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 |
What is another name for downhill skiing?
|
Downhill skiing is also known as alpine skiing
|
|
null | false
| null |
What is the state capital of Nevada?
|
The state capital of Nevada is Carson City. Carson City is named after Kit Carson and was founded in 1858.
|
|
null | false
| null |
What is a woodwind instrument?
|
Woodwind instruments are a family of musical instruments that produce sound by splitting the air blown on a sharp edge, such as a reed. These instruments can be made of any materials, not just wood.
|
|
null | false
| 85
|
There have been many advances in machine learning methods which help machines understand human behavior better than ever. One of the most important aspects of human behavior is emotion. If machines could detect human emotional expressions, it could be used to improve on verity of applications such as marketing BIBREF0 , human-computer interactions BIBREF1 , political science BIBREF2 etc.
Emotion in humans is complex and hard to distinguish. There have been many emotional models in psychology which tried to classify and point out basic human emotions such as Ekman's 6 basic emotions BIBREF3 , Plutchik's wheel of emotions BIBREF4 , or Parrott's three-level categorization of emotions BIBREF5 . These varieties show that emotions are hard to define, distinguish, and categorize even for human experts.
By adding the complexity of language and the fact that emotion expressions are very complex and context dependant BIBREF6 , BIBREF7 , BIBREF8 , we can see why detecting emotions in textual data is a challenging task. This difficulty can be seen when human annotators try to assign emotional labels to the text, but using various techniques the annotation task can be accomplished with desirable agreement among the annotators BIBREF9 .
Based on this assumption we propose a deep recurrent neural network architecture to detect discrete emotions in a tweet dataset. The code can be accessed at GitHub [https://github.com/armintabari/Emotion-Detection-RNN].
|
What does the author propose to detect discrete emotions in a tweet dataset?
|
A deep recurrent neural network architecture.
|
null | false
| null |
The potential uses of AI in government are wide and varied, with Deloitte considering that "Cognitive technologies could eventually revolutionize every facet of government operations". Mehr suggests that six types of government problems are appropriate for AI applications:
- Resource allocation - such as where administrative support is required to complete tasks more quickly.
- Large datasets - where these are too large for employees to work efficiently and multiple datasets could be combined to provide greater insights.
- Experts shortage - including where basic questions could be answered and niche issues can be learned.
- Predictable scenario - historical data makes the situation predictable.
- Procedural - repetitive tasks where inputs or outputs have a binary answer.
- Diverse data - where data takes a variety of forms (such as visual and linguistic) and needs to be summarised regularly.
|
What types of problems in government could AI be used to solve
|
The potential uses of AI in government are wide and varied but some appropriate areas to consider AI applications include: Resource allocation; Large datasets; Repetitive tasks; Answering basic questions; Summarising data from multiple forms; Predicting future scenarios based on historical data.
|
null | false
| null |
In optics, a dispersive prism is an optical prism that is used to disperse light, that is, to separate light into its spectral components (the colors of the rainbow). Different wavelengths (colors) of light will be deflected by the prism at different angles. This is a result of the prism material's index of refraction varying with wavelength (dispersion). Generally, longer wavelengths (red) undergo a smaller deviation than shorter wavelengths (blue). The dispersion of white light into colors by a prism led Sir Isaac Newton to conclude that white light consisted of a mixture of different colors.
|
What is a dispersive prism?
|
A dispersive prism is an optical prism that disperses the light's different wavelengths at different angles. When white light is shined through a dispersive prism it will separate into the different colors of the rainbow.
|
null | false
| null |
What is a piano?
|
A piano is a keyboard instrument that uses wooden hammers to strike a string which produces a sound. The hammers on a piano are coated with a soft material so they do not damage the string they hit when a keyboard key is pressed. You have to play this keyboard instrument with one or two hands in sync on a keyboard and then the keys hit in unified succession you get music or a string of sounds that is enjoyable to listen to.
|
|
null | false
| null |
Best place to visit in the world
|
1. Switzerland
2. Maldives
3, Hawaii
|
|
null | false
| null |
What is an exacta?
|
An exacta is a horse racing bet. It is when you bet on two horses to come in first place and second place consecutively in a given race. You are able to make the bet on almost all races. The payout will depend on the horses odds at the time the race has begun.
|
|
null | false
| null |
Where can I learn more about Kubernetes?
|
The Kubernetes docs themselves are very good - https://kubernetes.io/. Many folks also enjoy course material from Coursera or Udemy - such as this course from Mumshad https://www.udemy.com/course/certified-kubernetes-administrator-with-practice-tests/
|
|
null | false
| 0
|
Although Neural Machine Translation (NMT) has dominated recent research on translation tasks BIBREF0, BIBREF1, BIBREF2, NMT heavily relies on large-scale parallel data, resulting in poor performance on low-resource or zero-resource language pairs BIBREF3. Translation between these low-resource languages (e.g., Arabic$\rightarrow $Spanish) is usually accomplished with pivoting through a rich-resource language (such as English), i.e., Arabic (source) sentence is translated to English (pivot) first which is later translated to Spanish (target) BIBREF4, BIBREF5. However, the pivot-based method requires doubled decoding time and suffers from the propagation of translation errors.
One common alternative to avoid pivoting in NMT is transfer learning BIBREF6, BIBREF7, BIBREF8, BIBREF9 which leverages a high-resource pivot$\rightarrow $target model (parent) to initialize a low-resource source$\rightarrow $target model (child) that is further optimized with a small amount of available parallel data. Although this approach has achieved success in some low-resource language pairs, it still performs very poorly in extremely low-resource or zero-resource translation scenario. Specifically, BIBREF8 reports that without any child model training data, the performance of the parent model on the child test set is miserable.
In this work, we argue that the language space mismatch problem, also named domain shift problem BIBREF10, brings about the zero-shot translation failure in transfer learning. It is because transfer learning has no explicit training process to guarantee that the source and pivot languages share the same feature distributions, causing that the child model inherited from the parent model fails in such a situation. For instance, as illustrated in the left of Figure FIGREF1, the points of the sentence pair with the same semantics are not overlapping in source space, resulting in that the shared decoder will generate different translations denoted by different points in target space. Actually, transfer learning for NMT can be viewed as a multi-domain problem where each source language forms a new domain. Minimizing the discrepancy between the feature distributions of different source languages, i.e., different domains, will ensure the smooth transition between the parent and child models, as shown in the right of Figure FIGREF1. One way to achieve this goal is the fine-tuning technique, which forces the model to forget the specific knowledge from parent data and learn new features from child data. However, the domain shift problem still exists, and the demand of parallel child data for fine-tuning heavily hinders transfer learning for NMT towards the zero-resource setting.
In this paper, we explore the transfer learning in a common zero-shot scenario where there are a lot of source$\leftrightarrow $pivot and pivot$\leftrightarrow $target parallel data but no source$\leftrightarrow $target parallel data. In this scenario, we propose a simple but effective transfer approach, the key idea of which is to relieve the burden of the domain shift problem by means of cross-lingual pre-training. To this end, we firstly investigate the performance of two existing cross-lingual pre-training methods proposed by BIBREF11 in zero-shot translation scenario. Besides, a novel pre-training method called BRidge Language Modeling (BRLM) is designed to make full use of the source$\leftrightarrow $pivot bilingual data to obtain a universal encoder for different languages. Once the universal encoder is constructed, we only need to train the pivot$\rightarrow $target model and then test this model in source$\rightarrow $target direction directly. The main contributions of this paper are as follows:
We propose a new transfer learning approach for NMT which uses the cross-lingual language model pre-training to enable a high performance on zero-shot translation.
We propose a novel pre-training method called BRLM, which can effectively alleviates the distance between different source language spaces.
Our proposed approach significantly improves zero-shot translation performance, consistently surpassing pivoting and multilingual approaches. Meanwhile, the performance on supervised translation direction remains the same level or even better when using our method.
In this paper, we explore the transfer learning in a common zero-shot scenario where there are a lot of source↔pivot and pivot↔target parallel data but no source↔target parallel data.
|
In what scenario, does transfer learning being explored?
|
In a common zero-shot scenario where there are a lot of source↔pivot and pivot↔target parallel data but no source↔target parallel data.
|
null | false
| null |
Cadoo: a children's version of the game that has some elements of tic-tac-toe. Sold between October 2007 and January 2008, Cadoo units were recalled in January 2008 by the U.S. Consumer Product Safety Commission because "the surface paint on the die contains excessive levels of lead, violating the federal lead paint standard."
|
What is Cadoo?
|
Cadoo is a children version of the Cranium party game. It has some elements of tic-tac-toe, a decoding glass, and a pile of cards that the player can draw from and performing actions accordingly. The game was sold between October 2007 and January 2008. However, it was recalled by the U.S. Consumer Product Safety Commission "the surface paint on the die contains excessive levels of lead, violating the federal lead paint standard."
|
null | false
| null |
What are the careers of your dream?
|
Professional Baseball Player, astronaut, president of the United States, actor, singer, CEO of major tech company, fighter pilot, nascar driver, artist
|
|
null | false
| null |
What is discrete manufacturing?
|
Discrete manufacturing is a branch of manufacturing where finished goods are distinct and can be easily counted. Discrete manufacturing uses various manufacturing processes like rolling, milling, cutting and assembly in order to produce the finished product. Examples of products due to discrete manufacturing include cars, furniture, televisions and aeroplanes.
|
|
null | false
| null |
Give me some examples of spoonerisms.
|
Spoonerisms are where the first one or two letters of a word are transposed with another word to create two new words with another meaning. Some examples of this are "boiled icicle", "bunny phone" or "must tree".
|
|
1905.11901
| false
| null |
We use the TED data from the IWSLT 2014 German INLINEFORM0 English shared translation task BIBREF38 . We use the same data cleanup and train/dev split as BIBREF39 , resulting in 159000 parallel sentences of training data, and 7584 for development.
To simulate different amounts of training resources, we randomly subsample the IWSLT training corpus 5 times, discarding half of the data at each step. Truecaser and BPE segmentation are learned on the full training corpus; as one of our experiments, we set the frequency threshold for subword units to 10 in each subcorpus (see SECREF7 ). Table TABREF14 shows statistics for each subcorpus, including the subword vocabulary.
FLOAT SELECTED: Table 1: Training corpus size and subword vocabulary size for different subsets of IWSLT14 DE→EN data, and for KO→EN data.
We use the TED data from the IWSLT 2014 German INLINEFORM0 English shared translation task BIBREF38 . We use the same data cleanup and train/dev split as BIBREF39 , resulting in 159000 parallel sentences of training data, and 7584 for development.
Table TABREF14 shows statistics for each subcorpus, including the subword vocabulary.
FLOAT SELECTED: Table 1: Training corpus size and subword vocabulary size for different subsets of IWSLT14 DE→EN data, and for KO→EN data.
|
what amounts of size were used on german-english?
|
Training data with 159000, 80000, 40000, 20000, 10000 and 5000 sentences, and 7584 sentences for development
|
null | false
| 414
|
Results: Table 1 reports attack success rates and median query counts, and Fig. 2 plots cumulative success counts against the maximum queries spent per example (CDFs, modulo normalisation).
|
How do the approaches stack in terms of median/average count in the new experiments?
|
We sometimes provide medians as summary statistics (as in Table 1) to aid the digestion of large amounts of information across competing methods. However, the chosen summary statistic is inherently arbitrary: the CDF curves are the main results and should be treated as such.
|
null | false
| null |
Classify each of the Pokémon characters into different types.
|
Normal, Fire, Water, Grass, Electric, Ice, Fighting, Poison, Ground, Flying, Psychic, Bug, Rock, Ghost, Dark, Dragon, Steel, Fairy.
|
|
null | false
| null |
In different cultures and situations, a clan usually has different meaning than other kin-based groups, such as tribes and bands. Often, the distinguishing factor is that a clan is a smaller, integral part of a larger society such as a tribe, chiefdom, or a state. In some societies, clans may have an official leader such as a chief, matriarch or patriarch; or such leadership role is performed by elders. In others, leadership positions may have to be achieved.
The term tribe is used in many different contexts to refer to a category of human social group. The predominant worldwide usage of the term in English is in the discipline of anthropology. Its definition is contested, in part due to conflicting theoretical understandings of social and kinship structures, and also reflecting the problematic application of this concept to extremely diverse human societies. The concept is often contrasted by anthropologists with other social and kinship groups, being hierarchically larger than a lineage or clan, but smaller than a chiefdom, nation or state. These terms are equally disputed. In some cases tribes have legal recognition and some degree of political autonomy from national or federal government, but this legalistic usage of the term may conflict with anthropological definitions.
In the United States, Native American tribes are legally considered to have "domestic dependent nation" status within the territorial United States, with a government-to-government relationship with the federal government.
|
What is the difference between a clan and a tribe
|
In different cultures and situations, a clan usually has different meaning than other kin-based groups, such as tribes and bands. Often, the distinguishing factor is that a clan is a smaller, integral part of a larger society such as a tribe, chiefdom, or a state. In some societies, clans may have an official leader such as a chief, matriarch or patriarch; or such leadership role is performed by elders. In others, leadership positions may have to be achieved.
The term tribe is used in many different contexts to refer to a category of human social group. The predominant worldwide usage of the term in English is in the discipline of anthropology. Its definition is contested, in part due to conflicting theoretical understandings of social and kinship structures, and also reflecting the problematic application of this concept to extremely diverse human societies. The concept is often contrasted by anthropologists with other social and kinship groups, being hierarchically larger than a lineage or clan, but smaller than a chiefdom, nation or state. These terms are equally disputed. In some cases tribes have legal recognition and some degree of political autonomy from national or federal government, but this legalistic usage of the term may conflict with anthropological definitions.
