hit songs and soundtracks to movies. The list of celebrities singing his songs include leading stars of Azerbaijan as well as vocalists from CIS and Europe such as Dima Bilan, Eldar Gasimov, Nikki Jamal, Glenys Vargas, Kevin Etienne, Elli, Aysel, Gunesh, Zulfiyya Khanbabayeva, Aygün Kazımova, Elnara, Röya, Elşad Xose, Sevda Alekperzade, Manana, Faig Aghayev, Tunzala Agayeva, Zamig Huseynov, Malik Kalantarli, etc.
Biography
He was born in 1980, October 21 in Baku. The parents of future composer were teaching him love for music from the very early years of his life. Even as a child Isa knew by heart the lyrics of many international hits and by the age of 6 he already was trying to perform them on the piano. In 1987, he went to the secondary school #189 in Baku. In 1990 he became a student of the Baku musical school #1 named after Vagif Mustafazade.
In 1995 he entered the Musical College named after Asaf Zeynalli. Afterwards, in 1997 he entered the Azerbaijan State University of Culture and Arts, and graduated with honors in the specialty in 2001: the conductor of the orchestra of folk musical instruments. In 2003, he got the Master of Arts degree.
Starting from 2002, top introduce a new music genre R & B to Azerbaijani market, he started working on it. In 2003, for the first time in Azerbaijan, he wrote a song in R & B style named "Qəm Otaği" for the singer Elnara, and by this aimed to popularize the genre of R & B in Azerbaijan.
In 2004 he became a member of the British Academy of Composers and songwriters.
In 2005, for the first time he synthesized Azerbaijani folklore (mugham) music with R & B style in song "Uçduq", performed by singer Zamig.
Since 2006 becomes a dissertator at Baku Music Academy named after Uzeyir Hajibeyov.
In 2006 he became a musical producer of the largest Azerbaijani television project in a reality show format called "Academy".
In 2008, he participated in Berlinale Talent Campus holding in the frame of 58th Berlin Film Festival, where he met and received a master class from two time "Oscar" winner composer Gustavo Santaolalla.
In 2008, he established the production center "Baku Music Factory", and became its general producer.
The first project of BMF production center wins third place at Eurovision Song Contest 2009 with the participation of the singer Aysel, under the management of Isa Melikov.
In 2009, he created Azerbaijan's first official digital publishing and record label "BMF Records".
In 2011, BMF Records project wins Eurovision Song Contest 2011 with the participation of Nigar Jamal featuring Eldar Gasimov took with «Running Scared» song.
|
Given this reference text about Isa Melikov, which events did Isa participate in 2008?
|
Isa participated in Berlinale Talent Campus, and established the production center Baku Music Factory in 2008
|
null | false
| 3
|
Deep Learning approaches have achieved impressive results on various NLP tasks BIBREF0 , BIBREF1 , BIBREF2 and have become the de facto approach for any NLP task. However, these deep learning techniques have found to be less effective for low-resource languages when the available training data is very less BIBREF3 . Recently, several approaches like Multi-task learning BIBREF4 , multilingual learning BIBREF5 , semi-supervised learning BIBREF2 , BIBREF6 and transfer learning BIBREF7 , BIBREF3 have been explored by the deep learning community to overcome data sparsity in low-resource languages. Transfer learning trains a model for a parent task and fine-tunes the learned parent model weights (features) for a related child task BIBREF7 , BIBREF8 . This effectively reduces the requirement on training data for the child task as the model would have learned relevant features from the parent task data thereby, improving the performance on the child task.
Transfer learning has also been explored in the multilingual Neural Machine Translation BIBREF3 , BIBREF9 , BIBREF10 . The goal is to improve the NMT performance on the source to target language pair (child task) using an assisting source language (assisting to target translation is the parent task). Here, the parent model is trained on the assisting and target language parallel corpus and the trained weights are used to initialize the child model. The child model can now be fine-tuned on the source-target language pairs, if parallel corpus is available. The divergence between the source and the assisting language can adversely impact the benefits obtained from transfer learning. Multiple studies have shown that transfer learning works best when the languages are related BIBREF3 , BIBREF10 , BIBREF9 . Several studies have tried to address lexical divergence between the source and the target languages BIBREF10 , BIBREF11 , BIBREF12 . However, the effect of word order divergence and its mitigation has not been explored. In a practical setting, it is not uncommon to have source and assisting languages with different word order. For instance, it is possible to find parallel corpora between English and some Indian languages, but very little parallel corpora between Indian languages. Hence, it is natural to use English as an assisting language for inter-Indian language translation.
To see how word order divergence can be detrimental, let us consider the case of the standard RNN (Bi-LSTM) encoder-attention-decoder architecture BIBREF13 . The encoder generates contextual representations (annotation vectors) for each source word, which are used by the attention network to match the source words to the current decoder state. The contextual representation is word-order dependent. Hence, if the assisting and the source languages do not have similar word order the generated contextual representations will not be consistent. The attention network (and hence the decoder) sees different contextual representations for similar words in parallel sentences across different languages. This makes it difficult to transfer knowledge learned from the assisting language to the source language.
We illustrate this by visualizing the contextual representations generated by the encoder of an English to Hindi NMT system for two versions of the English input: (a) original word order (SVO) (b) word order of the source language (SOV, for Bengali). Figure FIGREF1 shows that the encoder representations obtained are very different. The attention network and the decoder now have to work with very different representations. Note that the plot below does not take into account further lexical and other divergences between source and assisting languages, since we demonstrated word order divergence with the same language on the source side.
To address this word order divergence, we propose to pre-order the assisting language sentences to match the word order of the source language. We consider an extremely resource constrained scenario, where we do not have any parallel corpus for the child task. We are limited to a bilingual dictionary for transfer information from the assisting to the source language. From our experiments, we show that there is a significant increase in the translation accuracy for the unseen source-target language pair.
From our experiments, we show that there is a significant increase in the translation accuracy for the unseen source-target language pair.
|
According to the experiment, what did the authors find that has an impact on the accuracy of translation?
|
The authors show that there is a significant increase in the translation accuracy for the unseen source-target language pair.
|
1911.03385
| false
| null |
fig:sentenceinput illustrates the process. Controls are calculated heuristically. All words found in the control word lists are then removed from the reference sentence. The remaining words, which represent the content, are used as input into the model, along with their POS tags and lemmas.
In this way we encourage models to construct a sentence using content and style independently. This will allow us to vary the stylistic controls while keeping the content constant, and successfully perform style transfer. When generating a new sentence, the controls correspond to the counts of the corresponding syntactic features that we expect to be realized in the output.
Controls are calculated heuristically. All words found in the control word lists are then removed from the reference sentence. The remaining words, which represent the content, are used as input into the model, along with their POS tags and lemmas.
In this way we encourage models to construct a sentence using content and style independently.
|
How they know what are content words?
|
The answers are shown as follows:
* words found in the control word lists are then removed
* The remaining words, which represent the content
|
null | false
| null |
1947: Two plane crashes in a 24-hour period, one with fatalities. Two people were killed as the result of a mid-air collision between two planes over the airfield. The accident occurred when William Gleason Sauerwein of Falls Church took off from the field in an Aeronca Champion at the same time that F.H. Kirchman, 48, of Vienna, Virginia, was attempting to land while flying a Stinson 105. Both Kirchman and his passenger, Agnes Louise Brady, 47, of Arlington, Virginia, were killed in the accident. Sauerwein was taken to Arlington County hospital with head injuries and a broken leg. Witnesses said that the accident occurred at 7pm when the Stinson cut into the tail of the Aeronca while the planes were approximately 75 feet off the ground. The next day, a second crash occurred when a Funk cabin airplane lost power shortly after takeoff. The pilot, Margaret Vaughn, of Washington, D.C., suffered severe head and leg injuries when the plane crashed into a wooded area about two miles west of the airport, near Gallows Road in Fairfax County. Vaughn was an experienced pilot and was a Link trainer instructor for the Civil Aeronautics Authority. The next day, the Civil Aeronautics Board announced they were conducting a "full inquiry" into the cause of the two crashes. The investigators were specifically concerned with the quality of the gasoline being dispensed at the airport and took samples from Vaughn's aircraft at the scene of the crash.
|
Who was injured or killed and in what ways in the accidents?
|
F.H. Kirchman and his passenger Agnes Louise Brady was also killed. William Gleason Sauerwein had a broken leg and head injuries. Margaret Vaughn had severe head and leg injuries.
|
null | false
| null |
Who is the largest employer in the world?
|
The United States Department of Defense (DoD, USDOD or DOD) is an executive branch department of the federal government of the United States charged with coordinating and supervising all agencies and functions of the U.S. government directly related to national security and the United States Armed Forces. The DoD is the largest employer in the world, with over 1.34 million active-duty service members (soldiers, marines, sailors, airmen, and guardians) as of June 2022. The DoD also maintains over 778,000 National Guard and reservists, and over 747,000 civilians bringing the total to over 2.87 million employees. Headquartered at the Pentagon in Arlington, Virginia, just outside Washington, D.C., the DoD's stated mission is to provide "the military forces needed to deter war and ensure our nation's security".
