It’s another beautiful Wednesday.
Hello everyone!
Siraj’s machine learning series is back!
Do you still remember last week’s discussion about “Voice assistants can analyze the same intent from different language expressions”?
NLP
In essence, whether it’s a voice assistant or speech recognition, it first converts speech into text, a process known as speech recognition. After converting speech into text, AI then performs “reading comprehension” on the text, ultimately identifying the intent hidden within the text.
Thus, behind every powerful voice assistant is a diligent AI reader. The core of the AI reader is the “natural language processing technology”, commonly referred to as NLP (Natural Language Processing).
NLP has a very wide range of applications. Search engines rely on NLP to understand the exact meaning of your queries, thus providing you with more accurate information. The Weibo platform uses NLP technology to analyze users’ tweets and comments, extracting the most popular topics at the moment. The company DuerOS utilizes NLP technology for in-car voice recognition, allowing you to listen to music and navigate without needing to operate manually.
Ray Kurzweil, Google’s Director of Technology, once said: “Language is the key to AI; a computer that can communicate indistinguishably with humans is true artificial intelligence.” The history and culture of humanity are recorded in our languages and texts; if AI can understand these languages and texts, it builds a bridge connecting AI and the human world.
Tokenization
There are 6,500 known languages in the world, each with its own unique grammar and rules. If we were to handle language solely through programming logic, it would create an immense workload that would be practically impossible to achieve in a reasonable timeframe.
So how can we find patterns in the vast expanse of languages? It’s time for machine learning to showcase its power.
Natural language processing issues are much more complex than general machine learning problems. This is because we cannot simply annotate each word’s meaning to derive the overall meaning of a sentence. For example, which direction does “Beijing West Station South Square East” refer to?
To solve such problems, we first need to label each word’s “part of speech” in the sentence, marking out all the “nouns”, “verbs”, “pronouns”, “adverbs”, etc. However, simply labeling each word’s part of speech is not enough; a piece of text often has multiple possible syntactic structures. A sentence containing 20-30 words can have hundreds of possible interpretations.
Syntax Parser
Therefore, we need a syntax parser to find all possible syntactic structures in a sentence and select the one with the best validity. In this episode, we will use Google’s “Parsey McParseface” as our syntax parser. It processes the relationships between words from left to right, establishing “relationship connection arcs” between the words.
At the end of the entire processing, a “root word” is generated. In the sentence shown above, the generated root word is “booked”. We can also obtain words that are related to the root word, such as “ticket”. By finding the “root word” and analyzing the connections between words, we can deduce the intent of the entire sentence. For example, in the application of voice assistants, for the phrase “I want to take a taxi to Sanlitun”, we can analyze that the root word is “take a taxi”, and the related word is “Sanlitun”. By identifying the user’s “action” and “purpose”, the voice assistant can provide services to the user.
If you want to see Siraj’s more interesting explanations and detailed operational steps, please click on the video.
[Recommended to watch under WI-FI]
Previous series videos:
Learn your first machine learning app in four minutes
How to build an AI composer? Continue learning new skills with Siraj
Learn this skill in five minutes; is it to put professional gamers out of work?
What movie do you want to watch? AI knows better than you—create a movie recommendation system in 5 minutes
Can programmers also hold art exhibitions? Learn how to create an AI painter in five minutes
Step-by-step guide to creating a chatbot
#Translation Group Management Operation Team Recruitment#
The Intelligent Translation Group was established in June 2016. In just over six months, nearly a hundred enthusiasts of translation and science have participated in the translation, completing:
* 92 abstracts of cutting-edge articles selected from Complex Digest;
* Two interviews with the author and translator of “The Three-Body Problem”;
* The translation of cutting-edge articles “How artificial intelligence is changing economic theory” and “big data fades to the algorithm economy”;
* Two chapters of “Computational Social Science: Discovery and Prediction” translated;
* The first chapter of “Pattern and Repertoire in History” totaling 54 pages translated.
Even after completing so many articles, we still have strong fighting power, and many participants want to join the translation activities. In 2017, we will have three groups focusing on subtitle translation, literature translation, and book translation, all working simultaneously, so everyone no longer has to worry about not having translation work during their free time.
To ensure the smooth completion of our translation work, we are currently recruiting members for the translation group management operation team, so we can collaborate to promote the translation work.
Joining the management operation team:
* You will have access to first-hand quality articles and translations;
* You will also have more opportunities to communicate with our expert teachers and industry leaders in artificial intelligence;
* You will have the chance to become a member of our Intelligent Club volunteer team.
* Outstanding translators will receive memorabilia from Intelligent (books or commemorative T-shirts or other items, even cash rewards).
* You will enter a rapidly ascending platform.
* You will meet like-minded partners and gain friendships and even love.
As long as you have one of the following skills, you can join:
* Solid English foundation with listening and translation skills;
* Video processing skills;
* Proficient in image editing and layout;
* Project management skills;
* Strong interest in the field of artificial intelligence.
Note: The workflow is clear and straightforward; completing small tasks regularly will not take up too much personal time.
Welcome to all science-loving and translation-loving friends to scan the QR code above to fill out the form. We look forward to your joining!
Click the link below to read the original text and learn about must-read books on artificial intelligence.