In the United States, Native American tribes are legally considered to have "domestic dependent nation" status within the territorial United States, with a government-to-government relationship with the federal government.
|
null | false
| null |
Who was the president of China before Xi?
|
It's Jingtao Hu. He was the president of China from 2003 to 2013.
|
|
null | false
| null |
What is the longest floating bridge in the world?
|
The 520 Bridge (Evergreen Point) in Washington State is the longest floating bridge in the world
|
|
null | false
| null |
What is called a Capital charge in finance?
|
If a unit trust manager takes the management charges out of the fund's capital instead of the income it has generated, it is called a capital charge.
|
|
null | false
| null |
How many vaccines should I get, if a dog bites me?
|
There is no common number that applies to all, so you should immediately seek medical assistance. And then, based on the severity of the bite, your health provider can recommend the number of vaccines, along with any other treatment.
|
|
null | false
| 229
|
In recent years, the use of word representations, such as word2vec BIBREF0 , BIBREF1 and GloVe BIBREF2 , has become a key “secret sauce” for the success of many natural language processing (NLP), information retrieval (IR) and machine learning (ML) tasks. The empirical success of word embeddings raises an interesting research question: Beyond words, can we learn fixed-length distributed representations for pieces of texts? The texts can be of variable-length, ranging from paragraphs to documents. Such document representations play a vital role in a large number of downstream NLP/IR/ML applications, such as text clustering, sentiment analysis, and document retrieval, which treat each piece of text as an instance. Learning a good representation that captures the semantics of each document is thus essential for the success of such applications.
In this paper, we introduce KeyVec, a neural network model that learns densely distributed representations for documents of variable-length. In order to capture semantics, the document representations are trained and optimized in a way to recover key information of the documents. In particular, given a document, the KeyVec model constructs a fixed-length vector to be able to predict both salient sentences and key words in the document. In this way, KeyVec conquers the problem of prior embedding models which treat every word and every sentence equally, failing to identify the key information that a document conveys. As a result, the vectorial representations generated by KeyVec can naturally capture the topics of the documents, and thus should yield good performance in downstream tasks.
We evaluate our KeyVec on two text understanding tasks: document retrieval and document clustering. As shown in the experimental section SECREF5 , KeyVec yields generic document representations that perform better than state-of-the-art embedding models.
In this paper, we introduce KeyVec, a neural network model that learns densely distributed representations for documents of variable-length.
|
What is the KeyVec introduced in this paper?
|
It is a neural network model that learns densely distributed representations for documents of variable-length.
|
null | false
| null |
What makes a good sales account executive?
|
A good Sales account executive will be able to multi-task, manage customer engagement activity for multiple teams, and expand relationships within a customer account by sharing the value proposition of the Company.
|
|
1703.09684
| false
| null |
VQA research began in earnest in late 2014 when the DAQUAR dataset was released BIBREF0 . Including DAQUAR, six major VQA datasets have been released, and algorithms have rapidly improved. On the most popular dataset, `The VQA Dataset' BIBREF1 , the best algorithms are now approaching 70% accuracy BIBREF2 (human performance is 83%). While these results are promising, there are critical problems with existing datasets in terms of multiple kinds of biases. Moreover, because existing datasets do not group instances into meaningful categories, it is not easy to compare the abilities of individual algorithms. For example, one method may excel at color questions compared to answering questions requiring spatial reasoning. Because color questions are far more common in the dataset, an algorithm that performs well at spatial reasoning will not be appropriately rewarded for that feat due to the evaluation metrics that are used.
VQA research began in earnest in late 2014 when the DAQUAR dataset was released BIBREF0
|
From when are many VQA datasets collected?
|
late 2014
|
null | false
| null |
The Tampa Bay Rowdies are an American professional soccer team based in St. Petersburg, Florida. The club was founded in 2008 and first took the pitch in 2010. Since 2017, the Rowdies have been members of the USL Championship in the second tier of the American soccer pyramid. They formerly played in USSF Division 2 (in 2010) and the North American Soccer League (NASL) (from 2011 to 2016), which were also second-tier leagues. The Rowdies play their home games at Al Lang Stadium on St. Petersburg's downtown waterfront.
The current club is a phoenix club of the original Tampa Bay Rowdies, who were active from 1975 until 1993, most notably in the original North American Soccer League. It shares its name, logo, and some of its club culture with the original club. The owners of the current club announced their intention to use the old Rowdies' trademarks at its introductory press conference in 2008. However, licensing issues forced the club to use the name FC Tampa Bay until December 2011, when it gained full rights to the Rowdies name and other intellectual property. The current Rowdies have always used the same green and yellow color scheme and "hoops" as the original team, even when they could not yet use the Rowdies name.
The Rowdies captured the NASL championship in Soccer Bowl 2012, and their team shield includes two stars: one for their 2012 win and one for the 1975 Soccer Bowl championship won by the original Rowdies. The club has had a long-standing rivalry with the Fort Lauderdale Strikers, with whom they have contested the Florida Derby since the original Rowdies and Strikers first met in 1977. The Rowdies were also named co-league champions in 2020 after winning the USL Regular Season title and Eastern Conference Championship, but the title game was canceled due to COVID-19. They were Eastern Conference Champions again in 2021, but lost in the title game.
In October 2018, the Tampa Bay Rays, the area's Major League Baseball franchise, announced plans to purchase the Rowdies and assume control of Al Lang Stadium.
|
Based on the reference text, what colors do the Tampa Bay Rowdies wear and what stadium do they play in?
|
The Tampa Bay Rowdies wear green and yellow and play in Al Lang Stadium in downtown St. Petersburg.
|
null | false
| null |
What are some popular search engines?
|
Google, Bing, DuckDuckGo, Yandex, Baidu
|
|
null | false
| 83
|
We developed the annotation guidelines jointly with an experienced annotator, who is a native Arabic speaker with a good knowledge of various Arabic dialects. We made sure that our guidelines were compatible with those of OffensEval2019. The annotator carried out all annotation. Tweets were given one or more of the following four labels: offensive, vulgar, hate speech, or clean. Since the offensive label covers both vulgar and hate speech and vulgarity and hate speech are not mutually exclusive, a tweet can be just offensive or offensive and vulgar and/or hate speech. The annotation adhered to the following guidelines:
Offensive tweets contain explicit or implicit insults or attacks against other people, or inappropriate language, such as:
Direct threats or incitement, ex: احرقوا> مقرات المعارضة> (“AHrqwA mqrAt AlmEArDp” – “burn the headquarters of the opposition”) and هذا المنافق يجب قتله> (“h*A AlmnAfq yjb qtlh” – “this hypocrite needs to be killed”).
Insults and expressions of contempt, which include: Animal analogy, ex: يا كلب> (“yA klb” – “O dog”) and كل تبن> (“kl tbn” – “eat hay”).; Insult to family members, ex: يا روح أمك> (“yA rwH Amk” – “O mother's soul”); Sexually-related insults, ex: يا ديوث> (“yA dywv” – “O person without envy”); Damnation, ex: الله يلعنك> (“Allh ylEnk” – “may Allah/God curse you”); and Attacks on morals and ethics, ex: يا كاذب> (“yA kA*b” – “O liar”)
Vulgar tweets are a subset of offensive tweets and contain profanity, such as mentions of private parts or sexual-related acts or references.
Hate speech tweets, a subset of offensive tweets containing offensive language targeting group based on common characteristics such as: Race, ex: يا زنجي> (“yA znjy” – “O negro”); Ethnicity, ex. الفرس الأنجاس> (“Alfrs AlAnjAs” – “Impure Persians”); Group or party, ex: أبوك شيوعي> (“Abwk $ywEy” – “your father is communist”); and Religion, ex: دينك القذر> (“dynk Alq*r” – “your filthy religion”).
Clean tweets do not contain vulgar or offensive language. We noticed that some tweets have some offensive words, but the whole tweet should not be considered as offensive due to the intention of users. This suggests that normal string match without considering contexts will fail in some cases. Examples of such ambiguous cases include: Humor, ex: يا عدوة الفرحة ههه> (“yA Edwp AlfrHp hhh” – “O enemy of happiness hahaha”); Advice, ex: لا تقل لصاحبك يا خنزير> (“lA tql lSAHbk yA xnzyr” – “don't say to your friend: You are a pig”); Condition, ex: إذا عارضتهم يقولون يا عميل> (“A*A EArDthm yqwlwn yA Emyl” – “if you disagree with them they will say: You are an agent”); Condemnation, ex: لماذا نسب بقول: يا بقرة؟> (“lmA*A nsb bqwl: yA bqrp?” – “Why do we insult others by saying: O cow?”); Self offense, ex: تعبت من لساني القذر> (“tEbt mn lsAny Alq*r” – “I am tired of my dirty tongue”); Non-human target, ex: يا بنت المجنونة يا كورة> (“yA bnt Almjnwnp yA kwrp” – “O daughter of the crazy one O football”); and Quotation from a movies or a story, ex: تاني يا زكي! تاني يا فاشل> (“tAny yA zky! tAny yA fA$l” – “again O Zaky! again O loser”). For other ambiguous cases, the annotator searched Twitter to find how actual users used expressions.
Table TABREF11 shows the distribution of the annotated tweets. There are 1,915 offensive tweets, including 225 vulgar tweet and 506 hate speech tweets, and 8,085 clean tweets. To validate the quality of annotation, a random sample of 100 tweets from the data, containing 50 offensive and 50 clean tweets, was given to additional three annotators. We calculated the Inter-Annotator Agreement between the annotators using Fleiss’s Kappa coefficient BIBREF23. The Kappa score was 0.92 indicating high quality annotation and agreement.
Tweets were given one or more of the following four labels: offensive, vulgar, hate speech, or clean.
|
What labels have been given to the tweets?
|
Tweets were given one or more of the following four labels: offensive, vulgar, hate speech, or clean.
|
null | false
| 104
|
Since the CoNLL annotations have 21 semantic roles in total, we use 21 roles in our model as well as the baseline. Following garg2012unsupervised, we set the number of PRs to 2 (excluding INLINEFORM0 , INLINEFORM1 and INLINEFORM2 ), and SRs to 21-2=19. Table TABREF27 shows the results.
In the first setting (Line 1), we train and test the monolingual model on the CoNLL data. We observe significant improvements in F1 score over the Baseline (Line 0) in both languages. Using the CoNLL 2009 dataset alone, titovcrosslingual report an F1 score of 80.9% (PU=86.8%, CO=75.7%) for German. Thus, our monolingual model outperforms their monolingual model in German. For English, they report an F1 score of 83.6% (PU=87.5%, CO=80.1%), but note that our English results are not directly comparable to theirs due to differences argument identification, as discussed in section SECREF25 . As their argument identification score is lower, perhaps their system is discarding “difficult” arguments which leads to a higher clustering score.
In the second setting (Line 2), we use the additional monolingual Europarl (EP) data for training. We get equivalent results in English and a significant improvement in German compared to our previous setting (Line 1). The German dataset in CoNLL is quite small and benefits from the additional EP training data. In contrast, the English model is already quite good due to a relatively big dataset from CoNLL, and good accuracy syntactic parsers. Unfortunately, titovcrosslingual do not report results with this setting.
The third setting (Line 3) gives the results of our multilingual model, which adds the word alignments in the EP data. Comparing with Line 2, we get non-significant improvements in both languages. titovcrosslingual obtain an F1 score of 82.7% (PU=85.0%, CO=80.6%) for German, and 83.7% (PU=86.8%, CO=80.7%) for English. Thus, for German, our multilingual Bayesian model is able to capture the cross-lingual patterns at least as well as the external penalty term in BIBREF6 . We cannot compare the English results unfortunately due to differences in argument identification.
We also compared monolingual and bilingual training data using a setting that emulates the standard supervised setup of separate training and test data sets. We train only on the EP dataset and test on the CoNLL dataset. Lines 4 and 5 of Table TABREF27 give the results. The multilingual model obtains small improvements in both languages, which confirms the results from the standard unsupervised setup, comparing lines 2 to 3.
These results indicate that little information can be learned about semantic roles from this parallel data setup. One possible explanation for this result is that the setup itself is inadequate. Given the definition of aligned arguments, only 8% of English arguments and 17% of German arguments are aligned. This plus our experiments suggest that improving the alignment model is a necessary step to making effective use of parallel data in multilingual SRI, for example by joint modeling with SRI. We leave this exploration to future work.
The third setting (Line 3) gives the results of our multilingual model, which adds the word alignments in the EP data.
|
What was added to the EP data in Line 1?
|
The word alignments.
|
2003.06044
| false
| null |
Dialogue act (DA) characterizes the type of a speaker's intention in the course of producing an utterance and is approximately equivalent to the illocutionary act of BIBREF0 or the speech act of BIBREF1. The recognition of DA is essential for modeling and automatically detecting discourse structure, especially in developing a human-machine dialogue system. It is natural to predict the Answer acts following an utterance of type Question, and then match the Question utterance to each QA-pair in the knowledge base. The predicted DA can also guide the response generation process BIBREF2. For instance, system generates a Greeting type response to former Greeting type utterance. Moreover, DA is beneficial to other online dialogue strategies, such as conflict avoidance BIBREF3. In the offline system, DA also plays a significant role in summarizing and analyzing the collected utterances. For instance, recognizing DAs of a wholly online service record between customer and agent is beneficial to mine QA-pairs, which are selected and clustered then to expand the knowledge base. DA recognition is challenging due to the same utterance may have a different meaning in a different context. Table TABREF1 shows an example of some utterances together with their DAs from Switchboard dataset. In this example, utterance “Okay.” corresponds to two different DA labels within different semantic context.
DA recognition is aimed to assign a label to each utterance in a conversation. It can be formulated as a supervised classification problem. There are two trends to solve this problem: 1) as a sequence labeling problem, it will predict the labels for all utterances in the whole dialogue history BIBREF13, BIBREF14, BIBREF9; 2) as a sentence classification problem, it will treat utterance independently without any context history BIBREF5, BIBREF15. Early studies rely heavily on handcrafted features such as lexical, syntactic, contextual, prosodic and speaker information and achieve good results BIBREF13, BIBREF4, BIBREF16.
