(Original title: 6 translators appeared out of thin air within a year. Do you think this matter is reliable?)
Text | Lina
Not long ago, the microblogging of the smart hardware circle was touched by a piece of news. IBM Watson and Australia’s Lingmo launched a Translate One2One headset capable of “simultaneous machine interpretation”. The headset not only supports simultaneous interpretation in English, Chinese, Japanese, French, Italian, German, Portuguese, and Spanish (3-5 seconds), but also does not require external Bluetooth or WiFi, priced at 179 US dollars. The official website shows that shipments will be made in July.
The author was @en on a microblogging circle. There are only two central ideas for everyone to ask - "Is it reliable?" and "I've bought it."
Coincidentally, just a few days ago, a domestic company named Double Monkey Technology also launched four translation machines at a stretch: Magic Brains, Magic Brains, Demon Brain Guides, and Demon Translation Officers. The four product models vary in size and application scenario, but claim to be capable of simultaneous interpretation in 28 languages.
According to statistics from the intellectual and the West, from May 2016 to June 2017, there were 6 companies that introduced a similar form of translator/translator headset, both of which we are familiar with, the domestic voice giant HKUST. There are also start-up companies from Australia, Japan, and the Netherlands. In these translation machine hardware, there are more translation technologies such as IBM Watson, Google, Baidu... and other startup companies such as Caiyun AI.
First, translation headphones
1, Australian start-up Lingmo - Translate One2One
Since the article originated from this translation headset, we will first talk about the combination of IBM Watson+ Australian startup Lingmo.
Machine translation steps can be roughly divided as follows: speech recognition - natural semantic understanding - translation - speech synthesis (also divided into: speech recognition - machine translation - speech synthesis). Translate One2One uses IBM Watson's Natural Language Understand (NLU) translation API, combining Lingmo's own hardware technology and machine learning applications.
The features of this headset, as described above, not only support simultaneous interpretation in English, Chinese, Japanese, French, Italian, German, Portuguese, and Spanish (3-5 seconds), but also does not require external Bluetooth. Or WiFi, priced at 179 US dollars, the official website shows that shipments will be in July.
IBM Watson can almost be regarded as a natural linguistic processing (NLP) ancestral character. As early as 2011, Watson was put on the US intelligence quiz show "Dangerous Edge." This Q&A program is different from traditional game setups in that the player is faced with the answer and needs to give the answer corresponding to this answer. The contestants must have knowledge of history, literature, politics, science, and popular culture. For example: When the clue given is "Father of the United States, cut down the cherry tree," the player can respond like this: "Who is George Washington?"
In addition, contestants sometimes need skills such as parsing hidden meanings, irony and riddles to get answers. For example, during a show, the contestant drew a category called “Chicks Dig Me”. The literal meaning is “Chicken likes me” but the real meaning is “Sister loves me”.
For Watson, the difficulty in participating in such quiz shows is not a reserve of knowledge—after all, Watson stores 4TB of data far beyond the human brain. The real difficulty is for the computer to learn to parse the human language, identify subtle meanings in the human language, and distinguish logics and clues such as sarcastic tone, riddles, word formation, psalm clues, and so on. Then it is the decomposition of the title, quick search of their own memory data, find the best answer.
As for the results of this competition, everyone knows that - Watson defeated two human champions and won first place.
Six years later, Watson didn't get much on TV, but he never stopped researching natural language processing (and other artificial intelligence). Today's Watson not only translates a variety of languages, but also conducts emotional analysis, personality traits, writing style categorization, and social orientation in people's language texts.
2. New York Waverley Lab - The Pilot Translator
In the article describing Translate One2One, there is often another similar translation headset - The Pilot Translator (also referred to as Pilot headset). The translation headset was launched in May 2016 by the New York-based startup Waverley Lab. It initially received support from 19,000 people on the Indiegogo crowdfunding platform and raised .5 million. Today, a total of 25,000 units are pre-sold and a total of US million has been raised. The first batch of products is expected to ship in the fall of 2017. The company also participated in the MWC World Mobile Conference held in Barcelona in the 3rd year of this year.