|
|
null | false
| null |
On what day did the invasion of Normandy, France occur in World War II?
|
June 6, 1944
|
|
null | false
| null |
What is a Mormon?
|
A Mormon typically refers to a member of the Church of Jesus Christ of Latter-day Saints. The word Mormon comes from a prophet in the Book of Mormon who lived in the Americas around 300 A.D. The Book of Mormon is part of the canon of scripture that members of the Church of Jesus Christ of Latter-day Saints use, in conjunction with the Bible and the Doctrine and Covenants. It is often viewed as a distinguishing difference between the Church of Jesus Christ of Latter-day Saints and other Christian denominations. Historically, this is why members of the Church of Jesus Christ of Latter-day Saints have been called Mormons.
|
|
null | false
| null |
Wine is an alcoholic drink typically made from fermented grapes. Yeast consumes the sugar in the grapes and converts it to ethanol and carbon dioxide, releasing heat in the process. Different varieties of grapes and strains of yeasts are major factors in different styles of wine. These differences result from the complex interactions between the biochemical development of the grape, the reactions involved in fermentation, the grape's growing environment (terroir), and the wine production process. Many countries enact legal appellations intended to define styles and qualities of wine. These typically restrict the geographical origin and permitted varieties of grapes, as well as other aspects of wine production. Wines can be made by fermentation of other fruit crops such as plum, cherry, pomegranate, blueberry, currant and elderberry.
|
What is wine?
|
Wine is an alcoholic drink typically made from fermented grapes.
|
null | false
| null |
A Moon landing is the arrival of a spacecraft on the surface of the Moon. This includes both crewed and robotic missions. The first human-made object to touch the Moon was the Soviet Union's Luna 2, on 13 September 1959.
|
What is the definition of a moon landing
|
A Moon landing is the arrival of a spacecraft on the surface of the Moon. This includes both crewed and robotic missions. The first human-made object to touch the Moon was the Soviet Union's Luna 2, on 13 September 1959.
The United States' Apollo 11 was the first crewed mission to land on the Moon, on 20 July 1969. There were six crewed U.S. landings between 1969 and 1972, and numerous uncrewed landings, with no soft landings happening between 22 August 1976 and 14 December 2013.
The United States is the only country to have successfully conducted crewed missions to the Moon, with the last departing the lunar surface in December 1972. All soft landings took place on the near side of the Moon until 3 January 2019, when the Chinese Chang'e 4 spacecraft made the first landing on the far side of the Moon.
|
1705.07368
| false
| null |
I first measured the effectiveness of the embeddings at the skip-gram's training task, predicting context words INLINEFORM0 given input words INLINEFORM1 . This task measures the methods' performance for predictive language modeling. I used four datasets of sociopolitical, scientific, and literary interest: the corpus of NIPS articles from 1987 – 1999 ( INLINEFORM2 million), the U.S. presidential state of the Union addresses from 1790 – 2015 ( INLINEFORM3 ), the complete works of Shakespeare ( INLINEFORM4 ; this version did not contain the Sonnets), and the writings of black scholar and activist W.E.B. Du Bois, as digitized by Project Gutenberg ( INLINEFORM5 ). For each dataset, I held out 10,000 INLINEFORM6 pairs uniformly at random, where INLINEFORM7 , and aimed to predict INLINEFORM8 given INLINEFORM9 (and optionally, INLINEFORM10 ). Since there are a large number of classes, I treat this as a ranking problem, and report the mean reciprocal rank. The experiments were repeated and averaged over 5 train/test splits.
I first measured the effectiveness of the embeddings at the skip-gram's training task, predicting context words INLINEFORM0 given input words INLINEFORM1 . This task measures the methods' performance for predictive language modeling. I used four datasets of sociopolitical, scientific, and literary interest: the corpus of NIPS articles from 1987 – 1999 ( INLINEFORM2 million), the U.S. presidential state of the Union addresses from 1790 – 2015 ( INLINEFORM3 ), the complete works of Shakespeare ( INLINEFORM4 ; this version did not contain the Sonnets), and the writings of black scholar and activist W.E.B. Du Bois, as digitized by Project Gutenberg ( INLINEFORM5 ). For each dataset, I held out 10,000 INLINEFORM6 pairs uniformly at random, where INLINEFORM7 , and aimed to predict INLINEFORM8 given INLINEFORM9 (and optionally, INLINEFORM10 ). Since there are a large number of classes, I treat this as a ranking problem, and report the mean reciprocal rank. The experiments were repeated and averaged over 5 train/test splits.
|
What is MRR?
|
The answers are shown as follows:
* mean reciprocal rank
|
null | false
| null |
The Patriots hold training camp and practices just outside of Gillette Stadium in Foxborough, with twin practice fields available for team use.
Prior to 2003, the Patriots held training camp and practice at numerous locations. From 1976 to 2002, the team held training camp at Bryant College in Smithfield, Rhode Island. From 1960 to 1961, then from 1969 to 1975, the Patriots held training camp at University of Massachusetts Amherst. Between 1962 and 1968, the Patriots held training camp at Phillips Academy in Andover, Massachusetts.
From 1971 until 2001, the Patriots would simply practice at Foxboro Stadium whenever the field was available, otherwise they would use the public football fields that were available in Foxborough.
|
Extract from this passage the the locations where the Patriots have held training camp and separate the answers with commas.
|
Outside Gillette Stadium, Bryant College, University of Massachusetts Amherst, Philips Academy, Foxboro Stadium, the public football fields available in Foxborough
|
null | false
| null |
Tell me whether these games are exclusives or multi-platform: The Legend of Zelda, Call of Duty, Sly Cooper, God of War, Counter-Strike, Mario, Pokemon, Halo, Dota, League of Legends.
|
The Legend of Zelda, Sly Cooper, God of War, Mario, Pokemon and Halo have been at least been historically advertised and classified as console exclusives: Zelda, Mario and Pokemon on Nintendo platforms, God of War and Sly Cooper on Sony Playstation platforms, Halo on Microsoft Xbox Platforms.
Lately, some of the games above, God of War and Halo, have been ported PC which makes them multi-platform games. God of War is considered as a timed exclusive to Playstation systems where the game was exclusive to the console for several years before being released to other platforms, in this case PC.
Halo has recently became PC and Xbox multi-platform dur to Microsoft's push to bring more of their console games to their PC gaming platform.
Rest of the games, Call of Duty, Counter-Strike, Dota and League of Legends are considered multi-platform as they have been released to multiple types of systems without ever being advertised as exclusive to one.
Be mindful that releasing a game to Windows PC and Apple MacOS is considered as multi-platform.
|
|
null | false
| null |
Tell me whether these states are in Eastern Time, Central Time, or neither
|
California, Oregon, Connecticut, Washington, Texas, Hawaii
|
|
null | false
| 46
|
Question answering (QA) has drawn a lot of attention in the past few years. QA tasks on images BIBREF0 have been widely studied, but most focused on understanding text documents BIBREF1 . A representative dataset in text QA is SQuAD BIBREF1 , in which several end-to-end neural models have accomplished promising performance BIBREF2 . Although there is a significant progress in machine comprehension (MC) on text documents, MC on spoken content is a much less investigated field. In spoken question answering (SQA), after transcribing spoken content into text by automatic speech recognition (ASR), typical approaches use information retrieval (IR) techniques BIBREF3 to find the proper answer from the ASR hypotheses. One attempt towards QA of spoken content is TOEFL listening comprehension by machine BIBREF4 . TOEFL is an English examination that tests the knowledge and skills of academic English for English learners whose native languages are not English. Another SQA corpus is Spoken-SQuAD BIBREF5 , which is automatically generated from SQuAD dataset through Google Text-to-Speech (TTS) system. Recently ODSQA, a SQA corpus recorded by real speakers, is released BIBREF6 .
To mitigate the impact of speech recognition errors, using sub-word units is a popular approach for speech-related downstream tasks. It has been applied to spoken document retrieval BIBREF7 and spoken term detection BIBREF8 The prior work showed that, using phonectic sub-word units brought improvements for both Spoken-SQuAD and ODSQA BIBREF5 .
Instead of considering sub-word features, this paper proposes a novel approach to mitigate the impact of ASR errors. We consider reference transcriptions and ASR hypotheses as two domains, and adapt the source domain data (reference transcriptions) to the target domain data (ASR hypotheses) by projecting these two domains in the shared common space. Therefore, it can effectively benefit the SQA model by improving the robustness to ASR errors in the SQA model.
