文/ 本·迪克森 譯/王越凡 審訂/張瓊

By Ben Dickson

Professional writing isn’t easy. As a blogger, journalist or reporter, you have to meet several challenges to stay at the top of your trade. You have to stay up to date with the latest developments and at the same time write timely,compelling and unique content.

[2] The same goes for scientists,researchers and analysts and other professionals whose job involves a lot of writing.

[3] With the deluge1deluge洪水，暴雨。of information being published on the web every day,things aren’t getting easier. You have to juggle speed, style, quality and content simultaneously if you want to succeed in reaching your audience.

[4] Fortunately, Artificial Intelligence,which is fast permeating2permeate滲透。every aspect of human life, has a few tricks up its sleeve to boost the efforts of professional writers.

專業(yè)寫作并非易事。無論是博主、新聞工作者，還是通訊記者，要維持自己在行業(yè)內(nèi)的權(quán)威優(yōu)勢(shì)，都需要面對(duì)許多挑戰(zhàn)。出色的專業(yè)寫作者必須與時(shí)俱進(jìn)，并能及時(shí)推出有力、獨(dú)特的寫作內(nèi)容。

[2]科學(xué)家、研究者、分析師和其他需要大量寫作的專業(yè)人士亦如此。

[3]如今，每天都有大量的信息發(fā)布在網(wǎng)上，專業(yè)寫作面臨的挑戰(zhàn)和壓力更甚。如果想成功贏得讀者關(guān)注，就必須同時(shí)兼顧速度、風(fēng)格、質(zhì)量和內(nèi)容。

[4]幸運(yùn)的是，人工智能正迅速滲透到人類生活的方方面面，為專業(yè)寫作者提供各種錦囊妙計(jì)。

智能校對(duì)

[5] 2014年，喬治·R. R.馬丁，著名的《冰與火之歌》作者，在一次采訪中解釋了他如何拒絕現(xiàn)代文字處理軟件，擺脫其惱人的自動(dòng)更正和拼寫檢查功能。

軟件供應(yīng)商一直試圖在軟件中添加校對(duì)功能來幫助寫作者。但是像馬丁這樣的作者會(huì)證明，對(duì)任何寫作能力較強(qiáng)的人而言，這些干涉令人憎惡。

[6]然而，隨著人工智能理解書面文本語境和作者意圖的水平不斷提高，上述現(xiàn)象正在發(fā)生改變。其中一例便是微軟文檔全新的“編輯”（Editor）功能，這項(xiàng)運(yùn)用人工智能的工具能提供的不止簡(jiǎn)單校對(duì)。

相較于代碼邏輯工具，“編輯”功能可以更好地理解文本中的細(xì)微差別。它不僅能標(biāo)出語法和語言規(guī)范方面的錯(cuò)誤，還能標(biāo)出不必要的復(fù)雜用詞和濫用的詞語。比如，當(dāng)你用“真的”一詞時(shí)，它能辨別你是在強(qiáng)調(diào)觀點(diǎn)還是提出疑問。

[7]當(dāng)它辨識(shí)出某一處錯(cuò)誤時(shí)，會(huì)就這一判斷給出有力說明，并提供智能建議。例如，如果它標(biāo)注出一個(gè)被動(dòng)語態(tài)的句子，就會(huì)提供另一個(gè)主動(dòng)語態(tài)的改寫版句式。

Smart proofreading

[5] In 2014, George R. R. Martin,the acclaimed3acclaimed著名的。writer of the Song of Ice and Fire saga, explained in an interview how he avoids modern word processors because of their pesky autocorrect and spell checkers.

Software vendors have always tried to assist writers by adding proofreading features to their tools. But as writers like Martin will attest, those efforts can be a nuisance4nuisance麻煩事，討厭的人。to anyone with morethan-moderate writing skills.

[6] However, that is changing as AI is getting better at understanding the context and intent of written text.One example is Microsoft Word’s new Editor feature, a tool that uses AI to provide more than simple proofreading.

Editor can understand different nuances5nuances細(xì)微差別。in your prose much better than code-and-logic tools do. It flags not only to grammatical errors and style mistakes, but also the use of unnecessarily complex words and overused terms. For instance, it knows when you’re using the word “really” to emphasize a point or to pose a question.

[7] It also gives eloquent6eloquent雄辯的。descriptions of its decisions and provides smart suggestions when it deems something as incorrect. For example if it marks a sentence as passive, it will provide a reworded version in active voice.

[9] Nonetheless AI-powered writing assistance is fast becoming a competitive market. Grammarly, a freemium grammar checker that installs as a browser extension, uses AI to help with all writing tasks on the web. Atomic Reach is another player in the space, which uses machine learning to provide feedback on the readability of written content.

