<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Alexei Rudak on Medium]]></title>
        <description><![CDATA[Stories by Alexei Rudak on Medium]]></description>
        <link>https://medium.com/@alexeirudak?source=rss-8f2249eba976------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*g6_XVjR1Co6P-_jI</url>
            <title>Stories by Alexei Rudak on Medium</title>
            <link>https://medium.com/@alexeirudak?source=rss-8f2249eba976------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 05:44:36 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@alexeirudak/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[ChatGPT Translator VS Lingvanex: which one is better?]]></title>
            <link>https://medium.com/@alexeirudak/chatgpt-translator-vs-mine-which-one-is-better-5816f5f0e89c?source=rss-8f2249eba976------2</link>
            <guid isPermaLink="false">https://medium.com/p/5816f5f0e89c</guid>
            <dc:creator><![CDATA[Alexei Rudak]]></dc:creator>
            <pubDate>Tue, 01 Aug 2023 09:24:46 GMT</pubDate>
            <atom:updated>2023-08-02T14:49:39.608Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*e_2AU5kUInRM5eB3" /></figure><p>I’ve seen many articles and rave reviews claiming that ChatGPT’s translation capabilities are on par with DeepL and Google, and sometimes even surpass them. As the founder of a company <a href="https://lingvanex.com/"><strong>Lingvanex</strong></a> that has been developing its own translation solutions for the past six years, I became curious about how true all of this is and how our solution compares to ChatGPT. Should I be worried about such a strong competitor?</p><p>To compare the quality of translation, we prepared test datasets for seven language pairs:</p><ul><li>English-Spanish</li><li>English-German</li><li>English-Russian</li><li>English-French</li><li>English-Italian</li><li>English-Portuguese</li><li>English-Finnish</li></ul><p>Each test dataset contains around 2500 lines and includes sentences of various themes, lengths, styles, and formatting, to eliminate text selection bias for a specific translator.</p><p>ChatGPT recently released the API version 4 for limited access. At the moment, access is only available to previously created accounts that have already paid for version 3.5. According to reviews, the new version has made significant advancements in terms of quality compared to version 3.5. We will check that too!</p><p>For testing, we will use two metrics: BLEU and COMET.</p><p><strong>BLEU</strong> — a widely recognized standard for testing the quality of translation. By default, we will use the Sacre Bleu version. This version is used in the MT machine translation conference and various international competitions. In this metric, the comparison of translations is based on the number of n-grams (combinations of words) that follow one another. The goal of the metric is to find the maximum matching combinations between the translation made by a human and the one made by a machine. The comparison begins with clusters of four words. If there are none, it searches for three-word n-grams. If further matches are not found, it can go down to one n-gram. Points are awarded for each sequence of words (tokens) that the program finds. The drawback of the metric is that it does not account for synonyms, and if the translation accurately conveys the thought but with different words, it will show 0.</p><p><strong>COMET</strong> — A metric designed to solve the problem of comparing synonyms, which metrics based on the symbolic comparison of two strings cannot handle. If the result of the translation is a semantically similar phrase, but described with different words, the metric will indicate comparable results. It should be noted that its result will also depend on the diversity of the language corpus used to construct the comparison classifier. This metric is widely used as an alternative to the BLEU metric.</p><p>Prompts we used for ChatGPT translation:</p><p><em>You are TranslateGPT. You translate user messages from English to Italian (Finnish / French / German / Portuguese / Russian / Spanish). You are the most accurate English to X translator in the world.</em></p><p>Below are graphs with test results:</p><p><strong>English-Finnish translation</strong></p><p>This language pair shows noticeable improvement in the translation quality of ChatGPT 4 compared to version 3.5. According to the COMET metric, ChatGPT4 slightly outperforms Lingvanex.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*GXOKxml-PM4w-iVKqScj_w.png" /></figure><p><strong>English-French translation</strong></p><p>When it comes to translating into German, the situation is the same as with French. However, Lingvanex’s lag in the COMET metric is minimal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xRzi4rLKi7de8QK8HhaGFw.png" /></figure><p><strong>English-German translation</strong></p><p>When it comes to translating into German, the situation is the same as with French. However, Lingvanex’s lag in the COMET metric is minimal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*7GkHosUAGtYEpW-2O_rAXQ.png" /></figure><p>Let’s compile all the differences in a table. In red font, we’ll indicate where ChatGPT falls short of Lingvanex. In green font, we’ll mark where it surpasses. The data is relevant as of July 31, 2023.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SRndhNWsn9vMeh5fk62w2w.png" /></figure><p>The Lingvanex translation price was calculated based on the cost of a month’s rent for a basic GPU server (150 dollars) + the monthly price for a Lingvanex language model translation (from 100 dollars), and the number of characters that can be translated in a month with this configuration.</p><p><strong>Conclusion:</strong></p><p>The test results show that while ChatGPT 3.5 is mostly inferior to Lingvanex in translation quality, according to the COMET metric, ChatGPT4 often matches Lingvanex. It’s worth noting that currently, translating large volumes of text with ChatGPT4 is very expensive. In order to perform the tests for this article and translate roughly 20,000 lines through ChatGPT4, a total of $45 was expended. The calculation of the translation cost can be confusing, as it’s difficult to estimate in advance, in tokens, how much you’ll end up paying for the translation.</p><p>At the moment, the translation speed of ChatGPT 4 is unstable, most likely depending on the current load on their servers. We had to take 3–4 second breaks between requests. Overall, the translation speed with the test dataset was about 8 words per second. Our solution enables the translation of several thousand words per second, even on weaker servers. In addition, we observed censorship in the translations: if a line contains obscene language, ChatGPT will not proceed to translate the entire sentence.</p><p>Therefore, ChatGPT is better used for stylistic translation of small volumes of text without special security requirements. Furthermore, styles and themes can be altered on the go. By carefully selecting prompts, you can achieve improved quality tailored to a specific task, but this requires going through a significant number of prompts.</p><p>Lingvanex translation solutions are better used where large volumes of translation are required at a low cost, with safety, speed, and stability.</p><p>I admit that for some language pairs, the difference in translation quality may be different, but testing all possible pairs is lengthy and costly.</p><p>In general, solutions from ChatGPT and Lingvanex are designed for different purposes and should be chosen depending on the task.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5816f5f0e89c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I spent a Million Bucks on Freelancers. My Honest Review]]></title>
            <link>https://medium.com/@alexeirudak/blowing-a-million-bucks-my-wild-journey-to-create-the-ultimate-machine-translation-tool-135147298f1f?source=rss-8f2249eba976------2</link>
            <guid isPermaLink="false">https://medium.com/p/135147298f1f</guid>
            <dc:creator><![CDATA[Alexei Rudak]]></dc:creator>
            <pubDate>Tue, 28 Mar 2023 15:13:42 GMT</pubDate>