Dialogue act (DA) characterizes the type of a speaker's intention in the course of producing an utterance and is approximately equivalent to the illocutionary act of BIBREF0 or the speech act of BIBREF1. The recognition of DA is essential for modeling and automatically detecting discourse structure, especially in developing a human-machine dialogue system. It is natural to predict the Answer acts following an utterance of type Question, and then match the Question utterance to each QA-pair in the knowledge base. The predicted DA can also guide the response generation process BIBREF2. For instance, system generates a Greeting type response to former Greeting type utterance.
DA recognition is aimed to assign a label to each utterance in a conversation. It can be formulated as a supervised classification problem. There are two trends to solve this problem: 1) as a sequence labeling problem, it will predict the labels for all utterances in the whole dialogue history BIBREF13, BIBREF14, BIBREF9; 2) as a sentence classification problem, it will treat utterance independently without any context history BIBREF5, BIBREF15.
|
What is dialogue act recognition?
|
The answers are shown as follows:
* DA recognition is aimed to assign a label to each utterance in a conversation. It can be formulated as a supervised classification problem.
|
null | false
| 20
|
The Russian and Ukrainian languages are mainly spoken in Russian Federation and the Ukraine and belong to the East-Slavic group of the Indo-European language family. They share many common morphosyntactic features: both are SVO languages with free word order and rich morphology, both use the Cyrillic alphabet and share many common cognates.
Both Russia and the Ukraine have common academic tradition that makes it easier to collect corpora, which are comparable in terms of both genre and strictly defined academic fields. We work with such a corpus of Russian and Ukrainian academic texts, initially collected for the purposes of cross-lingual plagiarism detection. This data is available online through a number of library services, but unfortunately cannot be republished due to copyright limitations.
The Ukrainian subcorpus contains about 60 thousand extended summaries (Russian and Ukrainian russian`автореферат', `avtoreferat') of theses submitted between 1998 and 2011. The Russian subcorpus is smaller in the number of documents (about 16 thousand, approximately the same time period), but the documents are full texts of theses, thus the total volume of the Russian subcorpus is notably larger: 830 million tokens versus 250 million tokens in the Ukrainian one. Generally, the texts belong to one genre that can be defined as post-Soviet expository academic prose, submitted for academic degree award process.
The documents were converted to plain text files from MS Word format in the case of the Ukrainian subcorpus and mainly from OCRed PDF files in the case of the Russian subcorpus. Because of this, the Russian documents often suffer from OCR artifacts, such as words split with line breaks, incorrectly recognized characters and so on. However, it does not influence the resulting model much, as we show below.
Both Ukrainian and Russian documents come with meta data allowing to separate them into academic fields, with economics, medicine and law being most frequent topics for the Ukrainian data and economics, history and pedagogy dominating the Russian data.
For evaluation, 3 topics were used, distant enough from each other and abundantly presented in both subcorpora: economics, law and history. We randomly selected 100 texts in each language for each topic. As an average length of Russian texts is significantly higher (them being full theses), we cropped them, leaving only the first 5 thousand words, to mimic the size of the Ukrainian summaries. These 600 documents in 3 classes are used as a test set (see Section "Experiment Design and Evaluation" for the description of the conducted experiments).
The corpora (including test set) were PoS-tagged. Each word was replaced with its lemma followed by a PoS-tag (`russianдиссертация_S', `russianдиссертацiя_N'). Functional parts of speech (conjunctions, pronouns, prepositions, etc.) and numerals were removed from the texts.
For evaluation, 3 topics were used, distant enough from each other and abundantly presented in both subcorpora: economics, law and history.
|
How many topics were used for evaluation?
|
3 topics.
|
1901.02262
| false
| null |
We conducted experiments on the two tasks of MS MARCO 2.1 BIBREF5 . The answer styles considered in the experiments corresponded to the two tasks. The NLG task requires a well-formed answer that is an abstractive summary of the question and ten passages, averaging 16.6 words. The Q&A task also requires an abstractive answer but prefers a more concise answer than the NLG task, averaging 13.1 words, where many of the answers do not contain the context of the question. For instance, for the question “tablespoon in cup”, the answer in the Q&A task will be “16”, and the answer in the NLG task will be “There are 16 tablespoons in a cup.” In addition to the ALL dataset, we prepared two subsets (Table 1 ). The ANS set consists of answerable questions, and the WFA set consists of the answerable questions and well-formed answers, where WFA $\subset $ ANS $\subset $ ALL.
The answer styles considered in the experiments corresponded to the two tasks. The NLG task requires a well-formed answer that is an abstractive summary of the question and ten passages, averaging 16.6 words. The Q&A task also requires an abstractive answer but prefers a more concise answer than the NLG task, averaging 13.1 words, where many of the answers do not contain the context of the question.
|
What do they mean by answer styles?
|
well-formed sentences vs concise answers
|
null | false
| null |
Tell me whether these are alcohol or not
|
Beer, Coca, Gin, Water, Wine, Coffee, Tea
|
|
null | false
| null |
Give me a bulleted list of the last seven World Series winners
|
- 2022: Houston Astros
- 2021: Atlanta Braves
- 2020: Los Angeles Dodgers
- 2019: Washington Nationals
- 2018: Boston Red Sox
- 2017: Houston Astros
- 2016: Chicago Cubs
|
|
null | false
| null |
The C++ language has two main components: a direct mapping of hardware features provided primarily by the C subset, and zero-overhead abstractions based on those mappings. Stroustrup describes C++ as "a light-weight abstraction programming language for building and using efficient and elegant abstractions"; and "offering both hardware access and abstraction is the basis of C++. Doing it efficiently is what distinguishes it from other languages." C++ inherits most of C's syntax.
|
Extract the two main component of C++ in a bulleted list based on this passage.
|
-direct mapping of hardware features provided primarily by the C subset
-zero-overhead abstractions based on those mappings
|
null | false
| null |
who is the prime minister of india
|
Ah, India is an awesome place to be and so are the people, same can be said for the prime minister also.
As of 2023 Mr. Narendra Damodardas Modi is the 14th and current prime minister of India since May 2014. He was born on 17 September 1950.
|
|
null | false
| 21
|
Zork1 is one of the first text-adventure games and heavily influences games released later in terms of narrative style and game structure. It is a dungeon crawler where the player must explore a vast world and collect a series of treasures. It was identified by BIBREF2 as a moonshot game and has been the subject of much work in leaning agents BIBREF12, BIBREF7, BIBREF11, BIBREF8. Rewards are given to the player when they collect treasures as well as when important intermediate milestones needed to further explore the world are passed. Figure FIGREF15 and Figure FIGREF1 show us a map of the world of Zork1 and the corresponding quest structure.
The bottleneck seen at a score of around 40 is when the player first enters the cellar on the right side of the map. The cellar is dark and you need to immediately light the lamp to see anything. Attempting to explore the cellar in the dark results in you being instantly killed by a monster known as a “grue”.
We make no changes from the graph update rules used by BIBREF6. Candidate interactive objects are identified by performing part-of-speech tagging on the current observation, identifying singular and proper nouns as well as adjectives, and are then filtered by checking if they can be examined using the command $examine$ $OBJ$. Only the interactive objects not found in the inventory are linked to the node corresponding to the current room and the inventory items are linked to the “you” node. The only other rule applied uses the navigational actions performed by the agent to infer the relative positions of rooms, e.g. $\langle kitchen,down,cellar \rangle $ when the agent performs $go$ $down$ when in the kitchen to move to the cellar.
Hyperparameters used for our agents are given below. Patience and buffer size are used for the policy chaining method as described in Section SECREF2. Cell step size is a parameter used for Go-Explore and describes how many steps are taken when exploring in a given cell state. Base hyperparameters for KG-A2C are taken from BIBREF6 and the same parameters are used for A2C.
The cellar is dark and you need to immediately light the lamp to see anything. Attempting to explore the cellar in the dark results in you being instantly killed by a monster known as a “grue”.
|
What is "grue" in the game Zork1?
|
It is a monster in the dark cellar.
|
null | false
| null |
How many types of elk are found in North America?
|
Four subspecies of elk can be found in North America. Tule Elk, Roosevelt Elk, Rocky Mountain Elk and Manitoba Elk.
|
|
null | false
| null |
What is the oldest country in the world?
|
Egypt is recognized as the oldest country in the world, its founding accepted to be approximately 3500 BC.
|
|
null | false
| null |
Identify the bird from the list: Zebra, Brambling, Blackfish
|
Brambling
|
|
null | false
| null |
Kia claims that the Stinger accelerates from zero to 100 km/h (62 mph) in 7.7, 6 and 4.9 seconds for the 2.2-liter diesel, 2.0-liter petrol and 3.3-liter petrol respectively. Schreyer reportedly drove a pre-production Stinger GT at a top speed of 269 km/h (167 mph) on the Autobahn.
During a test by Car and Driver, an all-wheel-drive U.S. spec GT 3.3T with Michelin Pilot Sport 4 tires achieved 0–60 mph (0–97 km/h) in 4.6 seconds on the track, reached 0.91 g on the skidpad and was able to stop from 70 mph (113 km/h) in 164 feet (50 m). According to this publication, the U.S. model's top speed is governed at 167 mph (269 km/h) per Kia specs. In tests conducted by Motor Trend, the four-cylinder U.S. spec Stinger 2.0 RWD on Bridgestone Potenza tires reached 60 mph (97 km/h) in 6.6 seconds, completed the 1⁄4-mile (0.4 km) run in 15 seconds and stopped from 60 mph (97 km/h) in 126 feet (38 m). The average lateral acceleration recorded in track testing was 0.85 g.
|
Given this paragraph, what is the top speed of a Kia Stinger?
|
The top speed of a Kia Stinger is 269km/h (167mph) according to this text.
|
null | false
| null |
Which country is the state of Illinois apart of?
|
The United States of America
|
|
null | false
| null |
Magic realism or magical realism is a style of literary fiction and art. It paints a realistic view of the world while also adding magical elements, often blurring the lines between fantasy and reality. Magic realism often refers to literature in particular, with magical or supernatural phenomena presented in an otherwise real-world or mundane setting, commonly found in novels and dramatic performances.: 1–5 Despite including certain magic elements, it is generally considered to be a different genre from fantasy because magical realism uses a substantial amount of realistic detail and employs magical elements to make a point about reality, while fantasy stories are often separated from reality. Magical realism is often seen as an amalgamation of real and magical elements that produces a more inclusive writing form than either literary realism or fantasy.
The term magic realism is broadly descriptive rather than critically rigorous, and Matthew Strecher (1999) defines it as "what happens when a highly detailed, realistic setting is invaded by something too strange to believe." The term and its wide definition can often become confused, as many writers are categorized as magical realists. The term was influenced by a German and Italian painting style of the 1920s which were given the same name. In The Art of Fiction, British novelist and critic David Lodge defines magic realism: "when marvellous and impossible events occur in what otherwise purports to be a realistic narrative - is an effect especially associated with contemporary Latin-American fiction (for example the work of the Colombian novelist, Gabriel García Marquez) but it is also encountered in novels from other continents, such as those of Günter Grass, Salman Rushdie and Milan Kundera. All these writers have lived through great historical convulsions and wrenching personal upheavals, which they feel they cannot be adequately represented in a discourse of undisturbed realism", citing Kundera's The Book of Laughter and Forgetting as an exemplar. " Michiko Kakutani writes that "The transactions between the extraordinary and the mundane that occur in so much Latin American fiction are not merely a literary technique, but also a mirror of a reality in which the fantastic is frequently part of everyday life." Magical realism often mixes history and fantasy, as in Salman Rushdie's Midnight's Children, in which the children born at midnight on August 15, 1947, the moment of India's independence, are telepathically linked.
Irene Guenther (1995) tackles the German roots of the term, and how an earlier magic realist art is related to a later magic realist literature; meanwhile, magical realism is often associated with Latin-American literature, including founders of the genre, particularly the authors Gabriel García Márquez, Isabel Allende, Jorge Luis Borges, Juan Rulfo, Miguel Ángel Asturias, Elena Garro, Mireya Robles, Rómulo Gallegos and Arturo Uslar Pietri. In English literature, its chief exponents include Neil Gaiman, Salman Rushdie, Alice Hoffman, Nick Joaquin, and Nicola Barker. In Bengali literature, prominent writers of magic realism include Nabarun Bhattacharya, Akhteruzzaman Elias, Shahidul Zahir, Jibanananda Das and Syed Waliullah. In Japanese literature, one of the most important authors of this genre is Haruki Murakami. In Kannada literature, the writers Shivaram Karanth and Devanur Mahadeva have infused magical realism in their most prominent works. In Polish literature, magic realism is represented by Olga Tokarczuk, the 2018 Nobel Prize laureate in Literature.
|
Based on the paragraph below, name some authors outside of Latin America who employ magical realism in their work.
|
While Latin American authors are most closely associated with magical realism, there are many others who adopt this style. Notable ones include Salman Rushdie, Neil Gaiman, Haruki Murakami and Olga Tokarczuk.
|
null | false
| null |
Blockbuster LLC, formerly known as Blockbuster Video, is an American brand currently owned by Dish Network. It was originally founded by David Cook in 1985 as a home video rental shops, but later grew into video game rentals, DVD-by-mail, streaming, video on demand, and cinema theater. Previously operated by Blockbuster Entertainment, Inc., the company expanded internationally throughout the 1990s. At its peak in 2004, Blockbuster consisted of 9,094 stores and employed approximately 84,300 people: 58,500 in the United States and 25,800 in other countries.
|
How many stores did Blockbuster have at its peak?
|
In 2004, Blockbuster consisted of 9,094 stores
|
null | false
| null |
The Buddhist Uprising of 1966 (Vietnamese: Nổi dậy Phật giáo 1966), or more widely known in Vietnam as the Crisis in Central Vietnam (Vietnamese: Biến động Miền Trung), was a period of civil and military unrest in South Vietnam, largely focused in the I Corps area in the north of the country in central Vietnam. The area is a heartland of Vietnamese Buddhism, and at the time, activist Buddhist monks and civilians were at the forefront of opposition to a series of military juntas that had been ruling the nation, as well as prominently questioning the escalation of the Vietnam War.