Waverley Lab was founded in 2014. According to CEO Andrew Ochoa, the original intention of setting up this company was... He met a French girl who wanted to communicate with the other party and was suffering from language barriers. At present, the company has five core members. In addition to CEO Andrew Ochoa, Jainam Shah, vice president of hardware engineering, William Goethals, vice president of manufacturing, Sergio Delrio Diaz, vice president of design, and Marion Guerriero, vice president of marketing.
(Waverley Lab's booth at MWC)
This wireless translation headset needs to be connected with a mobile phone App, using Bluetooth to connect, and both sides of the conversation wear a headset. The current pre-sale version supports simultaneous interpretation in five languages (English, French, Italian, Portuguese, Spanish). The official website shows that users can also purchase additional Chinese, German, Japanese, Polish, Greek, and other extras this fall. Languages.
Of course, just like all other wireless headphones, you can use it directly to listen to music, currently available in red, white and black colors. And the most amazing thing is that this headset you can also find a lot of unofficial pre-sale links on the magical Taobao......
Second, the translation machine
In addition to translating headphones, a flat rectangular translator is a common form of translation hardware. Some will bring a small screen to display the recognized/composite text on the screen, while others have no screen and can only perform speech translation.
1. HKUST News - Xiao Translation Machine
This Xiao translation machine can be regarded as "net red translator", at the 2016 annual conference on November 23 last year, HKUST launched a "Xiao translation machine", which can achieve the Chinese and English. Mutual translation or translation between Chinese and Uighur.
This is a flat, rectangular device that resembles a small remote control. It has a red, blue, and two-language stylus. Before speaking English or Uighur, you need to press the corresponding button. On March 10 this year, some leaders of the country also experienced the translation of the translator. The news was also touched by a small screen.
2, double monkey technology - magic brain series translator
Double Monkey Technology is a relatively magical (mysterious) company. The information that can be found on the Internet is almost zero except for official propaganda. Just last week, they had just held a press conference in Hainan. They introduced four products: Magic Brain, Magic Brain, Demon Brain Guide, and Demon Translation Officer. Among them, the Magic Brain Stroke is similar to the other translation machines described in this column. In addition to the simultaneous interpretation of 28 languages, it also has recording and voice/word conversion functions. The price is RMB 1999.
Double Monkey Technology was established in 2015, the company is located in Hainan International Tourism Island, the product is headquartered in Shenzhen Nanshan Science and Technology Park. According to the official website of the company, the company has invested more than 30 million yuan in total, and currently owns the complete independent intellectual property rights of Dasheng Culture Series (Grand Saint Magic Brains + Dasheng IP + Dasheng Speech Library + Dasheng Semantics).
3, Divider - Quasi-translator
Quasi-children translation machine by the split tower technology in Beijing this year in May this year, a crowded screen with a screen translation machine, crowdfunding price of about 699-799 yuan. Like other translators, the Quasi-Translator requires the user to speak and press the button, and then the machine translates. Both the words spoken and the translated sentences are displayed on the screen. At present, the quasi-child translator supports the mutual translation of languages such as Chinese, English, Japanese, Korean, and Thai. There is no need to connect a mobile phone, but it still needs to rely on a network connection.
Split Tower Technology was established on July 15, 2016. Both founder CEO Guan Lei and CTO Zhang Ming are masters of Tsinghua University. In May 2017, the company announced the completion of the tens of millions of angel round of financing, led by the Eastern National Lion. Jingdong Crowdfunding of the Qualcomm Translator has now been completed. According to the crowdfunding page, 5086 users support this crowdfunding plan.
4, Logbar - ili (咿哩) translator
Compared with its similar products, the ili(trans) translation machine built by Japan's Logbar Company can be said to be one of the earliest products to be announced but it has the least relative function. As early as January 2016, Logbar launched the ili interpreter at CES that year, but it was not until May of this year that the company announced the first batch of products for sale and was priced at US9. At present, the first ili translators have all been sold out and are expected to be shipped in October.