Domain adaptation has been successfully applied on computer vision BIBREF9 and speech recognition BIBREF10 . It is also widely studied on NLP tasks such as sequence tagging and parsing BIBREF11 , BIBREF12 , BIBREF13 . Recently, adversarial domain adaptation has already been explored on spoken language understanding (SLU). Liu and Lane learned domain-general features to benefit from multiple dialogue datasets BIBREF14 ; Zhu et al. learned to transfer the model from the transcripts side to the ASR hypotheses side BIBREF15 ; Lan et al. constructed a shared space for slot tagging and language model BIBREF16 . This paper extends the capability of adversarial domain adaptation for SQA, which has not been explored yet.
Instead of considering sub-word features, this paper proposes a novel approach to mitigate the impact of ASR errors.
|
What is the highlight of their approach?
|
They propose a novel approach to mitigate the impact of ASR errors in stead of considering sub-word features.
|
null | false
| null |
How many events were completed in first modern Olympic Games?
|
There were 43 events which included shot put, boxing, running...
|
|
null | false
| 22
|
Evaluation-I: Strategy Formulation Ability. Table 5 shows the list of inference strategies formulated by LiLi for various INLINEFORM0 and INLINEFORM1 , which control the strategy formulation of LiLi. When INLINEFORM2 , LiLi cannot interact with user and works like a closed-world method. Thus, INLINEFORM3 drops significantly (0.47). When INLINEFORM4 , i.e. with only one interaction per query, LiLi acquires knowledge well for instances where either of the entities or relation is unknown. However, as one unknown entity may appear in multiple test triples, once the entity becomes known, LiLi doesn’t need to ask for it again and can perform inference on future triples causing significant increase in INLINEFORM5 (0.97). When INLINEFORM6 , LiLi is able to perform inference on all instances and INLINEFORM7 becomes 1. For INLINEFORM8 , LiLi uses INLINEFORM9 only once (as only one MLQ satisfies INLINEFORM10 ) compared to INLINEFORM11 . In summary, LiLi’s RL-model can effectively formulate query-specific inference strategies (based on specified parameter values). Evaluation-II: Predictive Performance. Table 6 shows the comparative performance of LiLi with baselines. To judge the overall improvements, we performed paired t-test considering +ve F1 scores on each relation as paired data. Considering both KBs and all relation types, LiLi outperforms Sep with INLINEFORM12 . If we set INLINEFORM13 (training with very few clues), LiLi outperforms Sep with INLINEFORM14 on Freebase considering MCC. Thus, the lifelong learning mechanism is effective in transferring helpful knowledge. Single model performs better than Sep for unknown relations due to the sharing of knowledge (weights) across tasks. However, for known relations, performance drops because, as a new relation arrives to the system, old weights get corrupted and catastrophic forgetting occurs. For unknown relations, as the relations are evaluated just after training, there is no chance for catastrophic forgetting. The performance improvement ( INLINEFORM15 ) of LiLi over F-th on Freebase signifies that the relation-specific threshold INLINEFORM16 works better than fixed threshold 0.5 because, if all prediction values for test instances lie above (or below) 0.5, F-th predicts all instances as +ve (-ve) which degrades its performance. Due to the utilization of contextual similarity (highly correlated with class labels) of entity-pairs, LiLi’s guessing mechanism works better ( INLINEFORM17 ) than blind guessing (BG). The past task selection mechanism of LiLi also improves its performance over w/o PTS, as it acquires more clues during testing for poorly performed tasks (evaluated on validation set). For Freebase, due to a large number of past tasks [9 (25% of 38)], the performance difference is more significant ( INLINEFORM18 ). For WordNet, the number is relatively small [3 (25% of 14)] and hence, the difference is not significant.
Evaluation-III: User Interaction vs. Performance. Table 7 shows the results of LiLi by varying clue acquisition rate ( INLINEFORM0 ). We use Freebase for tuning INLINEFORM1 due to its higher number of unknown test relations compared to WordNet. LiLi’s performance improves significantly as it acquires more clues from the user. The results on INLINEFORM2 outperforms ( INLINEFORM3 ) that on INLINEFORM4 . Table 8 shows the results of LiLi on user responses to MLQ’s and CLQ’s. Answering MLQ’s and CLQ’s is very hard for simulated users (unlike crowd-sourcing) as often INLINEFORM5 lacks the required triple. Thus, we attempt to analyze how the performance is effected if the user does not respond at all. The results show a clear trend in overall performance improvement when the user responds. However, the improvement is not significant as the simulated user’s query satisfaction rate (1% MLQs and 10% CLQs) is very small. But, the analysis shows the effectiveness of LiLi’s guessing mechanism and continual learning ability that help in achieving avg. +ve F1 of 0.57 and 0.62 on FB and WN respectively with minimal participation of the user.
Evaluation-II: Predictive Performance. Table 6 shows the comparative performance of LiLi with baselines. To judge the overall improvements, we performed paired t-test considering +ve F1 scores on each relation as paired data.
|
How do they judge the overall improvements of the predictive performance?
|
They performed paired t-test considering +ve F1 scores on each relation as paired data.
|
null | false
| 176
|
The pre-trained language model, BERT BIBREF0 has led to a big breakthrough in various kinds of natural language understanding tasks. Ideally, people can start from a pre-trained BERT checkpoint and fine-tune it on a specific downstream task. However, the original BERT models are memory-exhaustive and latency-prohibitive to be served in embedded devices or CPU-based online environments. As the memory and latency constraints vary in different scenarios, the pre-trained BERT model should be adaptive to different requirements with accuracy retained to the largest extent. Existing BERT-oriented model compression solutions largely depend on knowledge distillation BIBREF1, which is inefficient and resource-consuming because a large training corpus is required to learn the behaviors of a teacher. For example, DistilBERT BIBREF2 is re-trained on the same corpus as pre-training a vanilla BERT from scratch; and TinyBERT BIBREF3 utilizes expensive data augmentation to fit the distillation target. The costs of these model compression methods are as large as pre-training and unaffordable for low-resource settings. Therefore, it is straight-forward to ask, can we design a lightweight method to generate adaptive models with comparable accuracy using significantly less time and resource consumption? In this paper, we propose LadaBERT (Lightweight adaptation of BERT through hybrid model compression) to tackle the raised questions. Specifically, LadaBERT is based on an iterative hybrid model compression framework consisting of weighting pruning, matrix factorization and knowledge distillation. Initially, the architecture and weights of student model are inherited from the BERT teacher. In each iteration, the student model is first compressed by a small ratio based on weight pruning and matrix factorization, and is then fine-tuned under the guidance of teacher model through knowledge distillation. Because weight pruning and matrix factorization help to generate better initial and intermediate status
in the knowledge distillation iterations, the accuracy and efficiency of model compression can be greatly improved.
We conduct extensive experiments on five public datasets of natural language understanding. As an example, the performance comparison of LadaBERT and state-of-the-art models on MNLI-m dataset is illustrated in Figure FIGREF1. We can see that LadaBERT outperforms other BERT-oriented model compression baselines at various model compression ratios. Especially, LadaBERT-1 outperforms BERT-PKD significantly under $2.5\times $ compression ratio, and LadaBERT-3 outperforms TinyBERT under $7.5\times $ compression ratio while the training speed is accelerated by an order of magnitude.
The rest of this paper is organized as follows. First, we summarizes the related works of model compression and their applications to BERT in Section SECREF2. Then, the methodology of LadaBERT is introduced in Section SECREF3, and experimental results are presented in Section SECREF4. At last, we conclude this work and discuss future works in Section SECREF5.
In this paper, we propose LadaBERT (Lightweight adaptation of BERT through hybrid model compression) to tackle the raised questions.
|
What method do the authors propose to tackle the raised questions?
|
The authors propose LadaBERT (Lightweight adaptation of BERT through hybrid model compression) to tackle the raised questions.
|
null | false
| 78
|
Social media platforms have made the spreading of fake news easier, faster as well as able to reach a wider audience. Social media offer another feature which is the anonymity for the authors, and this opens the door to many suspicious individuals or organizations to utilize these platforms. Recently, there has been an increased number of spreading fake news and rumors over the web and social media BIBREF0. Fake news in social media vary considering the intention to mislead. Some of these news are spread with the intention to be ironic or to deliver the news in an ironic way (satirical news). Others, such as propaganda, hoaxes, and clickbaits, are spread to mislead the audience or to manipulate their opinions. In the case of Twitter, suspicious news annotations should be done on a tweet rather than an account level, since some accounts mix fake with real news. However, these annotations are extremely costly and time consuming – i.e., due to high volume of available tweets Consequently, a first step in this direction, e.g., as a pre-filtering step, can be viewed as the task of detecting fake news at the account level.
The main obstacle for detecting suspicious Twitter accounts is due to the behavior of mixing some real news with the misleading ones. Consequently, we investigate ways to detect suspicious accounts by considering their tweets in groups (chunks). Our hypothesis is that suspicious accounts have a unique pattern in posting tweet sequences. Since their intention is to mislead, the way they transition from one set of tweets to the next has a hidden signature, biased by their intentions. Therefore, reading these tweets in chunks has the potential to improve the detection of the fake news accounts.