During the rule of the Catholic Ngô Đình Diệm, the discrimination against the majority Buddhist population generated the growth of Buddhist institutions as they sought to participate in national politics and gain better treatment. In 1965, after a series of military coups that followed the fall of the Diệm regime in 1963, Air Marshal Nguyễn Cao Kỳ and General Nguyễn Văn Thiệu finally established a stable junta, holding the positions of Prime Minister and figurehead Chief of State respectively. The Kỳ-Thiệu regime was initially almost a feudal system, being more of an alliance of warlords than a state as each corps commander ruled his area as his own fiefdom, handing some of the taxes they collected over to the government in Saigon and keeping the rest for themselves. During that time, suspicion and tension continued between the Buddhist and Catholic factions in Vietnamese society.
|
Can you tell me what the buddhist uprising is about in a few sentences?
|
The Buddhist Uprising was a civil and military crisis in South Vietnam, during 1966. At that time, civilians and Buddhist monks strongly questioned the activities schemes of military officers and the reasoning behind the Vietnam War.
The Buddhist population was discriminated against during the rule of Catholics, which spurred them into forming their own political party. After a series of military. coups in 1965, they successfully gained command of the political power. Tension remained between the two religious power.
|
null | false
| null |
Why is November 14 celebrated as world diabetes day?
|
Sir Frederick Banting who discovered Insulin in 1922 was born on November 14 which is why November 14 is celebrated as world diabetes day
|
|
null | false
| null |
How do you train for a marathon?
|
A lot of people who run marathons will start training about six months before the marathon. You would typically start by running short lengths, a couple times a week, such as 2 to 3 miles a day, 2 to 3 days per week after a couple weeks of that, you would build up to six or 7 miles per day Often marathon runners will run one day a week longer, such as 10 miles on a Saturday or Sunday. About one month out is when you shored train the longest, such as a 15 mile run on the weekend. Within about two weeks of the race you want to start cooling down and resting. usually a day or two before before the race you will start carb loading, which is when you eat a lot of carbohydrates. The day of the race is usually a celebratory event, as running a marathon is a big milestone!
|
|
null | false
| null |
How do I find my Tesla charging station?
|
Access charge settings by touching Controls > Charging when Tesla is in Park. You can also touch the battery icon on the touchscreen to access charge settings.
|
|
null | false
| null |
What’s so interesting about Pokemon Go?
|
Pokemon Go is so interesting because I can walk around (or even drive around) to catch a rich variety of Pokemons almost everywhere. Moreover, there are people playing the game around the globe, so when I want to battle, I can do it even in another country! However, most importantly, Pokemon Go is one of my son's favorite game. We often play it together and talk about it together.
|
|
null | false
| null |
What is a unit trust?
|
People can invest in unit trusts by buying units. The managers of the trust use the money people invest to buy investments. The fund manager values the fund's assets from time to time and puts a new price on the fund's units.
|
|
null | false
| null |
Describe a trip to Tokyo. What should I be prepared for? What should I eat? Any can't-miss attractions?
|
Tokyo is a massive city and even if I were to give you hundreds of recommendations, they would not begin to scratch the surface. What follows is a rundown of good things to know, food to eat, and places you may want to visit.
General tips and tricks:
- Bring cash: Do a currency exchange and pick up plenty of Yen at the airport — Tokyo is a very cash-heavy town and your local currency won’t cut it.
- Take the Narita Express: If you’re flying into Narita, the airport is far from the city center and the easiest way to get there is via Narita Express (abbreviated to N’EX), run by JR East. You have to book an actual seat on the Narita Express, and if you don't speak Japanese, don't try using one of the machines at the airport, go to an actual desk to speak to someone who can help you buy a round-trip ticket.
- Use the Metro: Hands down, the best way to get around Tokyo. If you stay in Shinjuku, Harajuku, Shibuya or Shinagawa, you’ll be on the Yamanote Line, which will get you almost everywhere you want to go (or to an easy transfer). You’ll need to buy a reusable Suica card — again, highly highly recommend having Yen with you so you can easily load up your card. All of the Suica machines have translation options.
- Make dinner reservations: It can be difficult to make reservations if you don't speak Japanse! If you’re staying at a hotel, use your concierge. If not, try one of following sites: Tablecheck, Tabelog and Retty.
- Make use of these helpful phrases:
Thank you (formal): Arigatou Gozaimasu (pronounced “Ah-di-got-oh Goze-eye-moss”)
I’m sorry / excuse me: Sumimasen (pronounced “sue-me-mah-sahn”)
Good morning: Ohayo Gozaimasu (pronounced “oh-hi-yo Goze-eye-moss”)
Good evening: Konbanwa (pronounced: “kon-bon-wah”)
What to eat:
- Omakase: If you enjoy sushi (which you probably do if you're visiting Japan!), don't miss the chance to enjoy an Omakase meal. Sushi chefs take their art seriously and whether you choose to go to a classic high-end restaurant like Sukiyabashi Jiro Roppongi (from Jiro Dreams of Sushi) or Nishiazabu Sushi Shin, or something more inventive like Udatsu, or the even more casual Sushipanchi, you will be treated to a true experience.
- Gyoza: Who doesn't love dumplings? You should be able to find casual gyoza options throughout the city.
Yakitori: Another casual dinning experience with tons of options. If you're interested in truly experiencing yakitori and aren't afraid of squeezing through alleyways for it, visit the Golden Gai or Piss Alley.
- Ramen: I don’t have specific recommendations for ramen because some of the best places are "holes in the wall" that you wander into. You can download the Ramen Beast app if you need direction.
Attractions and experiences to consider:
- Meiji Jingu: This is probably the most famous shrine in Tokyo (in Shibuya, near Harajuku) and definitely worth visiting.
- Senso-ji: Famous Buddhist temple.
- Imperial Palace: The residential place the Emperors and their families since 1868.
- Sumo wrestling tournaments: Grand tournaments only happen a few times a year, but check the schedule and see if you can catch a day of matches! It’s fascinating and even more enjoyable if you read up on the rituals in advance.
- Public baths or sentos: Experience Japanese bathing culture, but be ready to de-robe and cover any tattoos!
|
|
null | false
| null |
What are some common errors when using Kafka with Spark?
|
Common errors include issues with configuration settings, such as the bootstrap servers or topic names. Other issues may arise from incompatibility between the Kafka and Spark versions or network connectivity problems.
|
|
null | false
| null |
What is the Big 10 conference?
|
Large universities in the United States have organized themselves into conferences primarily for playing each other in sports. One of the longest-standing conferences is called the "Big 10" because it originally included 10 midwestern universities. It has since grown to include several east coast universities and will soon welcome two schools from California, so that the Big 10 will actually be comprised of 16 universities. This supersized conference has secured huge television contracts that make it one of the two most powerful conferences in the nation.
|
|
null | false
| null |
Why did Kim Kardashian and Ye divorce?
|
Kim Kardashian and Ye experienced marital stress from many causes. The initial commonly cited reason by social outlets for their separation and later divorce was tensions that arose from Ye's erratic behavior on social media and in public spheres.
|
|
null | false
| null |
Who are the tallest people in the world?
|
The Dutch are the tallest people in the world, with an average height between 175-176cm (depending on your source). The Dutch man drive this number up, with an average of 183 cm, the dutch ladies are only 171 cm on average.
Why they are so tall is unknown. Some think it is a combination of genetics, universal healthcare, and dairy at that delicious Gouda cheese consumption - others think the local delicacies called "bitterballen" might have something to do with it.
|
|
null | false
| null |
Elzéar Emmanuel Arène Abeille de Perrin (3 January 1843, Marseille – 9 October 1910, Marseille) was a French entomologist.
Abeille de Perrin was a lawyer in Marseille. He gave all his free time to entomology and was a member of the Société entomologique de France for twenty years. He was especially interested in the cave species of the Pyrenees. His best known publications are Monographie des malachites (1869), Études sur les coléoptères cavernicoles, suivies de la description de 27 coléoptères nouveaux français (1872), Notes sur les leptodirites (1878), and Synopsis critique et synonymique des chrysides de France (1878).
His collection of Palearctic, Coleoptera, Hymenoptera, Diptera, and Orthoptera is conserved in Muséum national d'histoire naturelle in Paris.
|
What is Elzéar Abeille de Perrin known for?
|
Elzéar Abeille de Perrin is best known as a French entomologist. He specialized in studying cave species in the Pyrenees. His best known publications are Monographie des malachites (1869), Études sur les coléoptères cavernicoles, suivies de la description de 27 coléoptères nouveaux français (1872), Notes sur les leptodirites (1878), and Synopsis critique et synonymique des chrysides de France (1878).
|
null | false
| null |
Identify which animal species is alive or extinct: Smilosuchus, Brown Bear
|
Brown Bear is alive, Smilosuchus is extinct.
|
|
null | false
| null |
Give me a bulleted list of the Seven Wonders of Portugal and their construction dates:
|
In Portugal, you can find these 7 wonders:
- Castle of Guimarães, 10th century
- Castle of Óbidos, 1195
- Batalha Monastery, 1385
- Alcobaça Monastery, 1153
- Jerónimos Monastery, 1502
- Pena Palace, 1854
- Belém Tower, 1521
|
|
null | false
| null |
Demeter (Δημήτηρ, Dēmḗtēr)Goddess of grain, agriculture, harvest, growth, and nourishment. Demeter, whose Roman counterpart is Ceres, is a daughter of Cronus and Rhea, and was swallowed and then regurgitated by her father. She is a sister of Zeus, by whom she bore Persephone, who is also known as Kore, i.e. "the girl." One of the central myths associated with Demeter involves Hades' abduction of Persephone and Demeter's lengthy search for her. Demeter is one of the main deities of the Eleusinian Mysteries, in which the rites seemed to center around Demeter's search for and reunion with her daughter, which symbolized both the rebirth of crops in spring and the rebirth of the initiates after death. She is depicted as a mature woman, often crowned and holding sheafs of wheat and a torch. Her symbols are the cornucopia, wheat-ears, the winged serpent, and the lotus staff. Her sacred animals include pigs and snakes.
|
From the passage identify the symbols of Demeter. Display the results in a comma separated format.
|
cornucopia, wheat-ears, the winged serpent, the lotus staff
|
null | false
| 493
|
In reinforcement learning (RL) for sequential decision making under uncertainty, existing methods proposed for considering mean-variance (MV) trade-off suffer from computational difficulties in computation of the gradient of the variance term. In this paper, we aim to obtain MV-efficient policies that achieve Pareto efficiency regarding MV trade-off. To achieve this purpose, we train an agent to maximize the expected quadratic utility function, in which the maximizer corresponds to the Pareto efficient policy. Our approach does not suffer from the computational difficulties because it does not include gradient estimation of the variance. In experiments, we confirm the effectiveness of our proposed methods.
Following and, we consider a portfolio composed of two asset types: a liquid asset with a fixed interest rate r l , and a non-liquid asset with a time-dependent interest rate taking either r low nl or r high nl , and the transition follows a switching probability p switch . An investor can sell the liquid asset at every time step t = 1, 2, . . . , T but the non-liquid asset can only be sold after the maturity W periods. This means that when holding 1 liquid asset, we obtain r l per period; when holding 1 non-liquid asset at the t-th period, we obtain r low nl or r high nl at the t + W -th period. Besides, the non-liquid asset has a risk of not being paid with a probability p risk ; that is, if the non-liquid asset defaulted during the W periods, we could not obtain any rewards by having the asset. An investor can change the portfolio by investing a fixed fraction w of the total capital M in the non-liquid asset at each time step. A typical investment strategy is to construct a portfolio using both liquid and non-liquid assets for decreasing the variance. In Figure, we plot performances under several hyperparameters, where the horizontal axis denotes the Var, and the vertical axis denotes the CR. Trained agents with a higher CR and lower Var are Pareto efficient. As the result shows, the EQUMRL returns more efficient portfolios than the others in almost all cases. We conjecture that this is because while the EQUMRL is an end-to-end optimization for obtaining an efficient agent, the other methods consist of several steps for solving the constrained optimization, where those multiple steps can be sources of the suboptimal result. We show CRs and Vars of some of their results in the upper, where we can confirm that the EQUMRL succeeded in minimizing the MSEs.
In this section, we introduce how to train a policy with the EQURL. We defined the objective function of the EQUMRL, and EQUMRL is an agnostic in learning method. As examples, we show an implementation based on REINFORCE and Actor-Critic (AC) methods. We use unbiased estimators of the gradients defined in (1).
For an episode k with the length n, the proposed algorithm replaces E π θ G and E π θ G 2 with the sample approximations n t=1 γ t−1 r(S t , A t ) and n t=1 γ t−1 r(S t , A t ) 2 , respectively; that is, the unbiased gradients are given as
(1). Therefore, for a sample approximation G k of E π θ G 2 at the episode k, we optimize the policy with ascending the unbiased gradient
CWe summarize the algorithm as the pseudo-code in Algorithm 1.
Here, we present three advantages of EQUMRL.The first advantage concerns computation. In EQUMRL, we do not suffer from the double sampling issue because the term
The third advantage is the ease of theoretical analysis. For example, referring to the results of, we derive the following result on the convergence of the gradient, which can be applied to simple policy gradient algorithms, not only our proposed REINFORCE-based algorithm. Theorem 1. Consider an update rule such that θ k+1 ← θ k + η i ∇E π θ k u trajectory (G; α, β) , where the learning rates η k are non-negative and satisfy
Suppose that (a) episode always finishes in finite horizon n; (b) the policy π θ has always bounded first and second partial derivatives. Then,
It is expected that non-asymptotic results can be derived by restricting the policy class and the optimization algorithm as and, but this is not the scope of this paper, which aims to provide a general framework.
