2023-05-29期ChatGPT的語言方式引發思考：人類如何習得語言（PDF版+Word版+音頻）

每日一篇|英語外刊社

(資料圖片僅供參考)

Johnson：The languageinstinctChatGPT’s way with words raises questions about how humansacquirelanguage【1】When deep blue, a chess computer,defeatedGarry Kasparov, a worldchampion, in1997many gasped in fear of machines triumphing over mankind. In theinterveningyears,artificialintelligence has done someastonishingthings, but none has managed tocapturethe publicimaginationin quite the same way. Now, though, theastonishmentof the Deep Blue moment is back, because computers areemployingsomething that humans consider theirdefiningability: language.【2】Or are they? Certainly, large language models (LLMS), of which the most famous is ChatGPT, produce what looks likeimpeccablehuman writing. But a debate hasensuedabout what the machines are actually doinginternally, what it is that humans, in turn, do when they speak—and, inside theacademy, about the theories of the world’s most famouslinguist, Noam Chomsky.【3】Although Professor Chomsky’s ideas have changedconsiderablysince he rose toprominencein the1950s, severalelementshave remainedfairlyconstant. He and his followers argue that human language is different in kind (not just degree of expressiveness) from all other kinds ofcommunication. All human languages are more similar to each other than they are to, say,whalesong or computer code. Professor Chomsky has frequently said a Martian visitor wouldconcludethat all humans speak the same language, withsurfacevariation.【4】Perhaps most notably, Chomskyan theories hold that children learn their native languages withastonishingspeed andeasedespite“thepovertyof thestimulus”: thesloppyandoccasionallanguage they hear in childhood. The onlyexplanationfor this can be that some kind of predisposition for language is built into the human brain.【5】Chomskyan ideas have dominated thelinguisticfield of syntax since their birth. But manylinguistsarestridentanti-Chomskyans. And some are nowseizingon thecapacitiesof llms to attack Chomskyan theories anew.【6】Grammar has ahierarchical,nestedstructureinvolving units within other units. Words formphrases, which formclauses, which formsentencesand so on. Chomskyan theorypositsa mental operation, “Merge”, which glues smaller units together to form larger ones that can then beoperatedon further (and so on). In a recent New York Times op-ed, the man himself (now94) and two co-authors said “we know” that computers do not think or use language as humans do, referringimplicitlyto this kind ofcognition. LLMS, in effect, merelypredictthe next word in a string of words.【7】Yet it is hard, for several reasons, tofathomwhat LLMS “think”. Details of theprogrammingand training data ofcommercialones like ChatGPT areproprietary. And not even theprogrammersknow exactly what is going on inside.【8】linguistshave, however, found clever ways to test LLMS’underlyingknowledge, in effecttrickingthem withprobingtests. And indeed, LLMS seem to learnnested,hierarchicalgrammaticalstructures, even though they areexposedto onlylinearinput, ie,stringsof text. They can handle novel words and grasp parts of speech. Tell ChatGPT that “dax” is a verb meaning to eat asliceof pizza byfoldingit, and the system deploys it easily: “After a long day at work, I like to relax and dax on asliceof pizza while watching my favourite TV show.” (Theimitativeelement can be seen in “dax on”, which ChatGPT probablypatternedon the likes of “chew on” or “munchon”.)【9】What about the “povertyof thestimulus”? After all, GPT-3(the LLMunderlyingChatGPT until the recentreleaseof GPT-4) isestimatedto be trained on about1,000times the data a human ten-year-old isexposedto. That leaves open the possibility that children have aninborntendencyto grammar, making them far moreproficientthan any LLM. In aforthcomingpaper inlinguisticInquiry, researchers claim to have trained an LLM on no more text than a human child isexposedto, finding that it can use even rare bits of grammar. But other researchers have tried to train an LLM on a database of only child-directed language (that is, oftranscriptsofcarersspeaking to children). Here LLMSfarefar worse. Perhaps the brain really is built for language, as Professor Chomsky says.【10】It is difficult to judge. Both sides of the argument are marshalling LLMS to make their case. Theeponymousfounder of his school of linguistics has offered only abrusqueriposte. For his theories to survive this challenge, his camp will have to put up a stronger defence.

①短語：1.原文：When deep blue, a chess computer, defeated Garry Kasparov, a world champion, in 1997 many gasped in fear of machines triumphing over mankind.

詞典: in fear of 害怕；擔憂

triumph over打??；戰勝

例句：We lived in constant fear of losing our jobs.

我們一直生活在擔心失去工作的陰影里。

Working side by side, we have the ability to solve the most insurmountable problems and to triumph overthe greatest of adversities.

只要攜手合作，我們就有能力解決最難以克服的問題，戰勝最大的逆境。

2.原文：Although Professor Chomsky’s ideas have changed considerably since he rose to prominence in the 1950s, several elements have remained fairly constant.

詞典: rise to prominence崛起；聲名鵲起

例句：As she rises to prominencein the international world of chess, she struggles with alcoholism and addiction.

隨著她在國際象棋界嶄露頭角，她與酗酒和毒癮作斗爭。

3.原文：And some are now seizing on the capacities of llms to attack Chomskyan theories anew.

詞典: seize on對…大為關注；抓住（可利用的事物）

例句：Newspapers seized onthe results as proof that global warming wasn"t really happening.