            <atom:updated>2023-03-29T19:15:24.540Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*AqLGjATBKk8ixPIkNQQzdw.png" /></figure><p>A week ago marked six years since I opened a company Lingvanex (<a href="http://www.lingvanex.com/">www.lingvanex.com</a>) to delve into machine translation. A lot has happened since then. Now, we have a team of 40 people, 120 servers, and over a million users. But it wasn’t always like this. I remember the days when there were just two of us, crammed into a tiny, poorly ventilated room. We had to open the windows every hour to keep from suffocating. We were bursting with hope, ideas, and a burning desire to create a project that could give Google a run for its money.</p><p>Back in 2017, we had a prototype translator for Linux that could already do some translations. We managed to assemble one for iOS as well. But things weren’t so straightforward with Windows. Clients wanted an installation for this OS, and we had to take action. To tackle this task, I had to recompile a bunch of libraries with a heap of dependencies, but I didn’t have the experience. When I went to freelance websites and posted a few ads, only a handful of applications came in. It became clear that the task wasn’t as simple as it seemed.</p><p>The internet was flooded with offers from outsourcing companies promising to develop projects of any complexity. As a software engineer by education, the claim “any complexity” piqued my curiosity, while also raising an eyebrow. I decided to give it a shot! Everything seemed legit on the surface, with communication conducted on behalf of a legal entity. Our company wanted to order the development, but first, we needed an estimate of how long the task might take.</p><p>It’s worth noting that the companies I spoke with varied in size. One of them was a well-known brand and a leader in Europe. I contacted their sales manager and was promised a follow-up. After a week without a response, I reached out again, and they assured me once more that they would definitely get in touch. They didn’t. Oh well!</p><p>Then I started looking for smaller companies, with up to 100 people, a good portfolio, and niche expertise. Before starting any conversation, I would always ask if it was true that they could handle projects of any complexity, as stated on their website. They would assure me that yes, they had done projects much more complex than mine, and they just so happened to have a programmer available who had been involved in the development of a quantum supercomputer. Great! Shall we get to work? Not quite! First, they needed two weeks to dive into the subject and required payment for researching the issue at a rate of $40 per hour. Only then would I get an answer as to whether or not they would take on the project. The offer was tempting, but of course, I wasn’t going to accept it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*2-Mn5I-d_Za8_cwWrwavjA.png" /></figure><p>One of the companies said they needed exactly 12 days to evaluate the task, and they would provide an answer on the 12th day at 1:45pm. This was unusual but kind of cool since they didn’t charge any money for studying my issue. Right on time at 1:45pm, they called me on Skype and provided a detailed answer about the task. They talked about neural networks, their configurations, Windows and Linux libraries, computers, processors, and even molecules. The answer was long and logical, but after 20 minutes, I just asked — “Can you do this, just answer ‘yes’ or ‘no’?” — of course, the answer was “no.”</p><p>Somewhere around the 30th call, fortune finally smiled upon me, and the company’s manager said, “Yes, we’ve done something similar, and we have a person who can complete this task in a couple of days, but it’s Friday evening, and he’s already left for the weekend.” I desperately clung to this last chance, promising good money, a steady stream of orders, and anything else I could think of — just get this done for me. Let him look at the task right now. — “Alright, you’ve convinced me.” While I was still on the call, the manager started calling this developer on the phone, simultaneously talking to him on Skype.</p><p>In the background, I could hear disgruntled phrases, a bit of swearing, children’s cries, and a rowdy group of people — clearly, this guy knew how to both work hard and play hard :) Nevertheless, their programmer was on Skype with me within an hour, listening to what needed to be done. According to his estimates, the task could be completed in one evening for $500, which delighted me. However, it wasn’t done that evening; on Saturday morning, he said it would take two more days. By Wednesday, the task was completed, and a month later, we released the first version of our translator for Windows. We then found freelancers who could assemble a Mac OS version and create a prototype. That was one of the few positive experiences in our journey.</p><p>When it came time to develop the Android version, I thought I’d figured out how to work with outsourcing. I allocated a budget of $200 for each test task to assess a potential company. Finding another contractor with a good reputation, we had a call with the manager, who introduced us to an energetic programmer named Brian. In one day, he efficiently created the main screen of the app, and everything seemed fine. But our tech lead decided to ask why he chose a particular approach. As the questions began, the programmer started to falter and went silent. “Brian, are you with us?” I inquired. “Yes!” That was all he could answer clearly. After a bit more mumbling from him, my patience finally wore thin.</p><p>- Who wrote this code, Brian?</p><p>- I’m not Brian!</p><p>- Then who are you?</p><p>- I’m Peter!</p><p>- And where is Brian?</p><p>- He went home!</p><p>- Why didn’t you say so?</p><p>- My boss said I couldn’t.</p><p>- Get him on the phone.</p><p>- He’s right here.</p><p>The manager picked up.</p><ul><li>Why are you all lying to me?</li></ul><p>- Calm down! Brian just had a stomachache, so I let him go. I’ll get you another developer.</p><p>- That’s fine, we can manage without him. I’ll just pay and …</p><p>At this point the manager started offering me every specialist he had on the payroll. I got the feeling he’d sell his grandmother right now if the hourly rate was reasonable. Eventually, I hung up the phone and sent him $200.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*RzXDOlM4FqdQCa6zJK12VA.png" /></figure><p>It would be funny if it weren’t so sad. In the end, we had to create the Android translator ourselves, and after a year, we released it, covering apps for iOS, Android, Windows, and Mac.</p><p><strong>Mr. Hong and his magic script</strong></p><p>After sorting out the apps, we needed to whip up a data filtering algorithm that would run on our servers at lightning speed. Lo and behold, we found our coding wizard in just a day, and the price? Just the cherry on top!</p><p>Mr. Hong’s profile was impressive. Glowing reviews, an impressive portfolio, and similar projects under his belt. He had a solid grasp of English and a clear understanding of the task. Ready to jump in immediately for a cool $600. Well then! Let’s get to work!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*NTc1qw6syx8mfrD3Jmq7uA.png" /></figure><p>In just a couple of days, he sent over a script that worked both quickly and efficiently. We ran a few tests, asked him how he tackled the task, and everything seemed reasonable. The script was a few thousand lines long, so we decided to dig into it later when we had the time. But, a couple of days later, the script stopped working, and we started searching for the cause.</p><p>Amidst the vast and tangled chunk of code, he had sneakily inserted an API call to some paid service. So, if you were to remove all the code except for this call, the program would still function the same. However, we discovered this far too late — the free trial period of that service had expired. Meanwhile, Mr. Hong had already received a 5-star rating and a glowing review. What a clever guy! It just goes to show that we really need to be more careful.</p><p><strong>Beach, Bikinis, and Neural Networks</strong></p><p>Aside from the negative experiences on freelance platforms, we’ve had our fair share of positive cases as well. Once, we had a task to improve the speech recognition quality for the German language, and we began searching for someone who could help us. When we posted the project, about 20 applicants responded, mainly consisting of professors, bespectacled hairy bearded fellows with advanced degrees and high rates. Overall, they all looked similar except for one application. The profile had dozens of completed similar projects with a rating — all five stars, absolutely all of them. No one else had such an impressive rating. The sheer number of technologies this person knew was truly incredible. A genius, really! But there was one detail that made me hesitate to choose them as the contractor. Their profile was named “Nicky” and the photo featured a girl in a swimsuit by the sea. Their hourly rate was three times cheaper than the others, and we couldn’t afford to pay more. I decided to take the risk.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*-ZR4s2BLd16zxzMhw4G3xw.png" /></figure><p>When she gave me her Skype profile, it showed up as “Scott,” and after saying hello, I asked, “What’s up with that?”