AC-based trajectory EQUMRL. Another implementation of the EQUMRL is to apply the AC algorithm. For an episode k with the length n, following, we train the policy by a gradient defined as
ω()
ω()
ω()
For more details, see. Remark 1 (Existing approaches). For the double sampling issue, and proposed multi-time-scale stochastic optimization. Their approaches are known to be sensitive to the choice of step-size schedules, which are not easy to control. proposed using the Legendre-Fenchel dual transformation with coordinate descent algorithm. First, based on Lagrangian relaxation, set an objective function as transformed the objective function as
and trained an agent by solving the optimization problem via a coordinate descent algorithm. However, this approach does not reflect the constraint η because the constraint condition η vanishes from the objective function. This problem is caused by their objective function based on the penalty function g
, where the first derivative does not include η. To avoid this problem, we need an iterative algorithm to decide an optimal δ or change g(x) from x but it is not obvious how to incorporate them into the approach. Remark 2 (Difference from existing MV approaches). Readers may assert that EQUMRL simply omits
2 from existing MV-controlled RL methods, which usually includes the explicit variance term in the objective function, and is the essentially the same. However, there are significant differences; one of the main findings of this paper is our formulation of a simpler RL problem to obtain an MV-efficient policy. Existing MV-controlled RL methods suffer from computational difficulties caused by the double sampling issue. However, we can obtain MV-efficient policy without going through the difficult problem. In addition, EQUMRL shows better performance in experiments even from the viewpoint of the constrained problem because it is difficult to choose parameters to avoid the double sampling issue in existing approaches. Thus, the EQUMRL has advantage in avoiding solving more difficult constrained problems for considering MV trade-off.
In this section, we introduce how to train a policy with the EQURL. We defined the objective function of the EQUMRL, and EQUMRL is an agnostic in learning method. As examples, we show an implementation based on REINFORCE and Actor-Critic (AC) methods. We use unbiased estimators of the gradients defined in (1).
For an episode k with the length n, the proposed algorithm replaces E π θ G and E π θ G 2 with the sample approximations n t=1 γ t−1 r(S t , A t ) and n t=1 γ t−1 r(S t , A t ) 2 , respectively; that is, the unbiased gradients are given as
(1). Therefore, for a sample approximation G k of E π θ G 2 at the episode k, we optimize the policy with ascending the unbiased gradient
CWe summarize the algorithm as the pseudo-code in Algorithm 1.
Here, we present three advantages of EQUMRL.The first advantage concerns computation. In EQUMRL, we do not suffer from the double sampling issue because the term
The third advantage is the ease of theoretical analysis. For example, referring to the results of, we derive the following result on the convergence of the gradient, which can be applied to simple policy gradient algorithms, not only our proposed REINFORCE-based algorithm. Theorem 1. Consider an update rule such that θ k+1 ← θ k + η i ∇E π θ k u trajectory (G; α, β) , where the learning rates η k are non-negative and satisfy
Suppose that (a) episode always finishes in finite horizon n; (b) the policy π θ has always bounded first and second partial derivatives. Then,
It is expected that non-asymptotic results can be derived by restricting the policy class and the optimization algorithm as and, but this is not the scope of this paper, which aims to provide a general framework.
AC-based trajectory EQUMRL. Another implementation of the EQUMRL is to apply the AC algorithm. For an episode k with the length n, following, we train the policy by a gradient defined as
ω()
ω()
ω()
For more details, see. Remark 1 (Existing approaches). For the double sampling issue, and proposed multi-time-scale stochastic optimization. Their approaches are known to be sensitive to the choice of step-size schedules, which are not easy to control. proposed using the Legendre-Fenchel dual transformation with coordinate descent algorithm. First, based on Lagrangian relaxation, set an objective function as transformed the objective function as
and trained an agent by solving the optimization problem via a coordinate descent algorithm. However, this approach does not reflect the constraint η because the constraint condition η vanishes from the objective function. This problem is caused by their objective function based on the penalty function g
, where the first derivative does not include η. To avoid this problem, we need an iterative algorithm to decide an optimal δ or change g(x) from x but it is not obvious how to incorporate them into the approach. Remark 2 (Difference from existing MV approaches). Readers may assert that EQUMRL simply omits
2 from existing MV-controlled RL methods, which usually includes the explicit variance term in the objective function, and is the essentially the same. However, there are significant differences; one of the main findings of this paper is our formulation of a simpler RL problem to obtain an MV-efficient policy. Existing MV-controlled RL methods suffer from computational difficulties caused by the double sampling issue. However, we can obtain MV-efficient policy without going through the difficult problem. In addition, EQUMRL shows better performance in experiments even from the viewpoint of the constrained problem because it is difficult to choose parameters to avoid the double sampling issue in existing approaches. Thus, the EQUMRL has advantage in avoiding solving more difficult constrained problems for considering MV trade-off.
In this paper, we aim to obtain MV-efficient policies that achieve Pareto efficiency regarding MV trade-off. To achieve this purpose, we train an agent to maximize the expected quadratic utility function, in which the maximizer corresponds to the Pareto efficient policy.****We conjecture that this is because while the EQUMRL is an end-to-end optimization for obtaining an efficient agent, the other methods consist of several steps for solving the constrained optimization, where those multiple steps can be sources of the suboptimal result.****For the double sampling issue, Tamar et al. (2012) and Prashanth & Ghavamzadeh (2013; 2016) proposed multi-time-scale stochastic optimization. Their approaches are known to be sensitive to the choice of step-size schedules, which are not easy to control (Xie et al., 2018).****MV-controlled RL methods suffer from computational difficulties caused by the double sampling issue. However, we can obtain MV-efficient policy without going through the difficult problem
|
Why does the proposed algorithm work better than the Tamar et al. and Xie et al.?
|
Xie et al. formulates the MVRL with a constrained optimization problem, but the method cannot consider the constraint because the constraint is not used in optimization. As a result, the objective is the same as ours, the expected quadratic utility function. However, their method includes complicated steps, which may not be necessary, as we pointed out. Tamar et al. applies computationally difficult steps in optimization, multi-time-scale stochastic optimization, which requires control of learning parameters. This difficulty of this optimization is well-known in RL, as pointed out by Xie et al. While we point out the computational difficulties of the method of Tamar et al., which is also mentioned in the other existing studies, we avoid explicitly mentioning the problem in Xie et al., as that would be a discussion of the correctness of that paper.
|
null | false
| null |
Classify the below based on the products they sell.
Adidas, nike, Pepsi, Sprite, Fanta, FILA
|
Adidas, nike, FILA - Shoe
Pepsi, Sprite, Fanta - Beverages
|
|
null | false
| 84
|
Traditional language models are estimated to maximize the likelihood of each word $x_i$ given the words that precede it, $p(x_i \mid x_{<i})$. Given a corpus that is annotated with shallow syntax, we propose to condition on both the preceding words and their annotations.
We associate with each word $x_i$ three additional variables (denoted $c_i$): the indices of the beginning and end of the last completed chunk before $x_i$, and its label. For example, in Fig. FIGREF8, $c_4=\langle 3, 3, \text{VP}\rangle $ for $x_4=\text{the}$. Chunks, $c$ are only used as conditioning context via $p(x_i \mid x_{<i}, c_{\leqslant i})$; they are not predicted. Because the $c$ labels depend on the entire sentence through the CRF chunker, conditioning each word's probability on any $c_i$ means that our model is, strictly speaking, not a language model, and it can no longer be meaningfully evaluated using perplexity.
A right-to-left model is constructed analogously, conditioning on $c_{\geqslant i}$ alongside $x_{>i}$. Following BIBREF2, we use a joint objective maximizing data likelihood objectives in both directions, with shared softmax parameters.
To reduce cost, we initialize our sequential CWRs h, using pretrained embeddings from ELMo-transformer.
|
How do authors initialize their sequential CWRs h?
|
Using pretrained embeddings from ELMo-transformer.
|
null | false
| 260
|
Affect is a term that subsumes emotion and longer term constructs such as mood and personality and refers to the experience of feeling or emotion BIBREF0 . BIBREF1 picard1997affective provides a detailed discussion of the importance of affect analysis in human communication and interaction. Within this context the analysis of human affect from text is an important topic in natural language understanding, examples of which include sentiment analysis from Twitter BIBREF2 , affect analysis from poetry BIBREF3 and studies of correlation between function words and social/psychological processes BIBREF4 . People exchange verbal messages which not only contain syntactic information, but also information conveying their mental and emotional states. Examples include the use of emotionally colored words (such as furious and joy) and swear words. The automated processing of affect in human verbal communication is of great importance to understanding spoken language systems, particularly for emerging applications such as dialogue systems and conversational agents.
Statistical language modeling is an integral component of speech recognition systems, with other applications such as machine translation and information retrieval. There has been a resurgence of research effort in recurrent neural networks for language modeling BIBREF5 , which have yielded performances far superior to baseline language models based on n-gram approaches. However, there has not been much effort in building neural language models of text that leverage affective information. Current literature on deep learning for language understanding focuses mainly on representations based on word semantics BIBREF6 , encoder-decoder models for sentence representations BIBREF7 , language modeling integrated with symbolic knowledge BIBREF8 and neural caption generation BIBREF9 , but to the best of our knowledge there has been no work on augmenting neural language modeling with affective information, or on data-driven approaches to generate emotional text.
Motivated by these advances in neural language modeling and affective analysis of text, in this paper we propose a model for representation and generation of emotional text, which we call the Affect-LM. Our model is trained on conversational speech corpora, common in language modeling for speech recognition applications BIBREF10 . Figure 1 provides an overview of our Affect-LM and its ability to generate emotionally colored conversational text in a number of affect categories with varying affect strengths. While these parameters can be manually tuned to generate conversational text, the affect category can also be automatically inferred from preceding context words. Specifically for model training, the affect category is derived from features generated using keyword spotting from a dictionary of emotional words, such as the LIWC (Linguistic Inquiry and Word Count) tool BIBREF11 . Our primary research questions in this paper are:
Q1:Can Affect-LM be used to generate affective sentences for a target emotion with varying degrees of affect strength through a customizable model parameter?
Q2:Are these generated sentences rated as emotionally expressive as well as grammatically correct in an extensive crowd-sourced perception experiment?
Q3:Does the automatic inference of affect category from the context words improve language modeling performance of the proposed Affect-LM over the baseline as measured by perplexity?
The remainder of this paper is organized as follows. In Section "Related Work" , we discuss prior work in the fields of neural language modeling, and generation of affective conversational text. In Section "LSTM Language Model" we describe the baseline LSTM model and our proposed Affect-LM model. Section "Experimental Setup" details the experimental setup, and in Section "Results" , we discuss results for customizable emotional text generation, perception studies for each affect category, and perplexity improvements over the baseline model before concluding the paper in Section "Conclusions and Future Work" .
Within this context the analysis of human affect from text is an important topic in natural language understanding, examples of which include sentiment analysis from Twitter, affect analysis from poetry and studies of correlation between function words and social/psychological processes.
|
Is the analysis of human affect from text an important topic in natural language understanding?
|
Yes, it is.
|
null | false
| null |
I am trying to book a flight from Singapore to Sydney, what shall I do if the flight is too expensive?
|
You will have the option to choose from local Asian low-cost airlines such as Scoot, Jetstar, or AirAsia which would provide cheaper flights options.
|
|
null | false
| null |
How to add two cells in Excel?
|
You can use the =cell1 + cell2 operation where cell1 is the first cell and cell2 is the 2nd cell
|
|
null | false
| 252
|
Although the sifted multi-task learning method outperforms previous state-of-the-art methods on two datasets (From Table TABREF27), we observe that the proposed method achieves more remarkable performance boosts on PHEME than on RumourEval. There are two reasons for our analysis according to Table TABREF24 and Table TABREF27. One is that the number of training examples in RumourEval (including 5,568 tweets) is relatively limited as compared with PHEME (including 105,354 tweets), which is not enough to train deep neural networks. Another is that PHEME includes more threads (6,425 threads) than RumourEval (325 threads) so that PHEME can offer more rich credibility features to our proposed method.
Although the sifted multi-task learning method outperforms previous state-of-the-art methods on two datasets (From Table 2), we observe that the proposed method achieves more remarkable performance boosts on PHEME than on RumourEval. There are two reasons for our analysis according to Table 1 and Table 2. One is that the number of training examples in RumourEval (including 5,568 tweets) is relatively limited as compared with PHEME (including 105,354 tweets), which is not enough to train deep neural networks. Another is that PHEME includes more threads (6,425 threads) than RumourEval (325 threads) so that PHEME can offer more rich credibility features to our proposed method.
|
Through their analysis, what are the reasons that their method achieves more remarkable performance boosts on PHEME than on RumourEval?
|
The number of training examples in RumourEval is relatively limited as compared with PHEME , which is not enough to train deep neural networks, and PHEME includes more threads than RumourEval so that PHEME can offer more rich credibility features to their method.
|
null | false
| null |
Nordic countries are the highest coffee-consuming nations when measured per capita, with consumption in Finland as the world's highest.
Finland – 26.45 lb (12.00 kg)
Norway – 21.82 lb (9.90 kg)
Iceland – 19.84 lb (9.00 kg)
Denmark – 19.18 lb (8.70 kg)
Netherlands – 18.52 lb (8.40 kg)
Sweden – 18.00 lb (8.16 kg)
Switzerland – 17.42 lb (7.90 kg)
Belgium – 15.00 lb (6.80 kg)
Luxembourg – 14.33 lb (6.50 kg)
Canada – 14.33 lb (6.50 kg)
|
From the passage provided, extract the per capita coffee consumption in kg for Belgium.
|
Belgium consumes 6.80 kg of coffee per capita.
|
null | false
| null |
Classify the following by how they are affected by a strong permanent magnet as one of "attract", "repel", "nothing": iron nail, paper clips, a tin can, a compass, a plastic cup, a glass window, a leather wallet, a brick, a magnet of opposite polarity, pyrolytic carbon sheet.
|
Items that are strongly or weakly attracted are called "ferromagnetic" and "paramagnetic", respectively. These include: iron nail, paper clips, a tin can, a compass.