[8]盡管“編輯”功能尚不完美，但慣用被動(dòng)語態(tài)的專業(yè)作家們已經(jīng)很好地接受了它。

[9]盡管如此，人工智能寫作輔助軟件正在迅速形成一個(gè)頗具競(jìng)爭(zhēng)力的市場(chǎng)。Grammarly是一款免費(fèi)增值語法檢測(cè)工具，可以安裝為瀏覽器擴(kuò)展，它運(yùn)用人工智能協(xié)助所有在線寫作。Atomic Reach是該領(lǐng)域中的另一家公司，它利用機(jī)器學(xué)習(xí)就寫作內(nèi)容的可讀性提供反饋。

書面文件的快速瀏覽

[10]良好的書寫內(nèi)容依托于良好的閱讀。在打開文字處理軟件奮筆疾書之前，我通常喜歡瀏覽一些對(duì)同一話題有著截然不同觀點(diǎn)的文章。問題在于，文章太多，而閱讀時(shí)間太少。當(dāng)你試著尋找同一話題在不同文章中的重點(diǎn)和差異時(shí)，閱讀就會(huì)變得乏味無趣。

[11]如今，人工智能正在以提供智能摘要的方式進(jìn)入閱讀領(lǐng)域。Salesforce公司的研究人員研發(fā)了一種人工智能算法，這種算法能夠生成描述長(zhǎng)文精髓的小段文本。盡管文本總結(jié)工具早已存在，但Salesforce運(yùn)用機(jī)器學(xué)習(xí)的方法超越了其他類似工具。這一系統(tǒng)把監(jiān)督學(xué)習(xí)和強(qiáng)化學(xué)習(xí)結(jié)合在一起，由此可以從人類訓(xùn)練者那里獲得幫助，并學(xué)會(huì)自己生成總結(jié)。

Quicker scanning of written documents

[10] Writing good content relies on good reading. I usually like to go through different articles describing conflicting opinions about a topic before I fire up my word processor. The problem is there’s so much material and so little time to read all of it. And things tend to get tedious7tedious單調(diào)乏味的。when you’re trying to find key highlights and differences between articles written about a similar topic.

[11] Now, Artificial Intelligence is making inroads in the field by providing smart summaries of documents. An AI algorithm developed by researchers at Salesforce generates snippets of text that describe the essence of long text.Though tools for summarizing texts have existed for a while, Salesforce’s solution surpasses others by using machine learning. The system uses a combination of supervised and reinforced learning to get help from human trainers and learn to summarize on its own.

[12]其他算法，如 Algorithmia公司的Summarizer，為軟件開發(fā)者提供了文獻(xiàn)資料庫，這些資料庫輕輕松松就能將文本摘要功能融合到軟件中。

這些工具可以幫助作者瀏覽大量文章，找到寫作的相關(guān)話題。也可以幫助編輯閱讀每天收到的大量電子郵件、宣傳單和新聞稿，幫助他們篩選出需要進(jìn)一步關(guān)注的郵件。收件箱里積壓的數(shù)百封未讀郵件讓我對(duì)此功能尤為心儀。

[13]自然語言處理（NLP）領(lǐng)域的進(jìn)步為這一趨勢(shì)的發(fā)展起到了不小的推動(dòng)作用。NLP幫助機(jī)器理解文本的大意以及不同元素和實(shí)體間的關(guān)系。

[14]平心而論，只有接近人類水平的智能才具備精確總結(jié)所需的常識(shí)性知識(shí)和語言水平。這項(xiàng)技術(shù)還有許多問題需

要解決，但它多少展示了未來的閱讀圖景。

[12] Other algorithms such as Algorithmia’s Summarizer provide developers with libraries that easily integrate8integrate使融入，使成為一體。text summary capabilities into their software.

These tools can help writers skim through a lot of articles and find relevant topics to write about. It can also help editors to read through tons of emails, pitches and press releases they receive every day. This way they’ll be better positioned to decide which emails need further attention. Having hundreds of unread emails in my inbox, I fully appreciate the value this can have.

[13] Advances in Natural Language Processing have contributed widely to this trend. NLP helps machines understand the general meaning of text and relations between different elements and entities.

[14] To be fair, nothing short of human level intelligence can have the commonsense knowledge and mastery of language required to provide fl awless summary of all text. The technology still has more than and few kinks to iron out9iron out解決。, but it shows a glimpse of what the future of reading might look like.

Smarter search engines, content-writing robots and beyond

[15] No matter how high-quality and relevant your content is, it’ll be of no use if you can’t reach out to the right audience. Unfortunately, old keywordbased search algorithms pushed online writers toward stuffing their content with keywords.

[16] “Although with Page Rank, Google did a great job in organizing the web, it also created a web where keywords ruled over content,” says Gennaro Cuofano,growth hacker at WordLift, a company that develops tools for semantic web.“Eventually, web writers ended up spending a significant amount of time improving the findability.” The trend resulted in poor quality writing getting higher search ranking.

[17] But thanks to Artificial Intelligence, search engines are able to parse and understand content, and the rules of search engine optimization have changed immensely in past years.

[18] “Since new semantic technologies are now mature enough to read human language, journalists and professional writers can finally go back to writing for people,” Cuofano says. This means you can expect more quality content to appear both on websites and search engine results.

更智能的搜索引擎，內(nèi)容寫作機(jī)器人及其他

[15]如果你無法找到合適的受眾，那么不論你的寫作內(nèi)容質(zhì)量多高、有多大意義，都毫無作用。不幸的是，過去基于關(guān)鍵詞搜索的算法迫使網(wǎng)絡(luò)寫作者在他們的內(nèi)容中填滿關(guān)鍵詞。

[16]“雖然谷歌利用Page Rank在網(wǎng)絡(luò)組織方面成績(jī)斐然，但它同時(shí)建構(gòu)了一個(gè)關(guān)鍵詞凌駕于內(nèi)容之上的網(wǎng)絡(luò)，”開發(fā)語義網(wǎng)絡(luò)工具的公司W(wǎng)ordLift的產(chǎn)品經(jīng)理金納羅·科法諾認(rèn)為，“網(wǎng)絡(luò)作家最終把大量時(shí)間花費(fèi)在提高檢索度上?！边@種趨勢(shì)導(dǎo)致了質(zhì)量低下的作品搜索排名卻更靠前的現(xiàn)象。

[17]幸好有了人工智能，搜索引擎能夠分析語法并理解內(nèi)容，其優(yōu)化規(guī)則也在過去幾年中發(fā)生了巨大改變。

[18]“由于新的語義技術(shù)已經(jīng)成熟到讀懂人類語言，新聞工作者和專業(yè)作家終于可以重新為大眾寫作了?！笨品ㄖZ說。這就意味著你可以在網(wǎng)站和搜索引擎的結(jié)果中讀到更多高質(zhì)量的內(nèi)容。

[19] Where do we go from here?“The next revolution (which is already coming) is the leap from NLP to a subset of it called NLU (Natural Language Understanding),” Cuofano says. “In fact, while NLP is more about giving structure to data, defining it and making it readable by machines;NLU instead is about taking unclear,unstructured and undefined inputs and transforming them to an output that is close to human understanding.”

[20] We’re already seeing glimmers of this next generation in AI-powered journalism. The technology is still in its infancy, but will not remain so indefinitely. Writing can someday become a full-time machine occupation,just like many other tasks that were believed to be the exclusive domain of human intelligence the past.

[21] How does this affect writing?“Currently, the web is a place where how-to articles, tutorials and guides are dominant,” Cuofano says. “This makes sense in an era where people are still in charge of most tasks. Yet in a future where AI takes over, wouldn’t it make more and more sense to write about‘why’ we do things? Thus, instead of focusing on content that has a short shelf life, we can focus again on content that has the capability to outlive us.” ■?

[19]我們將走向何方？“下一場(chǎng)革命（即將到來）是從NLP到它的一個(gè)子集NLU（自然語言理解）的飛躍?！笨品ㄖZ說，“事實(shí)上，NLP更多的是構(gòu)建數(shù)據(jù)、定義數(shù)據(jù)，使它能夠被機(jī)器讀?。欢鳱LU則是把含混、松散和未定義的輸入轉(zhuǎn)換為便于人類理解的輸出?！?/p>

[20]我們已經(jīng)在人工智能驅(qū)動(dòng)的新聞寫作中看到了新一代技術(shù)的曙光。這項(xiàng)技術(shù)仍處于初期階段，但絕不會(huì)停滯不前。終有一天，寫作會(huì)和許多過去被認(rèn)為人類智能專屬的領(lǐng)域一樣，由機(jī)器全職替代。

[21]這會(huì)對(duì)寫作產(chǎn)生什么樣的影響呢？“目前，網(wǎng)絡(luò)被指導(dǎo)性文章、教程和指南所占據(jù)，”科法諾說，“在人類負(fù)責(zé)大部分工作任務(wù)的時(shí)代，這個(gè)現(xiàn)象合情合理。然而，在人工智能接管的未來，關(guān)于‘為何’的寫作不是會(huì)越有意義嗎？因此，我們可以再次專注于恒久傳世而非短暫易逝的內(nèi)容?！?□