. Scott explained that the photo was of his wife, and he’d provide the contact of a person from China who would do everything we needed for $300, with the language model training being on his dime. It was practically impossible to complete the task for that price, as renting servers for this task would cost a few thousand dollars, but Mr. Li assured us that he’d work at a loss to gather positive reviews and ratings. He had his own servers. Lots of servers! Mind-blowing! In just two days, he sent us a trained language model that provided excellent German recognition quality.</p><p><strong>Hello, I’m Your Long-Lost Aunt!</strong></p><p>At the dawn of our company’s founding, we relied heavily on outsourcing and freelancers, which we later deeply regretted. People kept changing, and with them, expertise was lost, as newcomers had to familiarize themselves with the project all over again. By 2019, I was completely fed up and decided to ditch them for good.</p><p>It was decided to assemble the dream team from scratch, and to quickly scout for experts, they started publishing articles about the project. The first article turned out to be quite the hit. The editor, in his infinite wisdom, suggested focusing on the million dollars earned from mobile apps as the headline — you know, to rake in those sweet, sweet views. At the end of the article, I left my email, personal phone number, and office address, inviting one and all to swing by, learn more about the project, and perhaps propose a cheeky partnership. And that’s when the wild ride began!</p><p>The phone was ringing off the hook. I spent the entire first day just answering it, afraid of missing something important. But folks offered us coffee, computers, grocery delivery, and even guitar lessons for the team. After lunch, a dapper gentleman knocked on our door, asking to speak with the company’s founder.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*r-GeRnGX4Ph3OmdL4k75Yw.png" /></figure><p>Sporting a snazzy jacket, a large briefcase, and a confident gaze, he seemed like an investor, a client, or at the very least a potential partner. He immediately asked for the founder, and when I emerged, he suggested we step aside to discuss a few matters. Within seconds, he whipped out a huge jar from his bag and asked, “Do you need honey?” — “What?!” — “Honey??” — “Man, we’re developing a neural network-based translator.” — “What the hell do we need honey for?”</p><p>He then proceeded to tell me about his bees, his apiary, and how this honey would do wonders for our team. But, frankly, I couldn’t care less. I headed back to the office to continue fielding calls.</p><p>An hour later, an elderly man walked in, carrying a massive chart detailing some mega-algorithm for learning English. He wanted to partner up with us. The partnership would involve us developing an app based on his work, at our expense, and then pocketing part of the future profits. “Very tempting,” I said sarcastically, before dashing off to resume my call-answering marathon.</p><p>Next, was a very weird call from a lady:</p><ul><li>Hey there, Alex!</li><li>Well hello there, who might you be?</li><li>It’s Helga!</li><li>Good day, Helga. Is there anything you need?</li><li>You.</li><li>And who are you?</li><li>I’m your relative!</li></ul><p>I had relatives with that name, but the voice was absolutely unfamiliar.</p><ul><li>Do you have a question?</li><li>An important one!</li><li>Well, spit it out, then!</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*sRi_z9oD2fAY6NSrTlitQg.png" /></figure><p>In response, there was silence, and I wasn’t in the mood for chatting with a stranger. A few seconds later, the conversation was over, and I removed the phone number and office address from my profile under the article. It felt more peaceful. Communicating via email and social media was easy, and I replied to everyone within the next two days. Overall, the outcome was positive — we found clients, partners, and team members. In short, articles work!</p><p><strong>To end with…</strong></p><p>When I first watched the TV show “Silicon Valley,” I thought it was hilarious and absurd. Our company had just opened, and I couldn’t believe that such things could happen in the IT business. Boy, was I wrong. It turns out that such shenanigans happen to many, and it only becomes funny when the company starts doing well. When it all begins, everyone’s stressed, and it feels different. Sometimes it seems that instead of developing and selling services, we should have just filmed a TV show about everything that happened to us. Then sell it to Netflix and make a billion. And maybe even snatch an Oscar, but that’s not certain.</p><p>Despite all the challenges, we persevered, improved our product’s quality, found serious clients, and expanded our team. The most important thing is your team — with the right people, you can tackle any project. I still believe in that. But you have to know how to pick the right folks.</p><p>Now we’re focused on two products:</p><p>a) Unlimited translation into 100 languages</p><p>b) Unlimited voice-to-text transcription in 90 languages</p><p>Why would anyone be interested in this?</p><p>Well, for starters, for $200/month, you can translate 30 billion characters and convert 30,000 hours of audio into text. That’s a whole lotta bang for your buck. And all your data stays with you. We spent years achieving these results. To get our free trial, shoot us an email at info@lingvanex.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=135147298f1f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I spent $1,000,000 to make my machine translator. Thoughts and Results]]></title>
            <link>https://medium.com/@alexeirudak/i-spent-1-000-000-to-make-my-machine-translator-thoughts-and-results-81cfb9cbc08f?source=rss-8f2249eba976------2</link>
            <guid isPermaLink="false">https://medium.com/p/81cfb9cbc08f</guid>
            <category><![CDATA[growth-hacking]]></category>
            <category><![CDATA[startup-lessons]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Alexei Rudak]]></dc:creator>
            <pubDate>Wed, 16 Feb 2022 07:22:04 GMT</pubDate>
            <atom:updated>2022-02-16T07:22:04.157Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*silfb3y6A4jNlwMRGTQePQ.jpeg" /></figure><p>I have been working on machine translation since 2012, and this business was not chosen by chance. Many years ago I noticed the big impact of localisation on revenue. In our case, support for only 15 languages instead of one helped us to increase the number of mobile app downloads to 35 million without advertising costs. It was the most effective money and time investment. And this was the main reason why I chose this business.</p><p>When I published my first story about the project, several hundred people wrote to me. I tried to answer everyone. Basically, there were questions about how to make money on mobile applications, how to start business, etc. A year later, several people said that their lives had changed a lot for the better. So, the advices helped.</p><p>The most important advice will be the next:</p><p>Before starting your own big business, try to find a company or people who can help you to get experience in the field you are interested in. Try to agree for smaller salary or even work for free. The main point is to get the experience you need — quickly. You need to face with real business problems, people and tasks. You can’t learn business using books and courses. Ideally, to get experience in several areas at once (development, marketing, management). Small startup companies of 10–15 people are suitable for this.</p><p>It is not easy to find a successful company where you will have the opportunity to work closely with an experienced manager, to watch on how to make decisions. Most likely, you will need to change several companies and spend a couple of years. The main idea is to quit your job quickly once you see no perspective there.</p><p>Learn from the mistakes of others. It takes a very long time to learn from your own. Therefore, choose your place of work very carefully.</p><p>The most important conclusion I have realized in 15 years of business — people is a key. Not an idea, not a market, not a technology, but a team. It doesn’t matter what is the competition on market. A good team will make a project of any complexity. Even if there are some mistakes at the beginning it will be always success at the end.</p><p>As a founder of the company you should be public and social. It’s necessary to write posts about your project everywhere, take part in events, make interviews. Try to avoid outsourcing of development to freelancers, and focus to gather your own team of professionals. To recruit such people, a personal example is very important. Experienced people want to work with the same as they are.