Items that are repelled by a magnet are either "paramagnetic" or themselves permanent magnets of the opposite polarity, including: a magnet of opposite polarity, pyrolytic carbon sheet. Items that are not affected by a permanent magnet are called "nonmagnetic" and include: a plastic cup, a glass window, a leather wallet, a brick.
|
|
null | false
| null |
Why do many people spend so much time on their smartphone everyday?
|
There are lots of reasons and one of them is that lot of people are now using smartphones for both work and entertainment. Applications have been developed to help improve work efficiency as well as providing easy access to all kinds of entertaining content.
|
|
null | false
| 135
|
For all experiments, we admitted only experienced MTurk workers who are based in the US. One HIT consisted of writing one text for the text collection, formulating 15 questions for the question collection, or finding 15 pairs of answers for the answer collection. We paid $0.50 per HIT for the text and question collection, and $0.60 per HIT for the answer collection.
More than 2,100 texts were paired with 15 questions each, resulting in a total number of approx. 32,000 annotated questions. For 13% of the questions, the workers did not agree on one of the 4 categories with a 3 out of 5 majority, so we did not include these questions in our dataset.
The distribution of category labels on the remaining 87% is shown in Table TABREF10 . 14,074 (52%) questions could be answered. Out of the answerable questions, 10,160 could be answered from the text directly (text-based) and 3,914 questions required the use of commonsense knowledge (script-based). After removing 135 questions during the validation, the final dataset comprises 13,939 questions, 3,827 of which require commonsense knowledge (i.e. 27.4%). This ratio was manually verified based on a random sample of questions.
We split the dataset into training (9,731 questions on 1,470 texts), development (1,411 questions on 219 texts), and test set (2,797 questions on 430 texts). Each text appears only in one of the three sets. The complete set of texts for 5 scenarios was held out for the test set. The average text, question, and answer length is 196.0 words, 7.8 words, and 3.6 words, respectively. On average, there are 6.7 questions per text.
Figure FIGREF21 shows the distribution of question types in the dataset, which we identified using simple heuristics based on the first words of a question: Yes/no questions were identified as questions starting with an auxiliary or modal verb, all other question types were determined based on the question word.
We found that 29% of all questions are yes/no questions. Questions about details of a situation (such as what/ which and who) form the second most frequent question category. Temporal questions (when and how long/often) form approx. 11% of all questions. We leave a more detailed analysis of question types for future work.
The average text, question, and answer length is 196.0 words, 7.8 words, and 3.6 words, respectively.
|
How long are the questions on average?
|
7.8 words.
|
null | false
| null |
Who went to space first?
|
Yuri Gagarin was the first human to go into space. However, animals were sent before that to test out.
|
|
null | false
| null |
The Act to provide for reconciliation pursuant to titles II and V of the concurrent resolution on the budget for fiscal year 2018, Pub. L. 115–97 (text) (PDF), is a congressional revenue act of the United States originally introduced in Congress as the Tax Cuts and Jobs Act (TCJA), that amended the Internal Revenue Code of 1986. Major elements of the changes include reducing tax rates for businesses and individuals, increasing the standard deduction and family tax credits, eliminating personal exemptions and making it less beneficial to itemize deductions, limiting deductions for state and local income taxes and property taxes, further limiting the mortgage interest deduction, reducing the alternative minimum tax for individuals and eliminating it for corporations, doubling the estate tax exemption, and set the penalty enforcing individual mandate of the Affordable Care Act (ACA) at $0.
The Act is based on tax reform advocated by congressional Republicans and the Trump administration. The nonpartisan Congressional Budget Office (CBO) reported that under the Act individuals and pass-through entities like partnerships and S corporations would receive about $1.125 trillion in net benefits (i.e. net tax cuts offset by reduced healthcare subsidies) over 10 years, while corporations would receive around $320 billion in benefits. The CBO estimated that implementing the Act would add an estimated $2.289 trillion to the national debt over ten years, or about $1.891 trillion after taking into account macroeconomic feedback effects, in addition to the $9.8 trillion increase forecast under the current policy baseline and existing $20 trillion national debt.[8
|
Given a reference text about the Tax Cuts and Jobs Act, provide a bulleted list of the main elements of the act.
|
The main elements of the Tax Cuts and Jobs Act are:
- Reducing tax rates for businesses and individuals
- Increasing the standard deduction and family tax credits
- Eliminating personal exemptions
- Limiting deductions for state and local income and property taxes
- Limiting the mortgage interest deduction
- Reducing the alternative minimum tax
- Doubling the estate tax exemption
- Setting the penalty enforcing individual mandate of the Affordable Care Act at $0
|
null | false
| 148
|
To recognize entailment between two questions INLINEFORM0 (premise) and INLINEFORM1 (hypothesis), we adapted the neural network proposed by Bowman et al. BIBREF13 . Our DL model, presented in Figure FIGREF20 , consists of three 600d ReLU layers, with a bottom layer taking the concatenated sentence representations as input and a top layer feeding a softmax classifier. The sentence embedding model sums the Recurrent neural network (RNN) embeddings of its words. The word embeddings are first initialized with pretrained GloVe vectors. This adaptation provided the best performance in previous experiments with RQE data.
GloVe is an unsupervised learning algorithm to generate vector representations for words BIBREF42 . Training is performed on aggregated word co-occurrence statistics from a large corpus, and the resulting representations show interesting linear substructures of the word vector space. We use the pretrained common crawl version with 840B tokens and 300d vectors, which are not updated during training.
Our DL model, presented in Figure 1, consists of three 600d ReLU layers, with a bottom layer taking the concatenated sentence representations as input and a top layer feeding a softmax classifier. The sentence embedding model sums the Recurrent neural network (RNN) embeddings of its words. The word embeddings are first initialized with pretrained GloVe vectors.
|
What model do the authors use in this paper?
|
The DL model consists of three 600d ReLU layers, with a bottom layer taking the concatenated sentence representations as input and a top layer feeding a softmax classifier. The sentence embedding model sums the Recurrent neural network (RNN) embeddings of its words. The word embeddings are first initialized with pretrained GloVe vectors.
|
null | false
| null |
What are some ways you can stop your toddler from crying?
|
Give your toddler a hug to comfort him/her; try to reconcile the fight between the siblings if that is the source of conflict; if all fails - cry louder than the toddler
|
|
null | false
| null |
No, They Can't: Why Government Fails – But Individuals Succeed is a 2012 book by John Stossel, the American consumer reporter, investigative journalist, author and libertarian columnist. It was published on April 10, 2012, and focuses on what Stossel sees as the failures of government intervention.
|
When was the "No, They Can't" book released?
|
The book "No, They Can't" was published on April 10, 2012.
|
null | false
| 247
|
To gain better insights into our proposed ConvMS-Memnet, we conduct further experiments to understand the impact on performance by using: 1) pre-trained or randomly initialized word embedding; 2) multiple hops; 3) attention visualizations; 4) more training epochs.
In our ConvMS-Memnet, we use pre-trained word embedding as the input. The embedding maps each word into a lower dimensional real-value vector as its representation. Words sharing similar meanings should have similar representations. It enables our model to deal with synonyms more effectively. The question is, “can we train the network without using pre-trained word embeddings?". We initialize word vectors randomly, and use an embedding matrix to update the word vectors in the training of the network simultaneously. Comparison results are shown in Table 3. It can be observed that pre-trained word embedding gives 2.59% higher F-measure compared to random initialization. This is partly due to the limited size of our training data. Hence using word embedding trained from other much larger corpus gives better results.
It is widely acknowledged that computational models using deep architecture with multiple layers have better ability to learn data representations with multiple levels of abstractions. In this section, we evaluate the power of multiple hops in this task. We set the number of hops from 1 to 9 with 1 standing for the simplest single layer network shown in Figure 4. The more hops are stacked, the more complicated the model is. Results are shown in Table 4. The single layer network has achieved a competitive performance. With the increasing number of hops, the performance improves. However, when the number of hops is larger than 3, the performance decreases due to overfitting. Since the dataset for this task is small, more parameters will lead to overfitting. As such, we choose 3 hops in our final model since it gives the best performance in our experiments.
Essentially, memory network aims to measure the weight of each word in the clause with respect to the emotion word. The question is, will the model really focus on the words which describe the emotion cause? We choose one example to show the attention results in Table 5:
Ex.2 家人/family 的/'s 坚持/insistence 更/more 让/makes 人/people 感动/touched
In this example, the cause of the emotion “touched” is “insistence”. We show in Table 5 the distribution of word-level attention weights in different hops of memory network training. We can observe that in the first two hops, the highest attention weights centered on the word “more". However, from the third hop onwards, the highest attention weight moves to the word sub-sequence centred on the word “insistence”. This shows that our model is effective in identifying the most important keyword relating to the emotion cause. Also, better results are obtained using deep memory network trained with at least 3 hops. This is consistent with what we observed in Section UID45 .
In order to evaluate the quality of keywords extracted by memory networks, we define a new metric on the keyword level of emotion cause extraction. The keyword is defined as the word which obtains the highest attention weight in the identified clause. If the keywords extracted by our algorithm is located within the boundary of annotation, it is treated as correct. Thus, we can obtain the precision, recall, and F-measure by comparing the proposed keywords with the correct keywords by: INLINEFORM0
Since the reference methods do not focus on the keywords level, we only compare the performance of Memnet and ConvMS-Memnet in Table 6. It can be observed that our proposed ConvMS-Memnet outperforms Memnet by 5.6% in F-measure. It shows that by capturing context features, ConvMS-Memnet is able to identify the word level emotion cause better compare to Memnet.
In our model, the training epochs are set to 20. In this section, we examine the testing error using a case study. Due to the page length limit, we only choose one example from the corpus. The text below has four clauses:
Ex.3 45天，对于失去儿子的他们是多么的漫长，宝贝回家了，这个春节是多么幸福。
Ex.3 45 days, it is long time for the parents who lost their baby. If the baby comes back home, they would become so happy in this Spring Festival.
In this example, the cause of emotion “happy” is described in the third clause.
We show in Table 7 the probability of each clause containing an emotion cause in different training epochs. It is interesting to see that our model is able to detect the correct clause with only 5 epochs. With the increasing number of training epochs, the probability associated with the correct clause increases further while the probabilities of incorrect clauses decrease generally.
To gain better insights into our proposed ConvMSMemnet, we conduct further experiments to understand the impact on performance by using: 1) pre-trained or randomly initialized word embedding; 2) multiple hops; 3) attention visualizations; 4) more training epochs.
|
How to conduct further experiments to understand the impact on performance?
|
1) pre-trained or randomly initialized word embedding; 2) multiple hops; 3) attention visualizations; 4) more training epochs.
|
null | false
| 303
|
Online social networks have become one of the most disruptive communication platforms, as everyday billions of individuals use them to interact with each other. Their penetration in our everyday lives seems ever-growing and has in turn generated a massive volume of publicly available data open to analysis. The digital footprints left across these multiple media platforms provide us with a unique source to study and understand how the linguistic phenotype of a given user is related to social attributes such as socioeconomic status (SES).
The quantification and inference of SES of individuals is a long lasting question in the social sciences. It is a rather difficult problem as it may depend on a combination of individual characteristics and environmental variables BIBREF0 . Some of these features can be easier to assess like income, gender, or age whereas others, relying to some degree on self-definition and sometimes entangled with privacy issues, are harder to assign like ethnicity, occupation, education level or home location. Furthermore, individual SES correlates with other individual or network attributes, as users tend to build social links with others of similar SES, a phenomenon known as status homophily BIBREF1 , arguably driving the observed stratification of society BIBREF2 . At the same time, shared social environment, similar education level, and social influence have been shown to jointly lead socioeconomic groups to exhibit stereotypical behavioral patterns, such as shared political opinion BIBREF3 or similar linguistic patterns BIBREF4 . Although these features are entangled and causal relation between them is far from understood, they appear as correlations in the data.
Datasets recording multiple characteristics of human behaviour are more and more available due to recent developments in data collection technologies and increasingly popular online platforms and personal digital devices. The automatic tracking of online activities, commonly associated with profile data and meta-information; the precise recording of daily activities, interaction dynamics and mobility patterns collected through mobile personal devices; together with the detailed and expert annotated census data all provide new grounds for the inference of individual features or behavioral patterns BIBREF5 . The exploitation of these data sources has already been proven to be fruitful as cutting edge recommendation systems, advanced methods for health record analysis, or successful prediction tools for social behaviour heavily rely on them BIBREF6 . Nevertheless, despite the available data, some inference tasks, like individual SES prediction, remain an open challenge.
The precise inference of SES would contribute to overcome several scientific challenges and could potentially have several commercial applications BIBREF7 . Further, robust SES inference would provide unique opportunities to gain deeper insights on socioeconomic inequalities BIBREF8 , social stratification BIBREF2 , and on the driving mechanisms of network evolution, such as status homophily or social segregation.
In this work, we take a horizontal approach to this problem and explore various ways to infer the SES of a large sample of social media users. We propose different data collection and combination strategies using open, crawlable, or expert annotated socioeconomic data for the prediction task. Specifically, we use an extensive Twitter dataset of 1.3M users located in France, all associated with their tweets and profile information; 32,053 of them having inferred home locations. Individual SES is estimated by relying on three separate datasets, namely socioeconomic census data; crawled profession information and expert annotated Google Street View images of users' home locations. Each of these datasets is then used as ground-truth to infer the SES of Twitter users from profile and semantic features similar to BIBREF9 . We aim to explore and assess how the SES of social media users can be obtained and how much the inference problem depends on annotation and the user's individual and linguistic attributes.
We provide in Section SECREF2 an overview of the related literature to contextualize the novelty of our work. In Section SECREF3 we provide a detailed description of the data collection and combination methods. In Section SECREF4 we introduce the features extracted to solve the SES inference problem, with results summarized in Section SECREF5 . Finally, in Section SECREF6 and SECREF7 we conclude our paper with a brief discussion of the limitations and perspectives of our methods.