In this work, we investigate the problem of discriminating between factual and non-factual accounts in Twitter. To this end, we collect a large dataset of tweets using a list of propaganda, hoax and clickbait accounts and compare different versions of sequential chunk-based approaches using a variety of feature sets against several baselines. Several approaches have been proposed for news verification, whether in social media (rumors detection) BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4, or in news claims BIBREF5, BIBREF6, BIBREF7, BIBREF8. The main orientation in the previous works is to verify the textual claims/tweets but not their sources. To the best of our knowledge, this is the first work aiming to detect factuality at the account level, and especially from a textual perspective. Our contributions are:
[leftmargin=4mm]
We propose an approach to detect non-factual Twitter accounts by treating post streams as a sequence of tweets' chunks. We test several semantic and dictionary-based features together with a neural sequential approach, and apply an ablation test to investigate their contribution.
We benchmark our approach against other approaches that discard the chronological order of the tweets or read the tweets individually. The results show that our approach produces superior results at detecting non-factual accounts.
In this work, we investigate the problem of discriminating between factual and non-factual accounts in Twitter.
|
What problem does the paper investigate?
|
The problem of discriminating between factual and non-factual accounts in Twitter.
|
null | false
| 133
|
A variety of approaches have been proposed for document quality assessment across different domains: Wikipedia article quality assessment, academic paper rating, content quality assessment in community question answering (cQA), and essay scoring. Among these approaches, some use hand-crafted features while others use neural networks to learn features from documents. For each domain, we first briefly describe feature-based approaches and then review neural network-based approaches. Wikipedia article quality assessment: Quality assessment of Wikipedia articles is a task that assigns a quality class label to a given Wikipedia article, mirroring the quality assessment process that the Wikipedia community carries out manually. Many approaches have been proposed that use features from the article itself, meta-data features (e.g., the editors, and Wikipedia article revision history), or a combination of the two. Article-internal features capture information such as whether an article is properly organized, with supporting evidence, and with appropriate terminology. For example, BIBREF3 use writing styles represented by binarized character trigram features to identify featured articles. BIBREF4 and BIBREF0 explore the number of headings, images, and references in the article. BIBREF5 use nine readability scores, such as the percentage of difficult words in the document, to measure the quality of the article. Meta-data features, which are indirect indicators of article quality, are usually extracted from revision history, and the interaction between editors and articles. For example, one heuristic that has been proposed is that higher-quality articles have more edits BIBREF6 , BIBREF7 . BIBREF8 use the percentage of registered editors and the total number of editors of an article. Article–editor dependencies have also been explored. For example, BIBREF9 use the authority of editors to measure the quality of Wikipedia articles, where the authority of editors is determined by the articles they edit. Deep learning approaches to predicting Wikipedia article quality have also been proposed. For example, BIBREF10 use a version of doc2vec BIBREF11 to represent articles, and feed the document embeddings into a four hidden layer neural network. BIBREF12 first obtain sentence representations by averaging words within a sentence, and then apply a biLSTM BIBREF13 to learn a document-level representation, which is combined with hand-crafted features as side information. BIBREF14 exploit two stacked biLSTMs to learn document representations.
Academic paper rating: Academic paper rating is a relatively new task in NLP/AI, with the basic formulation being to automatically predict whether to accept or reject a paper. BIBREF2 explore hand-crafted features, such as the length of the title, whether specific words (such as outperform, state-of-the-art, and novel) appear in the abstract, and an embedded representation of the abstract as input to different downstream learners, such as logistic regression, decision tree, and random forest. BIBREF15 exploit a modularized hierarchical convolutional neural network (CNN), where each paper section is treated as a module. For each paper section, they train an attention-based CNN, and an attentive pooling layer is applied to the concatenated representation of each section, which is then fed into a softmax layer.
Content quality assessment in cQA: Automatic quality assessment in cQA is the task of determining whether an answer is of high quality, selected as the best answer, or ranked higher than other answers. To measure answer content quality in cQA, researchers have exploited various features from different sources, such as the answer content itself, the answerer's profile, interactions among users, and usage of the content. The most common feature used is the answer length BIBREF16 , BIBREF17 , with other features including: syntactic and semantic features, such as readability scores. BIBREF18 ; similarity between the question and the answer at lexical, syntactic, and semantic levels BIBREF18 , BIBREF19 , BIBREF20 ; or user data (e.g., a user's status points or the number of answers written by the user). There have also been approaches using neural networks. For example, BIBREF21 combine CNN-learned representations with hand-crafted features to predict answer quality. BIBREF22 use a 2-dimensional CNN to learn the semantic relevance of an answer to the question, and apply an LSTM to the answer sequence to model thread context. BIBREF23 and BIBREF24 model the problem similarly to machine translation quality estimation, treating answers as competing translation hypotheses and the question as the reference translation, and apply neural machine translation to the problem. Essay scoring: Automated essay scoring is the task of assigning a score to an essay, usually in the context of assessing the language ability of a language learner. The quality of an essay is affected by the following four primary dimensions: topic relevance, organization and coherence, word usage and sentence complexity, and grammar and mechanics. To measure whether an essay is relevant to its “prompt” (the description of the essay topic), lexical and semantic overlap is commonly used BIBREF25 , BIBREF26 . BIBREF27 explore word features, such as the number of verb formation errors, average word frequency, and average word length, to measure word usage and lexical complexity. BIBREF28 use sentence structure features to measure sentence variety. The effects of grammatical and mechanic errors on the quality of an essay are measured via word and part-of-speech $n$ -gram features and “mechanics” features BIBREF29 (e.g., spelling, capitalization, and punctuation), respectively. BIBREF30 , BIBREF31 , and BIBREF32 use an LSTM to obtain an essay representation, which is used as the basis for classification. Similarly, BIBREF33 utilize a CNN to obtain sentence representation and an LSTM to obtain essay representation, with an attention layer at both the sentence and essay levels.
Quality assessment of Wikipedia articles is a task that assigns a quality class label to a given Wikipedia article, mirroring the quality assessment process that the Wikipedia community carries out manually.
|
What is the Wikipedia article quality assessment in this paper？
|
Quality assessment of Wikipedia articles is a task that assigns a quality class label to a given Wikipedia article, mirroring the quality assessment process that the Wikipedia community carries out manually.
|
null | false
| null |
Categorize each of the following with their associated holiday, either "Christmas", "Easter" or "Halloween". Bunny, egg, cobwebs, candy bucket, candy basket, lights, Santa
|
Bunny: Easter, egg: Easter, cobwebs: Halloween, candy bucket: Halloween, Candy basket: Easter, lights: Christmas, Santa: Christmas
|
|
null | false
| null |
Where is the best place to vacation in the Bahamas?
|
Harbour Island is a small island in the Bahamas, accessible via the airport at North Eleuthera. Flights leave Miami, Fort Lauderdale or Nassau, several times daily. From Florida, Harbour Island is a 45min paddle jumper flight to North Eleuthera and then a 10 minute speed boat.
Harbour Island has 8 hotels, Coral Sands, Pink Sands, Ramora Bay, Valentines, The Rockhouse, and The Dunmore. The Landing. All are boutique hotels. Coral Sands is about 90 rooms and the largest. There are no cars allowed in Harbour Island. Everyone travels by golf cart. Restaurants are great on the Island, and no visit is complete without a visit to Daddy D's nightclub.
|
|
null | false
| null |
The primary housing areas for first-year students are McCormick Road Dormitories, often called "Old Dorms", and Alderman Road Dormitories, often called "New Dorms". The 1970s-era Alderman Road Dorms are being fully replaced with brand new dormitory buildings in the same area. The replacements feature hall-style living arrangements with common areas and many modern amenities. Instead of being torn down and replaced like the original New Dorms, the Old Dorms will see a $105 million renovation project between 2017 and 2022. They were constructed in 1950, and are also hall-style constructions but with fewer amenities. The Old Dorms are closer to the students' classes.
|
According to this paragraph, what are the "Old Dorms" at the University of Virginia?
|
The "Old Dorms" at the University of Virginia often refer to the primary housing area for first-year students on McCormick Road.
|
null | false
| null |
What is the largest city in North America?
|
The largest city in North America by population is Mexico City.
|
|
null | false
| null |
How many movies did director Quentin Tarantino make?
|
As of March 2023, Quentin Tarantino has made 10 movies. They are, in chronological order:
1. Reservoir Dogs
2. Pulp Fiction
3. Jackie Brown
4. Kill Bill: Volume 1
5. Kill Bill: Volume 2
6. Death Proof
7. Inglourious Basterds
8. Django Unchained
9. The Hateful Eight
10. Once Upon a Time in Hollywood
|
|
null | false
| null |
Name some of the famous battles from the American Revolutionary War
|
Famous battles from the American Revolutionary War include Battle of Bunker Hill, Battles of Lexington and Concord, and Yorktown
|
|
null | false
| null |
Which is a species of fish? Opah or Oprah
|
Opah
|
|
null | false
| null |
Suggest some ranged weapons for role playing games.
|
Ranged weapons include the bow, shotgun, rifle and cannon.
|
|
null | false
| null |
Why is University of Wisconsin Madison the best college?
|
UW-Madison has the best work/life balance. It's consistency ranked as one of the top academic universities in the world as well as party schools in the U.S. No wonder there are so many Fortune 500 CEOs from UW-Madison!
|
|
null | false
| null |
What is the current best TV show?
|
Ted Lasso because of it's heart wrenching plot, amazingly developed characters & an overall positive message. It also provides example of relatability in the characters to your own personal life.
|
|
null | false
| null |
Who are U2
|
U2 are an Irish rock band from Dublin, formed in 1976. The group consists of Bono (lead vocals and rhythm guitar), the Edge (lead guitar, keyboards, and backing vocals), Adam Clayton (bass guitar), and Larry Mullen Jr. (drums and percussion). Initially rooted in post-punk, U2's musical style has evolved throughout their career, yet has maintained an anthemic quality built on Bono's expressive vocals and the Edge's chiming, effects-based guitar sounds. Bono's lyrics, often embellished with spiritual imagery, focus on personal and sociopolitical themes. Popular for their live performances, the group have staged several ambitious and elaborate tours over their career.
|
|
null | false
| 64
|
Deep learning systems have shown a lot of promise for extractive Question Answering (QA), with performance comparable to humans when large scale data is available. However, practitioners looking to build QA systems for specific applications may not have the resources to collect tens of thousands of questions on corpora of their choice. At the same time, state-of-the-art machine reading systems do not lend well to low-resource QA settings where the number of labeled question-answer pairs are limited (c.f. Table 2 ). Semi-supervised QA methods like BIBREF0 aim to improve this performance by leveraging unlabeled data which is easier to collect.
In this work, we present a semi-supervised QA system which requires the end user to specify a set of base documents and only a small set of question-answer pairs over a subset of these documents. Our proposed system consists of three stages. First, we construct cloze-style questions (predicting missing spans of text) from the unlabeled corpus; next, we use the generated clozes to pre-train a powerful neural network model for extractive QA BIBREF1 , BIBREF2 ; and finally, we fine-tune the model on the small set of provided QA pairs.
Our cloze construction process builds on a typical writing phenomenon and document structure: an introduction precedes and summarizes the main body of the article. Many large corpora follow such a structure, including Wikipedia, academic papers, and news articles. We hypothesize that we can benefit from the un-annotated corpora to better answer various questions – at least ones that are lexically similar to the content in base documents and directly require factual information.
We apply the proposed system on three datasets from different domains – SQuAD BIBREF3 , TriviaQA-Web BIBREF4 and the BioASQ challenge BIBREF5 . We observe significant improvements in a low-resource setting across all three datasets. For SQuAD and TriviaQA, we attain an F1 score of more than 50% by merely using 1% of the training data. Our system outperforms the approaches for semi-supervised QA presented in BIBREF0 , and a baseline which uses the same unlabeled data but with a language modeling objective for pretraining. In the BioASQ challenge, we outperform the best performing system from previous year's challenge, improving over a baseline which does transfer learning from the SQuAD dataset. Our analysis reveals that questions which ask for factual information and match to specific parts of the context documents benefit the most from pretraining on automatically constructed clozes.
Deep learning systems have shown a lot of promise for extractive Question Answering (QA), with performance comparable to humans when large scale data is available. However, practitioners looking to build QA systems for specific applications may not have the resources to collect tens of thousands of questions on corpora of their choice. At the same time, state-of-the-art machine reading systems do not lend well to low-resource QA settings where the number of labeled questionanswer pairs are limited (c.f. Table 2). Semisupervised QA methods like (Yang et al., 2017) aim to improve this performance by leveraging unlabeled data which is easier to collect. In this work, we present a semi-supervised QA system which requires the end user to specify a set of base documents and only a small set of question-answer pairs over a subset of these documents. Our proposed system consists of three stages. First, we construct cloze-style questions (predicting missing spans of text) from the unlabeled corpus; next, we use the generated clozes to pre-train a powerful neural network model for extractive QA (Clark and Gardner, 2017; Dhingra et al., 2017); and finally, we fine-tune the model on the small set of provided QA pairs.
|
What is the purpose of the paper?
|
They come up with an approach to generate resource of a question answer pairs for the benefit of QA tasks .
|
null | false
| 310
|
Quality estimation (QE) refers to the task of measuring the quality of machine translation (MT) system outputs without reference to the gold translations BIBREF0 , BIBREF1 . QE research has grown increasingly popular due to the improved quality of MT systems, and potential for reductions in post-editing time and the corresponding savings in labor costs BIBREF2 , BIBREF3 . QE can be performed on multiple granularities, including at word level, sentence level, or document level. In this paper, we focus on quality estimation at word level, which is framed as the task of performing binary classification of translated tokens, assigning “OK” or “BAD” labels.
Early work on this problem mainly focused on hand-crafted features with simple regression/classification models BIBREF4 , BIBREF5 . Recent papers have demonstrated that utilizing recurrent neural networks (RNN) can result in large gains in QE performance BIBREF6 . However, these approaches encode the context of the target word by merely concatenating its left and right context words, giving them limited ability to control the interaction between the local context and the target word.
In this paper, we propose a neural architecture, Context Encoding Quality Estimation (CEQE), for better encoding of context in word-level QE. Specifically, we leverage the power of both (1) convolution modules that automatically learn local patterns of surrounding words, and (2) hand-crafted features that allow the model to make more robust predictions in the face of a paucity of labeled data. Moreover, we further utilize stacked recurrent neural networks to capture the long-term dependencies and global context information from the whole sentence.
We tested our model on the official benchmark of the WMT18 word-level QE task. On this task, it achieved highly competitive results, with the best performance over other competitors on English-Czech, English-Latvian (NMT) and English-Latvian (SMT) word-level QE task, and ranking second place on English-German (NMT) and German-English word-level QE task.
QE research has grown increasingly popular due to the improved quality of MT systems, and potential for reductions in post-editing time and the corresponding savings in labor costs (Specia, 2011; Turchi et al., 2014).
|
Why is QE research so popular?
|
Due to the improved quality of MT systems, and potential for reductions in post-editing time and the corresponding savings in labor costs.
|
null | false
| 171
|
This paper presents the first user study on perceived appropriateness of system responses after verbal abuse. We put strategies used by state-of-the-art systems to the test in a large-scale, crowd-sourced evaluation. The full annotated corpus contains 2441 system replies, categorised into 14 response types, which were evaluated by 472 raters - resulting in 7.7 ratings per reply.
Our results show that: (1) The user's age has an significant effect on the ratings. For example, older users find jokes as a response to harassment highly inappropriate. (2) Perceived appropriateness also depends on the type of previous abuse. For example, avoidance is most appropriate after sexual demands. (3) All system were rated significantly higher than our negative adult-only baselines - except two data-driven systems, one of which is a Seq2Seq model trained on “clean" data where all utterances containing abusive words were removed BIBREF1. This leads us to believe that data-driven response generation need more effective control mechanisms BIBREF30.
This paper presents the first user study on perceived appropriateness of system responses after verbal abuse.
|
Is this paper the first user study on the perceived appropriateness of system responses after verbal abuse?
|
Yes, it is.
|
null | false
| 486
|
Given the language instructions and the affordance-aware semantic representation (i.e., semantic map) acquired during exploration, the agent executes the subgoals sequentially. It uses a planning module (which consumes the semantic map) for navigation subgoals and an object interaction transformer for other subgoals.
|
What is the difference between planned and prediction actions?
|
In the execution phase, we refer to the actions obtained from the planner as “planned actions” for the navigation subgoals and actions obtained from the Object Interaction Transformer as “predicted actions” for the non-navigation subgoals.
|
null | false
| null |
The campuses' dominant architectural themes are Federal, Classical and Antebellum style. Though there have been many additions, changes, and augmentations, the University of Georgia's campus maintains its historic character. In 2000, the entire campus was designated as an arboretum, the University of Georgia Campus Arboretum. It is estimated to be home to about 9,000 trees with over 154 identified species including native trees such magnolias, red oaks, white oaks, and beeches, as well as non-native trees such as the North Africa Atlas cedar, the Chinese parasol and royal paulownia, and the Japanese zelkova and black pine
|
Based on the following paragraph about the campuses of the University of Georgia, which types of trees can be found on the campus?
|
On the campuses of the University of Georgia one can find native tree species such as magnolias, red oaks, white oaks, and beeches. There are also non-native tree species to be found such as the North Africa Atlas cedar, the Chinese parasol and royal paulownia, and the Japanese zelkova and black pine.
|
null | false
| null |
Tell me whether these are a breed of dog or cat: Australian Shepherd, Calico, Maine Coon, Golden Retriever, Corgi
|
dog, cat, cat, dog, dog
|
|
null | false
| null |
Cricket is a bat-and-ball game played between two teams of eleven players on a field at the centre of which is a 22-yard (20-metre) pitch with a wicket at each end, each comprising two bails balanced on three stumps. The batting side scores runs by striking the ball bowled at one of the wickets with the bat and then running between the wickets, while the bowling and fielding side tries to prevent this (by preventing the ball from leaving the field, and getting the ball to either wicket) and dismiss each batter (so they are "out"). Means of dismissal include being bowled, when the ball hits the stumps and dislodges the bails, and by the fielding side either catching the ball after it is hit by the bat, but before it hits the ground, or hitting a wicket with the ball before a batter can cross the crease in front of the wicket. When ten batters have been dismissed, the innings ends and the teams swap roles. The game is adjudicated by two umpires, aided by a third umpire and match referee in international matches. They communicate with two off-field scorers who record the match's statistical information.
|
What is game of Cricket?
|
Cricket is a game played using a bat and a ball, with 11 players in each team. The surface in the centre of cricket filed is called Pitch which is 22 yards long. The batting starts with 2 player on the pitch, a batter and runner. Bowling side chooses a baller and rest of the 10 players are spread across the filed to catch the ball. Once ten batters from batting side are dismissed, the other sides gets opportunity to bat. The runs are scored by running between each end of the wicket and hitting the ball outside the boundary lines of the filed.
|
null | false
| null |
What are a few dishes you can cook with potatoes?
|
There are a lot of great dishes you can cook with potatoes. A few common dishes include mashed potatoes, baked potatoes, and french fries.
|
|
1911.09483
| false
| null |
During inference, we adopt beam search with a beam size of 5 for De-En, En-Fr and En-Vi translation tasks. The length penalty is set to 0.8 for En-Fr according to the validation results, 1 for the two small datasets following the default setting of BIBREF14. We do not tune beam width and length penalty but use the setting reported in BIBREF0. The BLEU metric is adopted to evaluate the model performance during evaluation.
During inference, we adopt beam search with a beam size of 5 for De-En, En-Fr and En-Vi translation tasks.
|
What are three main machine translation tasks?
|
The answers are shown as follows:
* De-En, En-Fr and En-Vi translation tasks
|
null | false
| null |
Tell me a number of ways that one could entertain a small group of 6 year old children.
|
There are a quite a few ways in which to entertain a group of young children depending on their interests. Most children enjoy games, so you could find some age-appropriate board games and help them to understand the rules and play through the game. Alternatively, if you have space, you could have them play hide-and-go-seek indoors or outside, or alternatively games like 'suck in the mud' are fun for outdoor activity. For something quieter, you could find some rolls of butchers paper and drawing tools or paints and provide a theme for children to paint a mural together.
|
|
null | false
| null |
What is the best way to make an AreoPress coffee
|
Ingredients: 30g of Coffee Medium Grind, 150-250ml of 86º Water
Steps: Rinse AreoPress filter and set aside. Invert AreoPress, place on scale, and fill with ground coffee. Tare the scale, start a timer and begin filling AreoPress with water making sure to saturate the grounds. Stop pouring when when the scale reads 100g, adjust your pour so that this process takes around 10s. Stir vigorously for 20s then place your filter and lid on the AreoPress. When the timer reads 40s press the air out of the AreoPress. When the timer reads 45s flip the areopress, placing it on top of your cup. Press the body of the AreoPress to push out the coffee. Now add between 50-150ml of water according to your taste and stir to combine. Finally, enjoy your coffee.
|
|
null | false
| 306
|
BERT (Bidirectional Encoder Representations from Transformers) BIBREF14 is a new language representation model, which uses bidirectional transformers to pre-train a large unlabeled corpus, and fine-tunes the pre-trained model on other tasks. BERT has been widely used and shows great improvement on various natural language processing tasks, e.g., word segmentation, named entity recognition, sentiment analysis, and question answering. We use BERT to extract contextual feature for each character instead of BiLSTM in the original work BIBREF13. To further improve the performance, we optimize the pre-training process of BERT by introducing a semantic-enhanced task.
Original google BERT is pre-trained using two unsupervised tasks, masked language model (MLM) and next sentence prediction (NSP). MLM task enables the model to capture the discriminative contextual feature. NSP task makes it possible to understand the relationship between sentence pairs, which is not directly captured by language modeling. We further design a semantic-enhanced task to enhance the performance of BERT. It incorporate previous sentence prediction and document level prediction. We pre-train BERT by combining MLM, NSP and the semantic-enhanced task together.
BERT (Bidirectional Encoder Representations from Transformers) [1] is a new language representation model, which uses bidirectional transformers to pretrain a large unlabeled corpus, and fine-tunes the pre-trained model on other tasks.
|
What is the BERT?
|
It is a new language representation model, which uses bidirectional transformers to pretrain a large unlabeled corpus, and fine-tunes the pre-trained model on other tasks.
|
null | false
| null |
The game of association football is played in accordance with the Laws of the Game, a set of rules that has been in effect since 1863 and maintained by the IFAB since 1886. The game is played with a football that is 68–70 cm (27–28 in) in circumference. The two teams compete to get the ball into the other team's goal (between the posts and under the bar), thereby scoring a goal. When the ball is in play, the players mainly use their feet, but may use any other part of their body, except for their hands or arms, to control, strike, or pass the ball. Only the goalkeepers may use their hands and arms, and only then within the penalty area. The team that has scored more goals at the end of the game is the winner. Depending on the format of the competition, an equal number of goals scored may result in a draw being declared, or the game goes into extra time or a penalty shootout.
|
Summarise the basic rules of soccer, without using examples from the text
|
Soccer is played between two teams of players who pass a spherical inflated ball between each other, predominantly using the feet and head, with the aim of getting the ball into the opponents goal (a rectangular opening at each end of the playing area) . The only player who can use their hands and arms is the goalkeeper, within a small area near their team's goal called the penalty area. At the end of the game, the team who has got the ball into their opponents goal most times is declared the winner.
|
null | false
| null |
The Awards ceremony for 63rd Grammy Awards was held at which place?
|
Los Angeles
|
|
null | false
| 17
|
Following developing news stories is imperative to making real-time decisions on important political and public safety matters. Given the abundance of media providers and languages, this endeavor is an extremely difficult task. As such, there is a strong demand for automatic clustering of news streams, so that they can be organized into stories or themes for further processing. Performing this task in an online and efficient manner is a challenging problem, not only for newswire, but also for scientific articles, online reviews, forum posts, blogs, and microblogs.
A key challenge in handling document streams is that the story clusters must be generated on the fly in an online fashion: this requires handling documents one-by-one as they appear in the document stream. In this paper, we provide a treatment to the problem of online document clustering, i.e. the task of clustering a stream of documents into themes. For example, for news articles, we would want to cluster them into related news stories.
To this end, we introduce a system which aggregates news articles into fine-grained story clusters across different languages in a completely online and scalable fashion from a continuous stream. Our clustering approach is part of a larger media monitoring project to solve the problem of monitoring massive text and TV/Radio streams (speech-to-text). In particular, media monitors write intelligence reports about the most relevant events, and being able to search, visualize and explore news clusters assists in gathering more insight about a particular story. Since relevant events may be spawned from any part of the world (and from many multilingual sources), it becomes imperative to cluster news across different languages.
In terms of granularity, the type of story clusters we are interested in are the group of articles which, for example : (i) Narrate recent air-strikes in Eastern Ghouta (Syria); (ii) Describe the recent launch of Space X's Falcon Heavy rocket.