</p><p>Posts about your project will help to find good teammates. You can’t find them using vacancies and recruiters. Well-established processes in management and a team of even 10 experts can do wonders.</p><p><strong>The beginning of 2020.</strong></p><p>At the beginning of 2020, we have made a great progress in the quality of translation.</p><p>Previously, to correct translation errors, the “Suggest correct translation” feature was added to our applications. A year later, we turned it off due to the fact that people started spamming and offering complete nonsense for correction. After analyzing thousands of records, I saw that few things turned out to be useful. A translation cache was also made. It works well on 1 and 2 word phrases and allows you to save about 30% of all requests.</p><p>We rented 20 dedicated servers with one GTX 1080 graphics card in each.</p><p>With the increase in dataset volume, we increased the number of layers, RNN and other parameters of the neural network. It was necessary to achieve a certain level of the quality of translation, to make users not go to competitors.</p><p>And then there was one problem. According to our calculations, with the current neural network parameters, it will take 28 years to train 200 language pairs. Even considering all the servers we had, it take so long. Nvidia GTX 1080 graphics cards were very slow for our tasks, and we haven’t enough money to rent expensive V100.</p><p>During the year, we participated in various startup competitions with the hope of winning free credits from sponsors. And one day it happend. A couple of months later, we won the AWS Startup Challenge and received 100,000 free credits that could be spent on training models. Instead of servers with 1 x GTX 1080, we began to take 8 x Nvidia V100, From the 28 years, the training time was reduced to 2 months, during which we overtrained our models with the best quality.</p><p>At the same time, I watched what our competitors were doing and analyzed the market. More and more projects in the field of machine translation began to appear, such as real-time translation of conferences, translation of video clips and podcasts. Companies that have been specializing in human translation for a long time have begun to invest in machine translation . The basis was the business impact of localization.</p><p>The improvement in quality of machine translation has made it possible to reduce the cost of human translation many times, because in localisation the main expense is the salary of people who speak foreign languages. The better the computer translates, the less time a person needs to correct mistakes behind it. The development of technology has already made it possible in some cases to completely replace human translation with machine translation.</p><p>On September 1, 2020, the world’s leading scientific journal “Nature” published a report that machine translation has already reached human quality in English-Czech news.</p><p>Despite the huge demand for translations, the use of human translation for this task is still expensive and takes a lot of time. The translation of one word by a human costs in average of 5–8 cents. In case of cheaper price of human translation, the more companies are willing to invest in localization. Using more and more high-quality machine translation and CAT tools, the price of word translation is decreasing every year, and the speed of delivery of localised solutions is growing. In future it will lead to an explosive growth of the translation market. In the picture below, I showed this idea.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m9KcNbgM9pRyA85Vm_qEag.png" /></figure><p><strong>Pivot to B2B market</strong></p><p>Every year more than 1 billion people search the Internet for solutions related to translations. The amount of data in different languages is increasing every day, and the share of English as the Internet language has decreased to 25%.</p><p>Until 2020, our focus was on mobile translation applications. Recently, the Appstore Search Optimization (SEO) for mobile applications (which helped us to get 40 million downloads and earn the first million $) changed it’s algorithm. The keywords in the Apple App store without purchasing paid installations has become ineffective. The user acquisition with paid traffic has become very expensive.</p><p>At the end of 2020, we decided to try the B2B market. We think that any international business needs a translation feature. This can be the translation of documents, a website, multilingual customer support, competitor analysis, and more.</p><p>In five years I got thousands of questions “Why Lingvanex is better than Google”. I tried to give different answers, but now I try to answer briefly — data privacy, functionality, price, quality of support service.</p><p>If you are a student or a tourist and you need to translate a small post for personal goals, then you can use free services like Google or Microsoft. But if you are a company, then it may not be safe to translate customer emails, invoices, certificates and other internal documents via free online services, because such companies collect your data in order to sell them to advertisers. Also to translate large volumes of text using Google API will be very expensive.</p><p>For such cases, we have made it possible to install translator on a corporate network and translate text and files offline. Moreover, everything happens very quickly and with the preservation of formatting. We have all languages that Google has.</p><p>The first product we made for B2B market was on-premise translation server. It allows you to translate text, HTML (websites), files into 120 languages with no limits. We have spent enough time to make translations ultra-fast. Now 1 billion characters of text can be translated in 1 day on one Nvidia GTX 1080 video card. Also, unlike the Cloud API (Google, Microsoft, etc.), all your data remains with you, because everything works without the Internet.</p><p>It was privacy that became the basis for our first B2B integration with the Vivaldi browser.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fse1hASATJyOnTbLfm21Jw.png" /></figure><p>The translation server needs Ubuntu OS and comes as a docker image. After the launch, the REST API becomes available, which can be integrated into your project.</p><p>The function of translating web pages has been added not only to the desktop, but also to the mobile version of the browser.</p><p>Once we finished with Vivaldi browser, we integrated server into the medical library. The client wanted to save costs on HDD and not to store the translated pages in the database. Therefore, translations of web pages occur dynamically as the user surf around the website.</p><p>I thought about using the server to create localized versions of websites like Stackoverflow, Quora and then to make money from advertising. Because I see lot’s of websites made this way at Google Search. But didn’t realized this idea yet.</p><p>The third project of server integration was a large e-commerce website in the Asian region. Automatic translation of product descriptions into 87 languages was required. Then we sold the server to a lead generation agency, business analytics company, a plugin for translating websites, an electronics plant and others. In general, the server will be useful wherever there is a lot of content and an international audience.</p><p>Gradually, our focus shifted to business solutions instead of mobile applications.</p><p>For several years, we have often been asked about the feature of speech transcription (converting voice into text). Therefore, speech recognition in 20 languages has been added to the translation server. Thus, the server can translate text, files, HTML and do voice transcription into text.</p><p>Now the translation server is our flagship product. A lot of time and effort has been invested in it. The price of the server begins from 2000 euros per year (depending on the number of languages and functions). It’s less than 200 euro / month. A free demo version of the server is available for 1 week.</p><p>More information <a href="https://lingvanex.com/translationserver/"><strong>ABOUT TRANSLATION SERVER HERE</strong></a></p><p>Recently we made translation SDK for iOS, Android, Windows and Mac OS. This allows you to connect the offline translation features to any mobile and desktop applications as a static library. To do this, we made separate compact models of languages (up to 100 MB). The same 120 languages are available. It is suitable for private messengers, browsers, legal and medical and other programs where data protection is required.