Specifically, we use an extensive Twitter dataset of 1.3M users located in France, all associated with their tweets and profile information; 32,053 of them having inferred home locations.
|
How many of the users have inferred home locations?
|
32053.
|
null | false
| null |
What came first the chicken or the egg?
|
At first glance, it may seem that a chicken must have come from an egg first. "Egg" is a word, that is used to describe the resulting product of procreation for many species. Based on the theory of evolution a non chicken egg at some point led to the birth of a chicken. It is important to note that an egg does not instantly turn into a chicken and there is a vast and potentially infinite spectrum of transitionary states. Somewhere in the infinite spectrum of this evolutionary transition a chicken and egg existed for a moment. One can argue that a chicken and its egg were created exactly at the same instant in time. We could further argue that the chicken and egg of yesterday is not the same chicken and egg of today.
|
|
1911.09845
| false
| null |
We find that BIBREF27 zhao2018unsupervised make use of a set of discrete variables that define high-level attributes of a response. Although they interpret meanings of the learned discrete latent variables by clustering data according to certain classes (e.g. dialog acts), such latent variables still have no exact meanings. In our model, we connect each latent variable with a word in the vocabulary, thus each latent variable has an exact semantic meaning. Besides, they focus on multi-turn dialogue generation and presented an unsupervised discrete sentence representation learning method learned from the context while our concentration is primarily on single-turn dialogue generation with no context information.
We find that BIBREF27 zhao2018unsupervised make use of a set of discrete variables that define high-level attributes of a response. Although they interpret meanings of the learned discrete latent variables by clustering data according to certain classes (e.g. dialog acts), such latent variables still have no exact meanings. In our model, we connect each latent variable with a word in the vocabulary, thus each latent variable has an exact semantic meaning. Besides, they focus on multi-turn dialogue generation and presented an unsupervised discrete sentence representation learning method learned from the context while our concentration is primarily on single-turn dialogue generation with no context information.
|
How does discrete latent variable has an explicit semantic meaning to improve the CVAE on short-text conversation?
|
The answers are shown as follows:
* we connect each latent variable with a word in the vocabulary, thus each latent variable has an exact semantic meaning.
|
null | false
| 244
|
In this paper, we are exploring the historical significance of Croatian machine translation research group. The group was active in 1950s, and it was conducted by Bulcsu Laszlo, Croatian linguist, who was a pioneer in machine translation during the 1950s in Yugoslavia.
To put the research of the Croatian group in the right context, we have to explore the origin of the idea of machine translation. The idea of machine translation is an old one, and its origin is commonly connected with the work of Rene Descartes, i.e. to his idea of universal language, as described in his letter to Mersenne from 20.xi.1629 BIBREF0. Descartes describes universal language as a simplified version of the language which will serve as an “interlanguage” for translation. That is, if we want to translate from English to Croatian, we will firstly translate from English to an “interlanguage”, and then from the “interlanguage” to Croatian. As described later in this paper, this idea had been implemented in the machine translation process, firstly in the Indonesian-to-Russian machine translation system created by Andreev, Kulagina and Melchuk from the early 1960s.
In modern times, the idea of machine translation was put forth by the philosopher and logician Yehoshua Bar-Hillel (most notably in BIBREF1 and BIBREF2), whose papers were studied by the Croatian group. Perhaps the most important unrealized point of contact between machine translation and cybernetics happened in the winter of 1950/51. In that period, Bar-Hillel met Rudolf Carnap in Chicago, who introduced to him the (new) idea of cybernetics. Also, Carnap gave him the contact details of his former teaching assistant, Walter Pitts, who was at that moment with Norbert Wiener at MIT and who was supposed to introduce him to Wiener, but the meeting never took place BIBREF3. Nevertheless, Bar-Hillel was to stay at MIT where he, inspired by cybernetics, would go to organize the first machine translation conference in the world in 1952 BIBREF3.
The idea of machine translation was a tempting idea in the 1950s. The main military interest in machine translation as an intelligence gathering tool (translation of scientific papers, daily press, technical reports, and everything the intelligence services could get their hands on) was sparked by the Soviet advance in nuclear technology, and would later be compounded by the success of Vostok 1 (termed by the USA as a “strategic surprise”). In the nuclear age, being able to read and understand what the other side was working on was of crucial importance BIBREF4. Machine translation was quickly absorbed in the program of the Dartmouth Summer Research Project on Artificial Intelligence in 1956 (where Artificial Intelligence as a field was born), as one of the five core fields of artificial intelligence (later to be known as natural language processing). One other field was included here, the “nerve nets” as they were known back then, today commonly known as artificial neural networks. What is also essential for our discussion is that the earliest programming language for artificial intelligence, Lisp, was invented in 1958 by John McCarthy BIBREF5. But let us take a closer look at the history of machine translation. In the USA, the first major wave of government and military funding for machine translation came in 1954, and the period of abundancy lasted until 1964, when the National Research Council established the Automatic Language Processing Advisory Committee (ALPAC), which was to assess the results of the ten years of intense funding. The findings were very negative, and funding was almost gone BIBREF4, hence the ALPAC report became the catalyst for the first “AI Winter”.
One of the first recorded attempts of producing a machine translation system in the USSR was in 1954 BIBREF6, and the attempt was applauded by the Communist party of the Soviet Union, by the USSR Committee for Science and Technology and the USSR Academy of Sciences. The source does not specify how this first system worked, but it does delineate that the major figures of machine translation of the time were N. Andreev of the Leningrad State University, O. Kulagina and I. Melchuk of the Steklov Mathematical Institute. There is information on an Indonesian-to-Russian machine translation system by Andreev, Kulagina and Melchuk from the early 1960s, but it is reported that the system was ultimately a failure, in the same way early USA systems were. The system had statistical elements set forth by Andreev, but the bulk was logical and knowledge-heavy processing put forth by Kulagina and Melchuk. The idea was to have a logical intermediate language, under the working name “Interlingua”, which was the connector of both natural languages, and was used to model common-sense human knowledge. For more details, see BIBREF6.
In the USSR, there were four major approaches to machine translation in the late 1950s BIBREF7. The first one was the research at the Institute for Precise Mechanics and Computational Technology of the USSR Academy of Sciences. Their approach was mostly experimental and not much different from today's empirical methods. They evaluated the majority of algorithms known at the time algorithms over meticulously prepared datasets, whose main strength was data cleaning, and by 1959 they have built a German-Russian machine translation prototype. The second approach, as noted by Mulić BIBREF7, was championed by the team at the Steklov Mathematical Institute of the USSR Academy of Sciences led by A. A. Reformatsky. Their approach was mainly logical, and they extended the theoretical ideas of Bar-Hillel BIBREF2 to build three algorithms: French-Russian, English-Russian and Hungarian-Russian. The third and perhaps the most successful approach was the one by A. A. Lyapunov, O. S. Kulagina and R. L. Dobrushin. Their efforts resulted in the formation of the Mathematical Linguistics Seminar at the Faculty of Philology in Moscow in 1956 and in Leningrad in 1957. Their approach was mainly information-theoretic (but they also tried logic-based approaches BIBREF7), which was considered cybernetic at that time. This was the main role model for the Croatian efforts from 1957 onwards. The fourth, and perhaps most influential, was the approach at the Experimental Laboratory of the Leningrad University championed by N. D. Andreev BIBREF7. Here, the algorithms for Indonesian-Russian, Arabic-Russian, Hindu-Russian, Japanese-Russian, Burmese-Russian, Norwegian-Russian, English-Russian, Spanish-Russian and Turkish-Russian were being built. The main approach of Andreev's group was to use an intermediary language, which would capture the meanings BIBREF7. It was an approach similar to KL-ONE, which would be introduced in the West much later (in 1985) by Brachman and Schmolze BIBREF8. It is also interesting to note that the Andreev group had a profound influence on the Czechoslovakian machine translation program BIBREF9, which unfortunately suffered a similar fate as the Yugoslav one due to the lack of funding.
Andreev's approach was in a sense "external". The modelling would be statistical, but its purpose would not be to mimic the stochasticity of the human thought process, but rather to produce a working machine translation system. Kulagina and Melchuk disagreed with this approach as they thought that more of what is presently called "philosophical logic" was needed to model the human thought process at the symbolic level, and according to them, the formalization of the human thought process was a prerequisite for developing a machine translation system (cf. BIBREF6). We could speculate that sub-symbolic processing would have been acceptable too, since that approach is also rooted in philosophical logic as a way of formalizing human cognitive functions and is also "internal" in the same sense symbolic approaches are.
There were many other centers for research in machine translation: Gorkovsky University (Omsk), 1st Moscow Institute for Foreign Languages, Computing Centre of the Armenian SSR and at the Institute for Automatics and Telemechanics of the Georgian SSR BIBREF7. It is worthwhile to note that both the USA and the USSR had access to state-of-the-art computers, and the political support for the production of such systems meant that computers were made available to researchers in machine translation. However, the results were poor in the late 1950s, and a working system was yet to be shown. All work was therefore theoretical work implemented on a computer, which proved to be sub-optimal.
What is also essential for our discussion is that the earliest programming language for artificial intelligence, Lisp, was invented in 1958 by John McCarthy [17].
|
Who invented the earliest programming language for artificial intelligence?
|
The earliest programming language for artificial intelligence, Lisp, was invented in 1958 by John McCarthy.
|
null | false
| null |
Over the years, experimentalists have constructed small-scale quantum computers using trapped ions and superconductors. In 1998, a two-qubit quantum computer demonstrated the feasibility of the technology, and subsequent experiments have increased the number of qubits and reduced error rates. In 2019, Google AI and NASA announced that they had achieved quantum supremacy with a 54-qubit machine, performing a computation that is impossible for any classical computer. However, the validity of this claim is still being actively researched.
|
Which year was quantum computer demonstrated to be possible?
|
1998
|
null | false
| null |
why peoples eating popcorn in the cinema?
|
People often eat popcorn in the cinema because it has become a popular snack food that is associated with movie-going. Popcorn is easy to eat with your hands and can be shared among a group, making it a convenient snack to enjoy while watching a film. In addition, the aroma of freshly popped popcorn is often enticing and can help to create a festive atmosphere in the cinema. Popcorn also has a long shelf life and can be stored in large quantities, making it an economical choice for cinemas. Overall, the tradition of eating popcorn in the cinema has become deeply ingrained in popular culture and is enjoyed by many movie-goers around the world.
|
|
null | false
| 249
|
The English and Dutch corpora were independently annotated for cyberbullying by trained linguists. All were Dutch native speakers and English second-language speakers. To demonstrate the validity of our guidelines, inter-annotator agreement scores were calculated using Kappa on a subset of each corpus. Inter-rater agreement for Dutch (2 raters) is calculated using Cohen's Kappa BIBREF53 . Fleiss' Kappa BIBREF54 is used for the English corpus ( INLINEFORM0 2 raters). Kappa scores for the identification of cyberbullying are INLINEFORM1 = 0.69 (Dutch) and INLINEFORM2 = 0.59 (English).
As shown in Table TABREF35 , inter-annotator agreement for the identification of the more fine-grained categories for English varies from fair to substantial BIBREF55 , except for defamation, which appears to be more difficult to recognise. No encouragements to the harasser were present in this subset of the corpus. For Dutch, the inter-annotator agreement is fair to substantial, except for curse and defamation. Analysis revealed that one of both annotators often annotated the latter as an insult, and in some cases even did not consider it as cyberbullying-related.
In short, the inter-rater reliability study shows that the annotation of cyberbullying is not trivial and that more fine-grained categories like defamation, curse and encouragements are sometimes hard to recognise. It appears that defamations were sometimes hard to distinguish from insults, whereas curses and exclusions were sometimes considered insults or threats. The analysis further reveals that encouragements to the harasser are subject to interpretation. Some are straightforward (e.g. `I agree we should send her hate'), whereas others are subject to the annotator's judgement and interpretation (e.g. `hahaha', `LOL').
To demonstrate the validity of our guidelines, inter-annotator agreement scores were calculated using Kappa on a subset of 9 each corpus.
|
What is used to demonstrate the validity?
|
Kappa is used to demonstrate validity.
|
null | false
| null |
How was the Grammy Award named?
|
The Grammy was named after the Gramophone, which was used to play music before records, tape, discs, or digital.
|
|
1904.04055
| false
| null |
KGR10, also known as plWordNet Corpus 10.0 (PLWNC 10.0), is the result of the work on the toolchain to automatic acquisition and extraction of the website content, called CorpoGrabber BIBREF19 . It is a pipeline of tools to get the most relevant content of the website, including all subsites (up to the user-defined depth). The proposed toolchain can be used to build a big Web corpus of text documents. It requires the list of the root websites as the input. Tools composing CorpoGrabber are adapted to Polish, but most subtasks are language independent. The whole process can be run in parallel on a single machine and includes the following tasks: download of the HTML subpages of each input page URL with HTTrack, extraction of plain text from each subpage by removing boilerplate content (such as navigation links, headers, footers, advertisements from HTML pages) BIBREF20 , deduplication of plain text BIBREF20 , bad quality documents removal utilising Morphological Analysis Converter and Aggregator (MACA) BIBREF21 , documents tagging using Wrocław CRF Tagger (WCRFT) BIBREF22 . Last two steps are available only for Polish.