A key challenge in handling document streams is that the story clusters must be generated on the fly in an online fashion: this requires handling documents one-by-one as they appear in the document stream.
|
What are the key challenges in handling document flow?
|
This requires handling documents one-by-one as they appear in the document stream.
|
null | false
| null |
An inter frame is a frame in a video compression stream which is expressed in terms of one or more neighboring frames. The "inter" part of the term refers to the use of Inter frame prediction. This kind of prediction tries to take advantage from temporal redundancy between neighboring frames enabling higher compression rates.
|
Explain inter frame
|
An inter frame is a frame in a video compression stream which is expressed in terms of one or more neighboring frames. The "inter" part of the term refers to the use of Inter frame prediction. This kind of prediction tries to take advantage from temporal redundancy between neighboring frames enabling higher compression rates.
|
null | false
| null |
Identify which car manufacturer is British or American: Lotus, Lincoln
|
Lotus is British, Lincoln is American
|
|
null | false
| null |
The aid of the Association was also given to the formation of a class of women for the study of advanced chemistry at the laboratory of the Girls' High School. This class was found to meet so great a want that a chemical laboratory has been fitted up in connection with the Massachusetts Institute of Technology, and furnished by money collected by the association with apparatus and instruments, such as microscopes, spectroscopes, &c. From time to time pecuniary aid has also been given to young women seeking an advanced education. But the chief work of the Association —the first undertaken and still carried on under its charge- is a plan of examinations known as the Harvard Examinations for Women. These originated in a desire to raise the standard of education in the schools for girls, and are held under the personal supervision of members of the Association. Examination papers, prepared by Professors at Harvard University, are provided by the Association, and those women who successfully pass this examination receive a certificate from the university stating the fact. Branch societies have been established in New York City, Philadelphia, and Cincinnati for the same purpose.
|
Given this paragraph about the Women's Education Association, which institutions of higher learning had affiliations with the Association?
|
The Massachusetts Institute of Technology and Harvard University
|
null | false
| null |
Identify which instrument is string or percussion: Mridangam, Gambus
|
Gambus is string, Mridangam is percussion.
|
|
null | false
| null |
"I'm the Urban Spaceman" was the Bonzo Dog Doo-Dah Band's most successful single, released in 1968. It reached #5 in the UK charts. The song was written by Neil Innes—who won an Ivor Novello Award in 1968 for the song—and produced by Paul McCartney under the pseudonym "Apollo C. Vermouth". The B-side was written by Vivian Stanshall. A well-known staging of the song involves Innes performing solo while a female tap dancer performs an enthusiastic but apparently under-rehearsed routine around him. This skit originally appeared in a 1975 edition of Rutland Weekend Television, with Lyn Ashley as the dancer, and was more famously revived in the 1982 film Monty Python Live at the Hollywood Bowl with Carol Cleveland taking over the role.
Leeds indie rock band Cud performed an extremely fast version (1:07 long) for a 1989 Peel Session. The recording appears on their albums Elvis Belt and BB Cudn't C.
|
From the passage provided, list the awards won. Include when they were won and by whom.
|
Neil Innes won the Ivor Novello Award in 1968 for writing the song "I'm the Urban Spaceman".
|
null | false
| null |
Classify each of the following as straight or curved: banana, baseball bat, hook, rainbow, convex lens, ruler
|
Banana: Curved
Baseball Bat: Straight
Hook: Curved
Rainbow: Curved
Convex Lens: Curved
Ruler: Straight
|
|
null | false
| null |
Marina Vishmidt (born 1976) is an American writer, editor and critic. She lectures at the Centre for Cultural Studies at Goldsmiths, University of London in the MA program Culture Industry, and teaches Art Theory in the MA Art Praxis at the Dutch Art Institute in Arnhem. Her research mainly concerns the relationship between art, value and labour. She further explores this through works on debt, social reproduction and artistic entrepreneurialism. In 2013, she completed her PhD entitled 'Speculation as a Mode of Production in Art and Capital' at the Queen Mary University of London.
|
Who is Marina Vishmidt
|
Marina Vishmidt (born 1976) is an American writer, editor and critic. She lectures at the Centre for Cultural Studies at Goldsmiths, University of London in the MA program Culture Industry, and teaches Art Theory in the MA Art Praxis at the Dutch Art Institute in Arnhem. Her research mainly concerns the relationship between art, value and labour.
She further explores this through works on debt, social reproduction and artistic entrepreneurialism. In 2013, she completed her PhD entitled 'Speculation as a Mode of Production in Art and Capital' at the Queen Mary University of London.
|
null | false
| null |
Think up a new seasoning blend that will be as popular as "everything but the bagel" seasoning.
|
A new popular seasoning could be a sour cream and onion seasoning reminiscent of the popular chip flavor. You can sprinkle it on a myriad of items for a delicious savory boost.
|
|
null | false
| 109
|
Sarcasm is defined as “a sharp, bitter, or cutting expression or remark; a bitter gibe or taunt”. As the fields of affective computing and sentiment analysis have gained increasing popularity BIBREF0 , it is a major concern to detect sarcastic, ironic, and metaphoric expressions. Sarcasm, especially, is key for sentiment analysis as it can completely flip the polarity of opinions. Understanding the ground truth, or the facts about a given event, allows for the detection of contradiction between the objective polarity of the event (usually negative) and its sarcastic characteristic by the author (usually positive), as in “I love the pain of breakup”. Obtaining such knowledge is, however, very difficult.
In our experiments, we exposed the classifier to such knowledge extracted indirectly from Twitter. Namely, we used Twitter data crawled in a time period, which likely contain both the sarcastic and non-sarcastic accounts of an event or similar events. We believe that unambiguous non-sarcastic sentences provided the classifier with the ground-truth polarity of those events, which the classifier could then contrast with the opposite estimations in sarcastic sentences. Twitter is a more suitable resource for this purpose than blog posts, because the polarity of short tweets is easier to detect (as all the information necessary to detect polarity is likely to be contained in the same sentence) and because the Twitter API makes it easy to collect a large corpus of tweets containing both sarcastic and non-sarcastic examples of the same event.
Sometimes, however, just knowing the ground truth or simple facts on the topic is not enough, as the text may refer to other events in order to express sarcasm. For example, the sentence “If Hillary wins, she will surely be pleased to recall Monica each time she enters the Oval Office :P :D”, which refers to the 2016 US presidential election campaign and to the events of early 1990's related to the US president Clinton, is sarcastic because Hillary, a candidate and Clinton's wife, would in fact not be pleased to recall her husband's alleged past affair with Monica Lewinsky. The system, however, would need a considerable amount of facts, commonsense knowledge, anaphora resolution, and logical reasoning to draw such a conclusion. In this paper, we will not deal with such complex cases.
Existing works on sarcasm detection have mainly focused on unigrams and the use of emoticons BIBREF1 , BIBREF2 , BIBREF3 , unsupervised pattern mining approach BIBREF4 , semi-supervised approach BIBREF5 and n-grams based approach BIBREF6 , BIBREF7 , BIBREF8 , BIBREF9 with sentiment features. Instead, we propose a framework that learns sarcasm features automatically from a sarcasm corpus using a convolutional neural network (CNN). We also investigate whether features extracted using the pre-trained sentiment, emotion and personality models can improve sarcasm detection performance. Our approach uses relatively lower dimensional feature vectors and outperforms the state of the art on different datasets. In summary, the main contributions of this paper are the following:
The rest of the paper is organized as follows: Section SECREF2 proposes a brief literature review on sarcasm detection; Section SECREF4 presents the proposed approach; experimental results and thorough discussion on the experiments are given in Section SECREF5 ; finally, Section SECREF6 concludes the paper.
Twitter is a more suitable resource for this purpose than blog posts, because the polarity of short tweets is easier to detect (as all the information necessary to detect polarity is likely to be contained in the same sentence) and because the Twitter API makes it easy to collect a large corpus of tweets containing both sarcastic and non-sarcastic examples of the same event.
|
Why Twitter data is a more suitable resource than blog data?
|
Because it is shorter and it is easier to get data containing both sarcastic and non-sarcastic examples of the same event by Twitter API.
|
null | false
| null |
Give me a list of the best movies I've never seen.
|
I would definitely recommend to watch Lord of the Rings. It is a great fantasy movie about friendship and how in the end, the good always prevails. The scale and grandeur, also the cast of the movie was superb. Any film where Tom Hanks is in the lead role should also be watched, as an actor he is simply amazing, and comes across so natural and convincing. If you like action movies, I can heartily recommend all the James Bond movies with Daniel Craig in the lead role. The cinematography, tension, music all play together so well. Speaking of cinematography, one would be amiss to not watch the latest movies of David Villeneuve, for example Inception or Dune. Timothee Chalamais, who is leading role in Dune, also did a great performance in 'The King' - if you liked the character in Dune for example. Other movies would be Interstellar, a great Sci-Fi movie, the Hunger Games trilogy and so many more.
|
|
null | false
| 444
|
Here, we introduce our architectures for the action-value estimators. First, we construct the estimators using the utility functions z 1 (w agt ), . . . , z N (w agt ) with agent-wise risk level w agt . Our main contribution in designing the estimators is twofold: (i) using IQN for the true action-value estimator Z jt (w env ) with environment-wise risk level w env and (ii) using a monotonic mixing network for the transformed action-value estimators Z tran with agent-wise risk level w agt .
Agent-wise utility function z i . In a partially observable setting, agents can learn better policies by using their action-observation history τ i instead of their current observation. We represent each agent-wise utility function as a DRQN that receives action-observation history as input and outputs z i for each action and w agt . The utility functions do not estimate the action-value, but aim to accurately extract the optimal action from the joint action-value.
True action-value estimator Z jt . The estimator Z jt aims to express distributions with additional representation power; note that the transformed action-value estimator has limited power due to the decentralization condition (equation 1). For the true action-value estimator, we employ a feed-forward network that takes the state s and set of utility functions z i for i ∈ N . Since we use z averaged over multiple w agt samples, the feed forward network is not conditioned on w agt .
To apply IQN for the true action-value estimator, we use an additional network φ that computes an embedding φ(w env ) for the sample point w env . We calculate the embedding of w env with cosine basis functions and utilize element-wise (Hadamard) product, as done in the IQN paper.
Table: Payoff matrix of the two-step game. N (µ, σ 2 ) denotes the normal distribution with a mean of µ and a variance of σ 2 . For the first step, agent-wise risk-seeking (cooperation) is crucial because the optimal action is (A, A) but taking action A is very risky because when the teammate chooses to be non-cooperative (i.e., by selecting action B or C), a catastrophic reward of -12 is provided. In the second step, handling environment-wise risks is highlighted since there are various combinations of mean and variance depending on joint actions, i.e., N (−1, 10) for the action (C, C) and N (−1, 0) for the action (B, B). Transformed action-value estimator Z tran . The architecture of the transformed action-value estimator is largely the same as the mixing network of. The transformed action-value estimator is expressed as follows:
where θ tran (s, w agt ) is a non-negative parameter obtained from state-dependent hypernetwork. A more detailed discussion is available in Appendix D.
Agent-wise utility function zi. In a partially observable setting, agents can learn better policies by using their action-observation history τi instead of their current observation. We represent each agent-wise utility function as a DRQN (Hausknecht & Stone, 2015) that receives action-observation history as input and outputs zi for each action and wagt. The utility functions do not estimate the action-value, but aim to accurately extract the optimal action from the joint action-value
|
What is the best way to think of the reward, versus the agent-wise utility?
|
They are different because the value of the agent-wise utility can be any value as long as the argmax action of the agent-wise utility and the joint action-value is the same. Conceptually, an agent-wise utility is similar to an actor in an actor-critic method. This is why we use the word “utility” rather than agent-wise “action-value”. In our revised draft, we further clarified this in Section 3.3.
|
1910.08772
| false
| null |
MonaLog utilizes two auxiliary sets. First, a knowledge base ${K}$ that stores the world knowledge needed for inference, e.g., semanticist $\le $ linguist and swim $\le $ move, which captures the facts that $[\![\mbox{\em semanticist}]\!]$ denotes a subset of $[\![\mbox{\em linguist}]\!]$, and that $[\![\mbox{\em swim}]\!]$ denotes a subset of $[\![\mbox{\em move}]\!]$, respectively. Such world knowledge can be created manually for the problem at hand, or derived easily from existing resources such as WordNet BIBREF22. Note that we do not blindly add all relations from WordNet to our knowledge base, since this would hinge heavily on word sense disambiguation (we need to know whether the “bank” is a financial institution or a river bank to extract its relations correctly). In the current implementation, we avoid this by adding x $\le $ y or x $\perp $ y relations only if both x and y are words in the premise-hypothesis pair. Additionally, some relations that involve quantifiers and prepositions need to be hard-coded, since WordNet does not include them: every $=$ all $=$ each $\le $ most $\le $ many $\le $ a few $=$ several $\le $ some $=$ a; the $\le $ some $=$ a; on $\perp $ off; up $\perp $ down; etc.
MonaLog utilizes two auxiliary sets. First, a knowledge base ${K}$ that stores the world knowledge needed for inference, e.g., semanticist $\le $ linguist and swim $\le $ move, which captures the facts that $[\![\mbox{\em semanticist}]\!]$ denotes a subset of $[\![\mbox{\em linguist}]\!]$, and that $[\![\mbox{\em swim}]\!]$ denotes a subset of $[\![\mbox{\em move}]\!]$, respectively. Such world knowledge can be created manually for the problem at hand, or derived easily from existing resources such as WordNet BIBREF22. Note that we do not blindly add all relations from WordNet to our knowledge base, since this would hinge heavily on word sense disambiguation (we need to know whether the “bank” is a financial institution or a river bank to extract its relations correctly). In the current implementation, we avoid this by adding x $\le $ y or x $\perp $ y relations only if both x and y are words in the premise-hypothesis pair. Additionally, some relations that involve quantifiers and prepositions need to be hard-coded, since WordNet does not include them: every $=$ all $=$ each $\le $ most $\le $ many $\le $ a few $=$ several $\le $ some $=$ a; the $\le $ some $=$ a; on $\perp $ off; up $\perp $ down; etc.
|
How do they select monotonicity facts?
|
They derive it from Wordnet
|
null | false
| null |
Classify each of the following as a professional golfer or professional football player: Jon Rahm, Patrick Mahomes, Tom Brady, Max Homa, Scottie Scheffler, Tiger Woods, Rory McIlroy, Josh Allen, Jacoby Brissett.
|
Jon Rahm is a professional golfer.
Patrick Mahomes is a professional football player.
Tom Brady is a professional football player.
Max Homa is a professional golfer.
Scottie Scheffler is a professional golfer.
Tiger Woods is a professional golfer.
Rory McIlroy is a professional golfer.
Josh Allen is a professional football player.
Jacoby Brissett is a professional football player.
|
|
null | false
| null |
In the field of video compression a video frame is compressed using different algorithms with different advantages and disadvantages, centered mainly around amount of data compression. These different algorithms for video frames are called picture types or frame types. The three major picture types used in the different video algorithms are I, P and B. They are different in the following characteristics:
I‑frames are the least compressible but don't require other video frames to decode.
P‑frames can use data from previous frames to decompress and are more compressible than I‑frames.
B‑frames can use both previous and forward frames for data reference to get the highest amount of data compression.
Three types of pictures (or frames) are used in video compression: I, P, and B frames.
An I‑frame (Intra-coded picture) is a complete image, like a JPG or BMP image file.
A P‑frame (Predicted picture) holds only the changes in the image from the previous frame. For example, in a scene where a car moves across a stationary background, only the car's movements need to be encoded. The encoder does not need to store the unchanging background pixels in the P‑frame, thus saving space. P‑frames are also known as delta‑frames.
A B‑frame (Bidirectional predicted picture) saves even more space by using differences between the current frame and both the preceding and following frames to specify its content.
P and B frames are also called Inter frames. The order in which the I, P and B frames are arranged is called the Group of pictures.
|
Briefly explain what are I frames, B frames, and P frames?
|
I frames, B frames, and P frames are video frame types used in video codecs. I frames can be decoded without reference to any other frames. They are self-contained descriptions of one frame of the video. P frames and B frames on the other hand reference other frames and must be decoded after the frames they reference have been decoded. P frames only reference frames from earlier in the video whereas B frames can reference frames from earlier or later. Decoding B frames therefore requires decoding the video frames in an order different from the playback order.
|
null | false
| null |
The 13th century also witnessed the Crown of Aragon, centred in Spain's north east, expand its reach across islands in the Mediterranean, to Sicily and Naples. Around this time the universities of Palencia (1212/1263) and Salamanca (1218/1254) were established. The Black Death of 1348 and 1349 devastated Spain.
|
Extract when were the universities of Palencia and Salamanca established
|
The University of Palencia was established between 1212 and 1263 and the University of Salamanca was established between 1218 and 1254.
|
1905.00563
| false
| null |
Since the setting is quite different from traditional adversarial attacks, search for link prediction adversaries brings up unique challenges. To find these minimal changes for a target link, we need to identify the fact that, when added into or removed from the graph, will have the biggest impact on the predicted score of the target fact. Unfortunately, computing this change in the score is expensive since it involves retraining the model to recompute the embeddings. We propose an efficient estimate of this score change by approximating the change in the embeddings using Taylor expansion. The other challenge in identifying adversarial modifications for link prediction, especially when considering addition of fake facts, is the combinatorial search space over possible facts, which is intractable to enumerate. We introduce an inverter of the original embedding model, to decode the embeddings to their corresponding graph components, making the search of facts tractable by performing efficient gradient-based continuous optimization. We evaluate our proposed methods through following experiments. First, on relatively small KGs, we show that our approximations are accurate compared to the true change in the score. Second, we show that our additive attacks can effectively reduce the performance of state of the art models BIBREF2 , BIBREF10 up to $27.3\%$ and $50.7\%$ in Hits@1 for two large KGs: WN18 and YAGO3-10. We also explore the utility of adversarial modifications in explaining the model predictions by presenting rule-like descriptions of the most influential neighbors. Finally, we use adversaries to detect errors in the KG, obtaining up to $55\%$ accuracy in detecting errors.
WN18 and YAGO3-10
Second, we show that our additive attacks can effectively reduce the performance of state of the art models BIBREF2 , BIBREF10 up to $27.3\%$ and $50.7\%$ in Hits@1 for two large KGs: WN18 and YAGO3-10.
|
What datasets are used to evaluate this approach?
|
The answers are shown as follows:
* WN18 and YAGO3-10
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.