</p><p>Also we began to offer Cloud API for translation. The cost starts from $3 per million characters, with the same number of languages and comparable quality as Google. The price depends on the volume. By default, the price is $5 per million characters, which is 4 times cheaper than Google. We can give up to 20 million characters to test for free.</p><p>Previously, our desktop applications for Windows and Mac can translate files only online, now we have separate installers that work completely without the Internet.</p><p>Finally</p><p>There are 18 people in our team, and I am pleased that despite all the difficulties, we have found an effective business model, which we are focusing now.</p><p>The rule of 10 000 hours always works. If you are engaged in any activity for a long time, you become a professional and then you begin to succeed. The main thing is to remember about this and not to give up in advance. And it’s better to get a job in a company with a similar project before and just adopt a positive experience. It’s much faster to succeed that way.</p><p>More about our projects <a href="https://lingvanex.com/platform/"><strong>HERE</strong></a></p><p>Write me to info@lingvanex.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=81cfb9cbc08f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I earned $1M with no business experience, and spent it to create a brand-new translator]]></title>
            <link>https://medium.com/@alexeirudak/how-i-earned-1m-with-no-business-experience-and-spent-it-to-create-a-brand-new-translator-54cbd68fa9b?source=rss-8f2249eba976------2</link>
            <guid isPermaLink="false">https://medium.com/p/54cbd68fa9b</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[translation]]></category>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[mobile-app-development]]></category>
            <dc:creator><![CDATA[Alexei Rudak]]></dc:creator>
            <pubDate>Mon, 23 Mar 2020 08:41:31 GMT</pubDate>
            <atom:updated>2022-03-17T16:43:55.156Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>How it all started</strong></h3><p>10 years ago I worked as a software engineer and later quit the job to start my own projects. To save some extra money, I went to a small hometown, where I used to work on a website for students, accounting software and mobile games at the same time. Having no business experience caused some issues with generating income, so all projects had to be closed. I came back to the capital to get a job, again. The story repeated itself a few times.</p><p>When I was broke again then faced a total economical crisis. I couldn’t find a job and it felt awful. It was a good reason to see the world through sober eyes. I had to honestly admit that I didn’t know what niche to choose for my business. Doing projects you like seemed like a way to nowhere.</p><p>The only thing I was capable of is creating mobile applications. Several years of work in tech companies allowed me to gain some useful experience, so I decided to make fundamentally different apps (games, music, art, health, lifestyle, languages) and test the market needs. Prepared sets of assets and code libraries made it possible to simply create applications on various topics: 2d games, GPS trackers, simple utilities, etc. Most of them had several pictures, 2 buttons, and only one function. But it was enough to test the idea and monetization model. For example, a running app tracked the person’s speed, distance and burned calories. Nothing more. Purchasing graphics on stocks and source code re-usage helped me to create hundreds of simple applications over 2 years.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1U5iX4R5hMeVN9mqXytYFQ.jpeg" /></figure><p>At first, the applications were free. Then I added ads and in-app purchases, picked up keywords and bright icons. Users started to download my apps.</p><p>When the income reached $30K per month, I offered a friend from a large tech company to work together since I was able to earn money on test applications. He said that they have only one application — a game with a monthly income of $60K and 25K users vs $30K revenue and $200K users I had. This completely changed my vision. Turned out that it was better to create one high-quality application than one hundred poor.</p><blockquote>10X income from high-quality apps is kinda obvious, but I was alone in a small city with no experience and a team of designers and marketers. I had to pay rent and try to make a living. My applications were designed to test market niches and advertising strategies to learn what exactly I should create. It just happened that some of my apps began to generate revenue. Now you can’t make money with such apps.</blockquote><p>Some applications differed in profit: translators, navigation for trucks, music simulators (piano, drums, guitar chords, players), as well as simple casual games.</p><p>Then I noticed that in just a month, translators had been downloaded more than 1M times taking 100th position in the ranking category. There are hundreds of languages in the world, and people enter a query for every language. The niche turned out to be promising.</p><p>About 40 simple translators were later created using Google API. It cost me $20 per 1M characters. Then appeared improved versions of apps, where I included ads, in-app purchases, and voice translation.</p><p>I earned enough money to move to big city and buy a house. By that time, I had 50–70 translation applications and 5M downloads in total. User growth increased the cost of the paid Google Translate API. So business profitability has seriously decreased. Paying users translated blocks of 1K characters at a time, which forced us to limit their requests. When they faced this translation limit, they left bad reviews and got their refunds. 70% of the income covered our expenses. With large translation volumes, this business wasn’t promising. To recoup expenses, it was necessary to add advertising to applications which always scares users. That’s why we needed our API for translation.</p><blockquote>I tried to ask startup community for advice and investments but got no support in response. Most people didn’t understand why we work in a market where there is already a giant leader — Google Translator.</blockquote><p>Besides Google, several companies provided an API for translation. I was ready to pay $30K for their technology licenses in 40 languages. This would allow me to translate an unlimited number of times for a fixed price and serve any number of users on my servers. But in response, I got the amount several times higher than expected. It was too expensive. I decided to recreate their technology for translation. I invited some friends to help me with development, but then most of them had families, small children, and loans. Everyone wanted stability and a good salary instead of joining a startup. I had no experience in public speaking, charisma, or a cool prototype application to get people interested. Earning $300K on test translation applications didn’t surprise anyone.</p><p>I turned to a friend who owns an outsourcing company. At the end of 2016, he allocated a team for me. I expected to solve the problem in six months on an outsourcing basis, not to depend on Google’s API.</p><h3>On the way to my translator</h3><p>The work has begun. In 2016, we found several opensource projects — Apertium, Joshua and Moses. It was a statistical machine translation suitable for simple texts. From 3 to 40 people supported these projects. Later it became clear that we need powerful servers and high-quality datasets which are expensive. Even after we spent money on hardware and a quality dataset for one of the translation pairs, the quality left much to be desired.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zCCJEQ5GdhcdcgfVtHQePg.jpeg" /></figure><p>Technically, it didn’t boil down to the “download dataset and train” scheme. It turned out that there are a million nuances that we were not even aware of. We tried a few more resources but didn’t achieve good results. Google and Microsoft don’t disclose their achievements. Nevertheless, the work continued, freelancers joined the company.</p><p>In March 2017, we found a project called <a href="https://opennmt.net/"><strong>Open NMT</strong></a> — a joint development of Systran, one of the leaders in the machine translation market, and Harvard University. The project was just launched and offered translation based on new technology — neural networks.</p><blockquote>Modern machine translation technologies belong to large companies and they are closed. Small players, realizing how difficult it is to integrate, don’t make such attempts. This inhibits market development. The quality of the translation among the leaders differed not so much from each other. Obviously, large companies also faced a shortage of enthusiasts, research projects, startups, and open source projects in order to take new ideas, hire people, cooperate.