各報紙紛紛以此結果為證據來證明全球變暖并沒有真正發生。

4.原文：The imitative element can be seen in “dax on”, which ChatGPT probably patterned on the likes of “chew on” or “munch on”.

詞典: pattern... on...模仿；仿效 （被動形式：be patterned on）

例句：The clothing is patterned onathletes" wear.

這些衣服是仿照運動員的穿著制作的。

②長難句

1.原文：Now, though, the astonishmentof the Deep Blue moment isback, because computers are employing something thathumans consider their defining ability: language.

分析：本句包含一個原因狀語從句和一個定語從句。主句為“the astonishmentof the Deep Blue moment isback”；though在本句中為副詞，表示“然而”；because為連詞，引導原因狀語從句“computers are employing something”；“thathumans consider their defining ability: language”為限制性定語從句，修飾something。

譯文：但現在，“深藍”那樣震撼的時刻又回來了，因為計算機正在運用人類認為是其決定性能力的東西：語言。

2.原文：Certainly, large language models(LLMS), of whichthe most famous is ChatGPT, producewhatlooks like impeccable human writing.

分析：本句包含一個非限制性定語從句和一個賓語從句。主句為“large language modelsproduce...”；“of whichthe most famous is ChatGPT”為非限制性定語從句，which指代LLMS；“whatlooks like impeccable human writing”為賓語從句。

譯文：當然，大型語言模型（LLMS）能像人類一樣寫作且無可挑剔，其中最有名的就是ChatGPT。

③寫作技巧：

Here LLMS fare far worse.

在這種情況下，大型語言模型表現要差得多。

生詞: fare n.旅費；路費；車費 v.進展；進行

fare這個詞常見的意思是“費用”，在本文中為熟詞僻義，表示“情況如何；表現如何”，可以替換perform。fare常和well/badly搭配，表示“情況好/情況不好”，其比較級為fare better/worst。

例句：It is hard to categorize about how many hours should be spent on everyday learning. What suffices for able students may be inadequate for those who fare worse.

很難準確地說每天需要學習多少小時。同樣的學習時間，對于聰明的學生來說夠了， 對于成績落后的學生來說可能不夠。

④背景知識：

Blue：深藍是一臺專門為國際象棋比賽打造的超級計算機，美國IBM公司的研究小組從1989年就開始對其進行開發，并經歷了多次升級和改良。深藍的主要特色在于它能在每秒鐘內運算超過兩億種走法，并從中篩選出最優解。同時，它也能利用大量的國際象棋數據庫，吸收并學習人類大師的經驗和技巧。1997年5月11日，那是一場人類智慧與機械智力之間較量的終場之日，落錘之地在美國紐約。深藍超級電腦在一場六局的對決中，以.5的總分戰勝了國際象棋大師——世界冠軍加里·卡斯帕羅夫，此戰成為了人工智能歷程中的一個重要時刻。它向我們展示了機器的可能性，同時也讓我們對未來充滿了期待和想象。

language models (LLM)：大型語言模型是一種人工智能模型，旨在理解和生成人類語言。它們在大量的文本數據上進行訓練，可以執行廣泛的任務，包括文本總結、翻譯、情感分析等等。LLM的特點是規模龐大，包含數十億的參數，幫助它們學習語言數據中的復雜模式。LLM是通常源自Transformer架構的Al模型，旨在理解和生成人類語言、代碼等。這些模型在大量文本數據上進行訓練，使它們能夠捕捉人類語言的復雜性和細微差別。LLM可以執行范圍廣泛的語言任務，從簡單的文本分米到文本生成，且有很高的準確性，流暢性和風格。在醫療保健行業，LLM被用于電子病歷處理、臨床試驗匹配和藥物發現。在金融領域，LLM被用干欺詐檢測，金融新聞的情緒分析，其至交易策略。憑借其多功能性和高性能的特性，基于Transformer的LLM正在成為各種行業和應用程序中越來越有價值的資產。

段落大意：

【1】計算機正在運用人類的決定性能力：語言。

【2】大型語言模型能和人類一樣寫文章，引發了諸多爭論。

【3】喬姆斯基教授表示，所有人類語言之間具有相似性。

【4】喬姆斯基理論認為，人類大腦中存在某種語言傾向。

【5】有些人正在利用大型語言模型的能力攻擊喬姆斯基理論。

【6】大型語言模型和人類的認知方式并不相同。

【7】我們很難理解大型語言模型的“想法”。

【8】語言學家利用探究性測試來了解大型語言模型的基礎知識。

【9】《語言學探索》的一篇論文對“刺激貧乏”給予說明。

【10】對于爭論，喬姆斯基學派必須提出更強有力的辯論觀點。

PS:各位研友下載請文末點擊閱讀原文.

考研英語雜志經濟學人英文外刊|2022年經濟學人英文雜志（2022年經濟學人周刊英雜志已更新完畢）

366外刊社每日分享英文雜志,考研英語雜志,考研英語外刊雙語精讀,經濟學人雜志,英文雜志下載。

文件格式:True PDF ，①.下載方式:網盤下載②.是否支持編輯:支持③.文字內容:支持選取復制★能否打印:支持打印★

戳“閱讀原文”下載外刊精讀PDF Doc Mp3