KGR10, also known as plWordNet Corpus 10.0 (PLWNC 10.0), is the result of the work on the toolchain to automatic acquisition and extraction of the website content, called CorpoGrabber BIBREF19 . It is a pipeline of tools to get the most relevant content of the website, including all subsites (up to the user-defined depth). The proposed toolchain can be used to build a big Web corpus of text documents. It requires the list of the root websites as the input. Tools composing CorpoGrabber are adapted to Polish, but most subtasks are language independent. The whole process can be run in parallel on a single machine and includes the following tasks: download of the HTML subpages of each input page URL with HTTrack, extraction of plain text from each subpage by removing boilerplate content (such as navigation links, headers, footers, advertisements from HTML pages) BIBREF20 , deduplication of plain text BIBREF20 , bad quality documents removal utilising Morphological Analysis Converter and Aggregator (MACA) BIBREF21 , documents tagging using Wrocław CRF Tagger (WCRFT) BIBREF22 . Last two steps are available only for Polish.
|
How was the KGR10 corpus created?
|
The answers are shown as follows:
* most relevant content of the website, including all subsites
|
null | false
| null |
What is mitosis
|
A type of cell division that results in two daughter cells each having the same number and kind of chromosomes as the "parent" chromosome.
|
|
null | false
| null |
What are the top three cloud providers.
|
- AWS
- Microsoft Azure
- Google Cloud Platform (GCP)
|
|
null | false
| 305
|
Human learning occurs through interaction BIBREF0 and multimodal experience BIBREF1 , BIBREF2 . Prior work has argued that machine learning may also benefit from interactive, multimodal learning BIBREF3 , BIBREF4 , BIBREF5 , termed virtual embodiment BIBREF6 . Driven by breakthroughs in static, unimodal tasks such as image classification BIBREF7 and language processing BIBREF8 , machine learning has moved in this direction. Recent tasks such as visual question answering BIBREF9 , image captioning BIBREF10 , and audio-video classification BIBREF11 make steps towards learning from multiple modalities but lack the dynamic, responsive signal from exploratory learning. Modern, challenging tasks incorporating interaction, such as Atari BIBREF12 and Go BIBREF13 , push agents to learn complex strategies through trial-and-error but miss information-rich connections across vision, language, sounds, and actions. To remedy these shortcomings, subsequent work introduces tasks that are both multimodal and interactive, successfully training virtually embodied agents that, for example, ground language in actions and visual percepts in 3D worlds BIBREF3 , BIBREF4 , BIBREF14 .
For virtual embodiment to reach its full potential, though, agents should be immersed in a rich, lifelike context as humans are. Agents may then learn to ground concepts not only in various modalities but also in relationships to other concepts, i.e. that forks are often in kitchens, which are near living rooms, which contain sofas, etc. Humans learn by concept-to-concept association, as shown in child learning psychology BIBREF1 , BIBREF2 , cognitive science BIBREF15 , neuroscience BIBREF16 , and linguistics BIBREF17 . Even in machine learning, contextual information has given rise to effective word representations BIBREF8 , improvements in recommendation systems BIBREF18 , and increased reward quality in robotics BIBREF19 . Importantly, scale in data has proven key in algorithms learning from context BIBREF8 and in general BIBREF20 , BIBREF21 , BIBREF22 .
To this end, we present HoME: the Household Multimodal Environment (Figure 1 ). HoME is a large-scale platform for agents to navigate and interact within over 45,000 hand-designed houses from the SUNCG dataset BIBREF23 . Specifically, HoME provides:
HoME is a general platform extensible to many specific tasks, from reinforcement learning to language grounding to blind navigation, in a real-world context. HoME is also the first major interactive platform to support high fidelity audio, allowing researchers to better experiment across modalities and develop new tasks. While HoME is not the first platform to provide realistic context, we show in following sections that HoME provides a more large-scale and multimodal testbed than existing environments, making it more conducive to virtually embodied learning in many scenarios.
HoME is also the first major interactive platform to support high fidelity audio, allowing researchers to better experiment across modalities and develop new tasks.
|
Is the HoME the first major interactive platform to support high-fidelity audio?
|
Yes, it is.
|
null | false
| null |
Skateboarding is an action sport originating in the United States that involves riding and performing tricks using a skateboard, as well as a recreational activity, an art form, an entertainment industry job, and a method of transportation. Skateboarding has been shaped and influenced by many skateboarders throughout the years. A 2009 report found that the skateboarding market is worth an estimated $4.8 billion in annual revenue, with 11.08 million active skateboarders in the world.
|
Without quoting directly from the text give me a summary of what skateboarding is.
|
Skateboarding is an action sport originating in the United States that involves riding and performing tricks using a skateboard, as well as a recreational activity, an art form, an entertainment industry job, and a method of transportation. Skateboarding has been shaped and influenced by many skateboarders throughout the years. A 2009 report found that the skateboarding market is worth an estimated $4.8 billion in annual revenue, with 11.08 million active skateboarders in the world. In 2016, it was announced that skateboarding would be represented at the 2020 Summer Olympics in Tokyo, for both male and female teams.
Since the 1970s, skateparks have been constructed specifically for use by skateboarders, freestyle BMXers, aggressive skaters, and more recently, scooters. However, skateboarding has become controversial in areas in which the activity, although illegal, has damaged curbs, stoneworks, steps, benches, plazas, and parks.
|
1805.04570
| false
| null |
We used the Universal Dependencies Treebank UD v2.1 BIBREF0 for our experiments. We picked four low-resource/high-resource language pairs, each from a different family: Danish/Swedish (da/sv), Russian/Bulgarian (ru/bg), Finnish/Hungarian (fi/hu), Spanish/Portuguese (es/pt). Picking languages from different families would ensure that we obtain results that are on average consistent across languages.
We used the Universal Dependencies Treebank UD v2.1 BIBREF0 for our experiments. We picked four low-resource/high-resource language pairs, each from a different family: Danish/Swedish (da/sv), Russian/Bulgarian (ru/bg), Finnish/Hungarian (fi/hu), Spanish/Portuguese (es/pt). Picking languages from different families would ensure that we obtain results that are on average consistent across languages.
|
What languages are explored?
|
The answers are shown as follows:
* Danish/Swedish (da/sv), Russian/Bulgarian (ru/bg), Finnish/Hungarian (fi/hu), Spanish/Portuguese (es/pt)
|
null | false
| null |
What kind of disease is wheat mildew
|
Wheat mildew is a wheat disease that affects the ear, and is brought on by causes somewhat similar to those that cause blight, though at a more advanced period of the season. If this disorder comes on immediately after the first appearance of the ear, the straw is also affected—but if the grain is nearly or fully formed, injury to the straw is less discernible. Wheat may mildew where the straw is perfectly fresh, but only rarely. A severe mildew, however, retards further grain and straw maturation and leads to death. Something akin to mildew is the gum, which, in all warm moist seasons, attaches itself to the ear, and often causes considerable damage. All these different disorders are generally accompanied by insects, and by minute parasitic vegetable growths, considered by many to be the cause of the damage. Their appearance, however, may justly be attributed to the diseased state of the plant; for wherever putrefaction takes place, either in animal or vegetable substances, the presence of these parasites will never be wanting.
Another disorder that affects wheat is brought on by excessive heat, which makes the plants suffer from malnutrition, and become sickly and feeble. In this atrophic state a kind of dust gathers on the stalks and leaves, which increases with the disease, till the plant is in a great measure worn out and exhausted. The only remedy in this case, and it is one that cannot easily be administered by the hand of man, is a plentiful supply of moisture, by which, if it is received before consumption is too far advanced, the crop is benefited in a degree proportional to the extent of nourishment received, and the stage at which the disease has arrived.
Some people have recommended using blighted and mildewed wheat for seed. This, however may be hazardous. Light or defective wheat does vegetate and produce a plant, but the plant may be of poor quality and incapable of withstanding winter blasts.
|
|
null | false
| null |
Is it safe to smoke at a gas station?
|
No, it is unsafe to smoke at a gas station because there is a risk that fuel could be ignited and cause a fire.
|
|
null | false
| null |
Write a positive Yelp review for a Chinese restaurant called Dim Sum Queen
|
We went to Dim Sum Queen for lunch today and absolutely loved it. The Xiao Long Bao was juicy and fresh, the pork buns were perfectly fluffy, and everything was so fast. Great ambiance as well, and lots of space for large groups. Definitely coming again!
|
|
null | false
| null |
On 24 January 1829, Archimandrite Georgije was consecrated Bishop of Pakrac, since the former Bishop Josef (Putnik) moved to the Timisoara Cathedral. Bishop Georgije remained there for a decade.
In 1839, Bishop Georgije was transferred to the Diocese of Bačka on 26 May. With the death of Metropolitan Stefan Stanković, he administered the Metropolitanate of Karlovci in the period from 31 July 1841 until the election of the new Metropolitan in 1842.
Bishop Georgije was a very educated man of his time and had a penchant for literature, translating, writing and collecting books. His contemporaries were men of letters such as Sava Mrkalj, Jovan Muškatirović, Sava Tekelija, Lukijan Mušicki, Pavle Solarić, and Jovan Savić.
In addition, he helped the poor and educated young people. He translated a German cookbook into Serbian in 1804.
Georgije Hranislav died on 22 June 1843. He was buried in the Cathedral of the Serbian Orthodox Church of Novi Sad
|
What were Georgije Hranislav hobbies?
|
Georgije Hranislav had several interesting hobbies, including: literature, translating, writing and collecting books.
|
null | false
| null |
How do I open a pomegranate without exploding?
|
The trick is to open the pomegranate in a bowl of water. This way, the water will contain any bursts of juice from the pomegranate seeds.
|
|
null | false
| null |
The Wagner Group (Russian: Группа Вагнера, tr. Gruppa Vagnera), also known as PMC Wagner (Russian: ЧВК «Вагнер», tr. ChVK «Vagner»; lit. 'Wagner Private Military Company'), is a Russian paramilitary organization. It is variously described as a private military company (PMC), a network of mercenaries, or a de facto private army of Russian President Vladimir Putin. The group operates beyond the law in Russia, where private military contractors are officially forbidden
|
Who are the Wagner group
|
They are a private military organization that is endorsed by Russian President Vladimir Putin and is being used in the ongoing war against Ukraine.
|
null | false
| 335
|
While the demand for physical and manual labor is gradually declining, there is a growing need for a workforce with soft skills. Which soft skill do you think would be the most valuable in your daily life? According to an article in Forbes BIBREF0 , 70% of employed Americans agree that public speaking skills are critical to their success at work. Yet, it is one of the most dreaded acts. Many people rate the fear of public speaking even higher than the fear of death BIBREF1 . To alleviate the situation, several automated systems are now available that can quantify behavioral data for participants to reflect on BIBREF2 . Predicting the viewers' ratings from the speech transcripts would enable these systems to generate feedback on the potential audience behavior.
Predicting human behavior, however, is challenging due to its huge variability and the way the variables interact with each other. Running Randomized Control Trials (RCT) to decouple each variable is not always feasible and also expensive. It is possible to collect a large amount of observational data due to the advent of content sharing platforms such as YouTube, Massive Open Online Courses (MOOC), or ted.com. However, the uncontrolled variables in the observational dataset always keep a possibility of incorporating the effects of the “data bias” into the prediction model. Recently, the problems of using biased datasets are becoming apparent. BIBREF3 showed that the error rates in the commercial face-detectors for the dark-skinned females are 43 times higher than the light-skinned males due to the bias in the training dataset. The unfortunate incident of Google's photo app tagging African-American people as “Gorilla” BIBREF4 also highlights the severity of this issue.
We address the data bias issue as much as possible by carefully analyzing the relationships of different variables in the data generating process. We use a Causal Diagram BIBREF5 , BIBREF6 to analyze and remove the effects of the data bias (e.g., the speakers' reputations, popularity gained by publicity, etc.) in our prediction model. In order to make the prediction model less biased to the speakers' race and gender, we confine our analysis to the transcripts only. Besides, we normalize the ratings to remove the effects of the unwanted variables such as the speakers' reputations, publicity, contemporary hot topics, etc.
For our analysis, we curate an observational dataset of public speech transcripts and other meta-data collected from the ted.com website. This website contains a large collection of high-quality public speeches that are freely available to watch, share, rate, and comment on. Every day, numerous people watch and annotate their perceptions about the talks. Our dataset contains 2231 public speech transcripts and over 5 million ratings from the spontaneous viewers of the talks. The viewers annotate each talk by 14 different labels—Beautiful, Confusing, Courageous, Fascinating, Funny, Informative, Ingenious, Inspiring, Jaw-Dropping, Long-winded, Obnoxious, OK, Persuasive, and Unconvincing.
We use two neural network architectures in the prediction task. In the first architecture, we use LSTM BIBREF7 for a sequential input of the words within the sentences of the transcripts. In the second architecture, we use TreeLSTM BIBREF8 to represent the input sentences in the form of a dependency tree. Our experiments show that the dependency tree-based model can predict the TED talk ratings with slightly higher performance (average F-score 0.77) than the word sequence model (average F-score 0.76). To the best of our knowledge, this is the best performance in the literature on predicting the TED talk ratings. We compare the performances of these two models with a baseline of classical machine learning techniques using hand-engineered features. We find that the neural networks largely outperform the classical methods. We believe this gain in performance is achieved by the networks' ability to capture better the natural relationship of the words (as compared to the hand engineered feature selection approach in the baseline methods) and the correlations among different rating labels.
We find that the neural networks largely outperform the classical methods.
|
Do the neural networks largely outperform the classical methods?
|
Yes, they do.
|
null | false
| null |
Which of the following are cities in the PNW in the US: Seattle, Portland, Hillsboro, Eugene, Tacoma, Vancouver, Surrey, Richmond.
|
Seattle, Portland, Hillsboro, Eugene, Tacoma, Vancouver are cities in PNW. Vancouver is also in Canada but there's a Vancouver in WA as well.
|
|
null | false
| null |
Verstappen is the son of racing drivers Jos Verstappen, who also competed in Formula One, and Sophie Kumpen. He had a successful run in junior karting and single-seater categories – including KF3, WSK World Series, KZ2 and European Formula 3 – beating several records.
At the 2015 Australian Grand Prix, when he was aged 17 years, 166 days, he became the youngest driver to compete in Formula One. After spending the 2015 season with Scuderia Toro Rosso, Verstappen started his 2016 campaign with the Italian team before being promoted to parent team Red Bull Racing after four races as a replacement for Daniil Kvyat. At the age of 18, he won the 2016 Spanish Grand Prix on his debut for Red Bull Racing, becoming the youngest-ever driver and the first Dutch driver to win a Formula One Grand Prix. After winning the 2021 Abu Dhabi Grand Prix, Verstappen became the first Dutch driver to win the Formula One World Championship, and the 34th Formula One World Drivers' Champion. He won his second consecutive Formula One championship the next season.
|
Extract the names of the Formula One teams Verstappen was a part of from the text. Separate them with a comma.
|
Scuderia Toro Rosso, Red Bull Racing
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.