</blockquote><p>Therefore, Systran made a bold move: they shared their developments in open source so that enthusiasts like me could get involved in this work. They created a forum where their experts began to help newcomers for free. And it brought a good return: startups, scientific works on translation began to appear since everyone could take the basis and conduct their experiments on the basis of it. Systran led this community. Then other companies joined.</p><p>At that time there was still no ubiquitous neural translation, and Open NMT offered their developments in this area, outperforming statistical machine translation in quality. I and other guys around the world could take these technologies and ask experts for advice. They willingly shared their experiences, and this allowed me to understand in which direction to move.</p><p>We took OpenNMT as a basis. This happened in early 2017 when it wasn’t perfect and contained only basic functions, purely for academic research. A lot of bugs were found, everything worked slowly, unstably and crashed under light loads. In our group chat, we all tested together, tracked errors, shared ideas, gradually increasing stability (then there were about 100 of us). At first, I wondered: how is it so, why does Systran grow its competitors? But over time, I understood the rules of the game, when more and more companies began to publish their developments for processing natural language in open source.</p><p>Even if everyone has the computing power to handle large datasets, the question of finding specialists in NLP (Natural Language Processing) is acute on the market. In 2017, this topic was much less developed than image and video processing. Fewer datasets, scientific papers, specialists, frameworks and more. There are even fewer people who are able to build a business and close any of their local niches from NLP research papers. Both top-tier companies like Google and smaller Systran players need to gain a competitive edge over players in their category.</p><h4>How do they solve it?</h4><p>It may seem strange, but in order to compete, they decide to add new players to the market. For them to appear there — you need to make the market attractive. The entry threshold is still high, and the demand for language processing technologies is growing fast (voice assistants, chatbots, translations, speech recognition, and analysis, etc.)</p><p>Nowadays there are public papers published by Google, Facebook, Alibaba. Their frameworks and datasets are open sources. Forums are created with Q&amp;A.</p><p>Large companies are interested in such startups as ours developing, capturing new niches and showing maximum growth. They are happy to buy NLP startups to strengthen their positions.</p><p>After all, even if you have all the datasets and algorithms in your hands — this doesn’t mean you will make a high-quality translator or another startup in the NLP vector. And even if you do, then it’s far from the fact that you get a large piece of the market pie. Therefore, you need help, and if someone succeeds, buy or merge.</p><p>In March 2018, Systran invited the entire community to Paris to exchange experiences, and also hosted a free master class on the main problems that startups in translation usually face. Everyone was interested to see each other in real life.</p><p>Everyone had various projects. Someone created a bot-learning English, which talks to you like a person. Others used Open NMT to summarize the text. A significant part of startups represented plugins for SDL Trados Studio, tailored to a specific topic (medicine, construction, metallurgy, etc.) or language to help translators save time on editing the translated text.</p><p>In addition to enthusiasts, the guys from eBay and Booking arrived in Paris, who created a translator on the same platform as us, but optimized it for auction and hotel descriptions.</p><p>Also in May 2017, Facebook posted its solution for machine translation of Fairseq into open source along with trained models for tests. But we decided to stay on OpenNMT, watching how the community grows.</p><h3>DeepL history</h3><p>In September 2017, while analyzing competitors, I learned a lot about <a href="https://www.deepl.com/"><strong>DeepL</strong></a>. They just started and already provided translation in 7 languages. DeepL was positioned as a tool for professional translators, helping to spend less time on proofreading after machine translation. Even a small quality change saves a lot of money for translation companies. They constantly monitor the API for machine translation from different suppliers, since the quality in many language pairs is different for everyone and there is no single leader. Although by the number of languages, Google has the most.</p><p>To demonstrate the translation quality, DeepL decided to run tests in some languages and published this <a href="https://techcrunch.com/2017/08/29/deepl-schools-other-online-translators-with-clever-machine-learning/"><strong>article</strong></a></p><p>Quality assessment was carried out by blind testing, when professional translators choose the best translation from Google, Microsoft, DeepL, Facebook. According to the results, DeepL won, the jury rated its translation as the most “literary”.</p><h4>How did it happen?</h4><p>Founders of DeepL own <a href="https://www.linguee.com/"><strong>Linguee</strong></a> startup — the largest database of links to translated texts. Most likely, they have a huge number of datasets gained by parsers, and to train them, you need more computing power.</p><p>In 2017, they published an <a href="https://www.top500.org/news/startup-launches-language-translator-that-taps-into-five-petaflop-supercomputer/?fbclid=IwAR3kPvVCYwnEHpIkAVUazy5O649g_K3aKiCFaaFbDyPfuCKyuRCmSX6O8To"><strong>article</strong></a> on how they had launched a 5-petaflops supercomputer in Iceland being 23rd in terms of performance in the world by that time. To train a large-high-quality model was only a matter of time. At that moment, it seemed that even if we buy high-quality datasets, we will still never be able to compete with them without such a super-computer.</p><p>But everything changed in March 2018. Nvidia launched a <a href="https://www.nvidia.com/en-us/data-center/dgx-2/"><strong>DGX-2</strong></a> computer sized nightstand and 2 petFlops (FP16) performance, which can now be leased starting at $5K per month.</p><blockquote>Having such a computer, you can train your models with giant datasets quickly, and also keep a large load on the API. This dramatically changes the power balance of the entire machine learning startup market and allows small companies to compete with giants in the big data field. It was the best offer on the market in a price-performance ratio.</blockquote><p>I started looking for information on <a href="https://slator.com/ma-and-funding/benchmark-capital-takes-13-6-stake-in-deepl-as-usage-explodes/"><strong>DeepL statistics</strong></a>. For 2018, Google had 500M MAU (monthly active users). DeepL has 50M.</p><p>At the end of 2018, 10% of the monthly Google audience used DeepL, and they were not particularly advertised anywhere. In just over a year, they captured 10% of the market using word of mouth.</p><p>I thought about it. If the DeepL team defeated Google with a team of 20 people, having a supercomputer of 5 petaFlops in 2017, and now you can cheaply rent a supercomputer in 2 petaFlops and buy high-quality datasets, how difficult will it be to achieve Google quality?</p><h3>Lingvanex Control Panel</h3><p>(Here are some technical details of how our translator is made. If you are non-technical person , please scroll down to read the rest of the story )</p><p>To quickly deal with translation and stop running tests from the console, a Dashboard was created that allowed us to do all tasks, from preparing and filtering data to deploying translation tests. In the picture below: on the right is a list of tasks and GPU servers on which models are being trained. In the center are the parameters of the neural network, and below are the datasets that will be used for training.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S_545jhVQ7Lsa9dsGwSVMg.jpeg" /></figure><p>Working on a new language began with datasets preparation. We took them from open sources such as Wikipedia, European Parliament, Paracrawl, Tatoeba and others. To reach an average translation quality, 5M translated lines are enough.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PbGtLp9_MGYrpCO377_rUw.jpeg" /></figure><p>Datasets are lines of text translated from one language to another. Then the tokenizer splits the text into tokens and creates dictionaries from them, sorted by the frequency of meeting the token. The token can be either single characters, syllables, or whole words.