Each ili translator is programmed to convert only one specific language to another. For example, when an ili translator is set to convert only English to Japanese, no other language can be converted. . And it can only support one-way conversations now. That is to say, if an ili is set to be converted to English in Japanese, it cannot be translated from Japanese to English....
Currently, ili supports translations in English, Spanish, Mandarin, and Japanese. However, because the translator does not have networking, its vocabulary is limited to words and phrases that the traveler may use, such as advice on transportation, shopping or dining.
Third, statistical machine translation + artificial neural network translation
The history of machine translation dates back to several decades. In 1954, Georgetown University and IBM jointly used the IBM-701 computer to complete the machine translation experiments for the first time in English and Russian. It opened the premise of machine translation research and evolved many different translation methods in decades.
With the birth and rise of the Internet, since the new century, the amount of verbal and verbal data generated by humans has exploded, and statistical methods have been fully applied. Internet companies such as Google, Baidu, and Microsoft have set up a machine translation research group and developed a machine translation system based on internet big data, so that machine translation is truly practical.
Since 2013, with the breakthrough in deep learning in terms of images, speech, etc., machine translation based on artificial neural networks (Nephine Machine Translation) is also gradually emerging, and the translation effect becomes better and better.
In a nutshell, statistical machine translation lets machines learn how to translate phrases or word phrases. For example, a combination can be "my blue car" => "my blue car" or "blue car" => "blue car ". Each combination gets a score that evaluates the correctness of the translation (used by the user). The computer will try to combine different combinations of phrases to produce sentence translations in order to maximize translation fluency. The collection of these combinations is similar to the 100 words that you use to travel around the US, but the traditional translation system has hundreds of millions of translation combinations.
Machine translation based on artificial neural networks is more like a black box. Most of these translation systems use an "encoder-decoder" model. An "encoder" converts each source word into a sequence of digital vectors, and a "decoder" generates a translation word by picking information from each vector. The "attention model" weights each vector to determine which encoding parts of the original sentence contribute to the next translated word. As with other applications of deep learning, we do not know what standards machines rely on, but in most cases machine translation based on artificial neural networks is more fluid.
At present, many companies use artificial neural network machine translation + statistical machine translation and other integrated technologies for translation.
Conclusion: Little Red Sea? Pseudo demand?
With the development of economy and the improvement of people’s living standards, more and more people have begun to choose to travel abroad as leisure and entertainment, and language communication obstacles are becoming more and more noticeable as a pain point for users. Nearly all of the translator/translator headsets mentioned above are aimed at tourism users who have low accuracy and high fault tolerance. Simultaneous interpretation of business meetings cannot be replaced by machines for the time being.
The demand for tourism translation exists, but we may have to put a question mark on the needs of the translator. On the one hand, is there a market for a single functional hardware product when smartphones are universally available and all needs can be met by the app? How effective are these translation hardware effects compared to Google Translate App and Baidu Translation App that we are familiar with? Moreover, at present, most products still stay in the whisper. To say a word, one sentence does not provide simultaneous interpreting, which is very similar to the use of App translation (almost all companies mentioned in the article provide free translation app).
On the other hand, if this translator needs to rely on the Internet, can the existing network level guarantee the smoothness of networks in remote areas? And if the translator does not need to rely on the Internet, can the effectiveness of translation and accuracy of translation be guaranteed? How many functions can enterprises achieve in the promotion process? ... These questions have time to verify.
Prior to this, you must first learn English well.
燑br>
100W Medical Power Supply
100W Medical Power Supply,100W Medical Device Power Supply,100W Medical Power Adapter,100W Rade Power Supplies
Shenzhen Longxc Power Supply Co., Ltd , https://www.longxcpower.com
برچسب:
،
ادامه مطلب
بازدید:
