[Annotated] How machines learned to speak human language

And what does that mean for the way people use computers ?

This past Christmas, millions of people will have opened boxes containing^[각주:1] gadgets^[각주:2] with a rapidly improving ability to use human language. Amazon's Echo device, featuring a digital assistant called Alexa, is now present in^[각주:3] over 5m homes. The Echo is a cylindrical^[각주:4] desktop computer with no interface apart from^[각주:5] voice. Ask Alexa for the weather, to play music, to order a taxi, to tell you about your commute^[각주:6] or to tell a corny^[각주:7] joke, and she will comply^[각주:8]. The voice-driven^[각주:9] digital assistants from America's computer giants (Google Assistant, Microsoft's Cortana and Apple's Siri) have also vastly improved. How did computers tackle the problems of^[각주:10] human language?

Once, the idea was to teach machines rules - for example, in translation, a set of grammar rules^[각주:11] for breaking down^[각주:12] the meaning of the source language^[각주:13], and another set for reproducing^[각주:14] the meaning in the target language^[각주:15]. But after a burst of optimism^[각주:16] in the 1950s, such systems could not be made to work on complex new sentences; the rule-based approach^[각주:17] would not scale up^[각주:18]. Funding for human-language technologies went into hibernation^[각주:19] for decades, until a renaissance in the 1980s.

In effect, language technologies teach themselves, via a form of pattern-matching. For speech recognition^[각주:20], computers are fed^[각주:21] sound files on the one hand, and human-written transcriptions^[각주:22] on the other. The system learns to predict which sounds should result in what transcriptions. In translation, the training data are source-language texts and human-made translations. The system learns to match the patterns between^[각주:23] them. One thing that improves both speech recognition and translation is a "language model^[각주:24]" - a bank of knowledge about what (for example) English sentences tend to look like. This narrows the systems' guesswork considerably^[각주:25]. Three things have made this approach take a big leap forward ^[각주:26]recently: First, computers are far more powerful. Second, they can learn from huge and growing stores of^[각주:27] data, whether publicly available on the internet or privately gathered by firms. Third, so-called "deep learning", which uses digital neural networks ^[각주:28]with several layers of digital "neurons" and connections between them, have become very good at learning from example.

All this means that computers are now impressively competent at^[각주:29] handling spoken requests that require a narrowly defined reply. "What's the temperature going to be in London tomorrow?" is simple (To be fair^[각주:30], you don't need to be a computer to know it is going to rain in London tomorrow). Users can even ask in more natural ways, such as, "Should I carry^[각주:31] an umbrella to London tomorrow?" (Digital assistants learn continually from the different ways people ask questions.) But ask a wide-open question ("Is there anything fun and inexpensive to do in London tomorrow?") and you will usually just get a list of search-engine results. As machine learning improves, and as users let their gadgets learn more about them specifically, such answers will become more useful. This has implications that trouble privacy advocates, but if the past few years of mobile-phone use are any indication^[각주:32], consumers will be sufficiently delighted by^[각주:33] the new features to make the trade-off^[각주:34].

1. box 안에 구성 요소를 설명하려 contain 이라는 동사를 활용한 것을 확인 [본문으로]
2. gadget ; 전자기기용품들을 표현하는 단어 [본문으로]
3. be present (in) ; ~에 있다, 존재하다 [본문으로]
4. cylindrical ; 원통[실린더]형의 [본문으로]
5. apart from ; [전치사] …외에는, …을 제외하고 [본문으로]
6. commute ; [명사] 통근 (거리) [본문으로]
7. corny ; [형용사] corn・ier , corni・est (비격식) 진부한 [본문으로]
8. comply ; ~ (with sth) (법・명령 등에) 따르다[준수하다] 참고 compliance [본문으로]
9. voice-driven ; 음성 인식의 [본문으로]
10. tackle a problem ; 문제를 다루다, 문제와 씨름하다 [본문으로]
11. grammar (rules) ; 문법 규칙 [본문으로]
12. break down ; 3. ~을 나누다[분류하다] ; 세분화하여 분석하다 [본문으로]
13. source language ; [언어] 기점(起點) 언어 ((번역되기 전의 원문 언어)); [컴퓨터] 원시[기본] 언어 [본문으로]
14. reproduce ; 2. [타동사][VN] 다시 만들어 내다, 재생[재현]하다 [본문으로]
15. target language ; (번역에서) 목표어(번역의 결과물 언어) [본문으로]
16. a burst of optimism ; (별안간) 생겨난 낙관주의 [본문으로]
17. rule-based approach ; 규칙을 기반으로한 접근 [본문으로]
18. scale up ; (크기·규모를) 확대하다[늘리다] [본문으로]
19. go into hibernation ; 동면에 들어가다 [본문으로]
20. speech recognition ; 음성 인식 [본문으로]
21. be fed ; 전달받다, 제공받다 // feed 5. GIVE ADVICE/INFORMATION | ~ sb sth | ~ sth to sb (충고・정보 등을) 주다 [본문으로]
22. transcription ; [명사] (구술된 내용 등을[의]) 글로 옮김[표기/인쇄(하기)] [본문으로]
23. match ; 3. FIND STH SIMILAR/CONNECTED | [타동사][VN] ~ sb/sth (to/with sb/sth) (관련 있거나 비슷한 것을) 연결시키다 [본문으로]
24. language model ; 언어 모형, 언어모델, 언어의 모델 [본문으로]
25. guesswork ; 짐작, 추측 [본문으로]
26. take a leap ; 도약하다 [본문으로]
27. a store of ; 축적된 어느 정도의 [본문으로]
28. neural networks ; [명사] (컴퓨터) 신경망 [본문으로]
29. competent at ; …에 있어서 유능한. [본문으로]
30. to be fair ; 공정하게 말하면 [본문으로]
31. 우산을 챙기다 라는 표현을 carry 동사를 이용해 표현 [본문으로]
32. be an indication ; 징조, 징표, 징후가 되다 [본문으로]
33. be delighted by / with / at ; 반색을 하다 [본문으로]
34. trade-off ; ~ (between sth and sth) (서로 대립되는 요소 사이의) 균형 [본문으로]