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bI_z8Lp-2v2iKpXE1TxeLw.jpeg" /></figure><p>After the datasets were uploaded to the database, it turned out they have a lot of words with errors and poor translation. To achieve good quality, they must be strongly filtered. You can also buy already high-quality filtered datasets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NE54-CyauJ9A7-GVzsIVZA.jpeg" /></figure><p>When the language is deployed to the API, you need to set a list of available functions for it (voice recognition, speech synthesis, image recognition, parser of a file, site, etc.)</p><h3>Continue the work</h3><p>In 2018 I spent my time on solving the problem of high-quality translation in the main European languages. I thought I needed another six months for everything to work out. I was limited in resources, only 2 people were involved in data science tasks. It was necessary to move fast. It seemed that the solution to the problem was something simple. I wasn’t satisfied with the translation quality. I had already spent about $450K earned on old translators, and it was necessary to decide on what to do next. Running this project alone and without investment, I realized how many management mistakes I made. But if a decision was made — it’s not the best time to give up!</p><p>I noticed that our community started talking about a new architecture for neural networks — Transformer. Everyone rushed to train neural networks based on this Transformer model and began to switch to Python (Tensorflow) instead of the old Lua (Torch). I decided to try it too.</p><p>We also took a new tokenizer, pre-processed the text, started filtering and marking up the data in a different way, otherwise processing the text after translation to correct errors. The rule of 10K hours worked: there were many steps to the goal, and at some point I realized that the translation quality was already enough to use it in the API for my applications. Each change added 2–4% of quality, which wasn’t enough for the critical mass where people continue to use the product instead of using competitors’ solutions.</p><p>Then we started connecting various tools that allowed us to further improve the quality of translation: named entity recognition, transliteration, specific dictionaries, a system for correcting errors in words. After 5 months of hard work, the quality in some languages ​​became much better and people began to complain less. It was a turning point. You can already sell the software, and since you have your API for translation, you can greatly reduce costs. You can increase sales or the number of users, because your only expense is computing power.</p><p>To train a neural network I needed a good computer. But we saved money. We rented 20 regular computers ( each equipped with a GTX1080 video card) and simultaneously launched 20 simple tests on them through the Lingvanex Control Panel. It took a week for each test, it was a long time. To achieve better quality, you had to run with other parameters that required more resources. We needed a cloud computing and more video cards on one machine. We decided to rent a cloud service Amazon 8 GPU V100 x 4. It was fast, but very expensive. We started the test at night, and in the morning got a bill for $1200. At that time, there were very few rental options for powerful GPU servers besides it. I had to abandon this idea and look for cheaper options. Maybe try to create my own?</p><p>We began to consult with the team and decided that it is possible to make a computer using several powerful GPUs and at a price of up to 10K dollars, which will solve our problems and pay off in a month. Two weeks later, everything was ready.</p><p>At the beginning of 2019, I finally assembled this computer at home and began to conduct many tests, without worrying about what I need to pay for cloud services. I began to notice that the English-Spanish translation is close to the Google translation according to the BLEU metric. The computer buzzed all night, it was impossible to sleep. It was necessary to ensure there were no errors in the console. In the morning, I ran a test to translate 100 sentences with lengths from 1 to 100 words and saw that it turned out to be a good translation, including long sentences. This night has changed everything. I saw the light at the end of the tunnel and realized I can still someday achieve a good translation quality.</p><p>Here you can find our <a href="https://lingvanex.com/demo/"><strong>Translator Demo</strong></a>:</p><h3>Improving application quality</h3><p>Earning money from mobile translator apps with one button and one function, I decided to improve their quality, as well as make a version for Android, Mac OS, Windows Desktop. I was hoping that when I have my translation API, I will finish the apps development to enter other markets. But competitors went much further. Some core functions and features were needed.</p><p>The first thing I decided to do was offline voice translation for mobile applications without Internet access. This was a personal issue. For example, you go to Germany, download only the German package to your phone (400 mb) and get a translation from English into German and vice versa. Internet access abroad could be an issue. Wifi is often not available, slow or otherwise not useable. At the time, in 2017, there were thousands of high-quality translation apps that required an internet connection to use Google API.</p><p><strong>Some technical notes</strong></p><blockquote>Since many had problems with the Lua (Torch) version of OpenNMT due to not popular programming language, the founders transferred the logic of the translate.lua script to the C++ version (CTranslate), which was used for more convenient translation experiments. On the Lua version, it was possible to train models, on the C version, to run. By May 2017, it was already possible to somehow use it as the basis of production for applications.</blockquote><blockquote>We ported CTranslate to work for applications and put it all in open source.</blockquote><blockquote>Here is the link to this thread:</blockquote><blockquote><a href="https://github.com/hunter-packages/onmt">https://github.com/hunter-packages/onmt</a></blockquote><blockquote>Porting CTranslate to different platforms is only the first step. It was necessary to understand how to make offline models of small size and normal quality to work on phones and computers. The first versions of the translation models occupied 2GB in the phone’s RAM, which obviously didn’t work.</blockquote><blockquote>I found guys in Spain with good experience in machine translation projects. For about 3 months, we jointly conducted research in the field of reducing the size of the neural network model for translation, to achieve 150 mb per language pair and then run on mobile phones.</blockquote><blockquote>The size had to be reduced so that in a certain size of the dictionary (for example, 30 thousand words) to embed as many options as possible for translating words of different lengths and topics.</blockquote><p>Later, the result of <a href="https://rua.ua.es/dspace/bitstream/10045/76108/1/EAMT2018-Proceedings_33.pdf?fbclid=IwAR1BxipmZMR8Rt0d32gcJ7BaFt1Tf1UEm9LkJCYytBJLgdtx3ujAPFCwE80"><strong>our research</strong></a> was made publicly available and presented at the European Machine Translation Association in Alicante (Spain) in May 2018, and one of the team members got PhD on it.</p><p>At the conference, many people wanted to buy a product, but only one language pair was ready (English — Spanish). Offline translation on neurons for phones was ready in March 2018, and it was possible to do it in all other languages ​​until the summer. But I hadn’t enough time and money. I had to pause this feature. A year later, I returned to it and completed it.</p><p>In addition to translating text, voice and pictures, we decided to add phone calls translation with transcripts, which competitors didn’t have. We knew that people in different countries often use mobile or landline phones to call support. And for call addresses there was no need to install the app. This function required a lot of time and expenses, so we put it in a separate application. This is how we launched the Phone Call Translator. Here you can check this app.</p><p><a href="https://apps.apple.com/app/apple-store/id1367461351?mt=8"><strong>Phone Call Translator for iOS</strong></a></p><p><a href="https://play.google.com/store/apps/details?id=com.nordicwise.translator_call&amp;referrer=utm_source%3DPH&amp;ref=producthunt"><strong>Phone Call Translator for Android</strong></a></p><p>Translation apps had one problem — they are not used every day. There are not many situations in life when you need to translate daily. But if you study the language, the use of a translator becomes frequent. To learn languages, we created the cards, when words are added to bookmarks on the site through the browser extension or movie subtitles, and then knowledge is consolidated using the mobile chat bot application or the smart speaker application that will check the selected words later. All Lingvanex applications are connected via single account, so you can start translating on a mobile application and continue on your computer.</p><p>Also added voice chats with translation. This will be useful for tourist groups, when the guide can speak their language, and each of the visitors will listen in translation. And finally — the translation of large files on the phone or computer.</p><h3>Business Strategy</h3><p>To create an API for your applications and invest a lot of money, you need to understand the volume and prospects of the machine translation market. In 2017, there was a forecast that the market will become $1.5B by 2023, although the market volume for all translations will be $70B (by 2023).</p><p>Why such a difference — about x50?</p><p>Let’s say the best machine translation now translates 80% of the text well. The remaining 20% ​​must be edited by a person. The biggest expenses in translation are proofreading, that is, people’s salaries.</p><blockquote>Increasing the translation quality even by 1% (up to 81% in our example) can figuratively reduce the cost of proofreading text by 1%. 1% of the difference between the market of all translation minus the machine will be (70–1.5 = $68.5B) or $685M already. The numbers and calculation above are given approximately to convey the essence.</blockquote><p>That is, even a 1% improvement in quality can significantly save large companies on translation services. As the quality of machine translation develops, more and more of it will replace the manual translation market and save on salary costs. It’s not necessary to try to cover all languages, you can choose a popular pair (English-Spanish) and one of the areas (medicine, metallurgy, petrochemicals, etc.). 100% quality — the perfect machine translation on all subjects —is impossible to attain in the short term. And each subsequent percentage of quality improvement will be more difficult.</p><p>However, this doesn’t prevent the machine translation market from occupying a significant part of the total market by 2023 (by analogy with DeepL, it imperceptibly grabbed 10% of the Google market), as large companies test various translator APIs every day. And improving the quality of one of them by a percentage point(for any language) will allow them to save money.</p><p>The large companies’ strategy of open-sourcing their products began to bear fruit. There are more startups, public papers, and people in the industry now, which allowed us to rock the market and achieve better translation quality, increasing our forecast for the machine translation market.</p><p>Every year, competitions are held on NLP tasks, where corporations, startups and universities compete who will have better translation in certain language pairs. By analyzing the list of winners, there is confidence that small resources can achieve excellent results.</p><h3>Company incorporation</h3><p>The project has grown. Applications have appeared not only for mobile platforms but also for computers, wearable devices, instant messengers, browsers, voice assistants. In addition to translating text, a translation of voice, pictures, files, websites and phone calls was created. Initially, I planned to make my translation API to use only for my applications. But then I decided to offer it to everyone. Competitors went ahead and had to keep up.</p><p>Until that time, I managed everything alone as an individual, hiring people to outsource. But the complexity of the product and the number of tasks began to grow rapidly, and it became obvious that you need to delegate functions and quickly hire people to your own team in your office. I called a friend, he quit his job and decided to open the <a href="https://lingvanex.com/"><strong>Lingvanex LLC</strong></a><strong> </strong>office in March 2019.</p><p>Until this very moment I never promoted the project I created. When I decided to assemble my team, I faced a hiring problem. Nobody believed that this could be done at all. I had to interview many people by talking for three hours about thousands of unobvious details. When the first article about the project came out, it became easier. I was always asked one question: “Why are you better than Google?”</p><p>At the moment, our goal is to achieve the quality of Google translation with the main European and Asian languages and after that to provide solutions for:</p><p>1) Translation of text and sites through our API x3 times cheaper than competitors, providing excellent support service and easy integration. For example, the cost of Google translation is $20 per a million characters, which is very expensive at volume.</p><p>2) High-quality thematic translation of documents on certain topics (medicine, metallurgy, law, etc.) by API, including with integration into tools for professional translators (such as SDL Trados)</p><p>3) Integration into the business processes of enterprises to run translation models on their servers under our license. This allows you to preserve data confidentiality, not depending on the volume of the translated text and optimize the translation for the specifics of a particular company.</p><blockquote>You can make translation quality better than competitors for specific language pairs or topics. You can do everything. This is a matter of company resources. With sufficient investment, there is no problem. What and how to do — it’s known that you just need hands and money.</blockquote><p>In fact, the NLP market is growing very rapidly in recognition, speech analysis, machine translation improvement, and can bring good profit for a small team. All hype here will begin in 2–3years when today’s market promotion by large companies will bear fruit. A series of mergers/acquisitions will begin. The main thing at this moment is to have a good product with an audience that you can sell.</p><h3>Summary</h3><p>Over the years I earned about $1M in revenue from simple mobile apps, and spent most of profit to create my own translator. In hindsight it’s now obvious that everything could’ve been done. Many managerial mistakes were made, but this is experience, and then there was no one to consult with. The article describes a very small part of this story, and sometimes it may not be clear why certain decisions were made. Ask questions in the comments below.</p><p>At the moment, we haven’t achieved the quality of Google translation, but I don’t see any problems doing this if the team has at least a few specialists in Natural Language Processing.</p><p>Here is the list of new translation apps that have been developed over 3 years and in which money was invested.</p><p><a href="https://apps.apple.com/app/apple-store/id1254981151?mt=8"><strong>Translator for iOS</strong></a></p><p><a href="https://play.google.com/store/apps/details?id=com.nordicwise.translator&amp;utm_source=site"><strong>Translator for Android</strong></a></p><p><a href="https://apps.apple.com/app/apple-store/id1254982908?mt=8"><strong>Translator for Mac OS</strong></a></p><p><a href="https://www.microsoft.com/en-us/p/lingvanex-home-translator-and-dictionary/9pgwf6lbx4s4?activetab=pivot:overviewtab"><strong>Translator for Windows</strong></a></p><p><a href="https://lingvanex.com/products/extensions/"><strong>Translator for Web Browser</strong></a></p><p><a href="https://lingvanex.com/products/translator-chatbots/"><strong>Translator Bots for Messengers</strong></a></p><p><a href="https://slack.com/signin?redir=%2Foauth%3Fclient_id%3D290017425858.291600150471%26redirect_uri%3D%26state%3D%26scope%3Dapp_mentions%253Aread%252Cchannels%253Ahistory%252Cchannels%253Aread%252Cchat%253Awrite%252Ccommands%252Cfiles%253Aread%252Cfiles%253Awrite%252Cgroups%253Ahistory%252Cgroups%253Aread%252Cim%253Ahistory%252Cim%253Aread%252Cim%253Awrite%252Cmpim%253Ahistory%252Cmpim%253Aread%252Cteam%253Aread%252Cusers%253Aread%26user_scope%3Dchannels%253Awrite%252Cchat%253Awrite%252Cgroups%253Awrite%26granular_bot_scope%3D1%26team%3D%26install_redirect%3D%26single_channel%3D0"><strong>Translator for Slack</strong></a></p><p>If you have any questions, ideas for joint partnerships, investments and offers — add me on <a href="https://www.facebook.com/alexei.rudak">Facebook</a> or <a href="https://www.linkedin.com/in/aliaksei-rudak-42b3a715/">LinkedIn</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54cbd68fa9b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>