シャドーイング練習: Living with Machines: The language of mechanisation - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

各モードは全画面の練習ルームで開きます。

54 文

文が短すぎたり長すぎる場合は、Editをタップして調整してください。

The written word is the closest thing we have to accessing people's feelings from the past.

Using methods from the field of computational linguistics, we're analyzing the way that people wrote about technological change in the long 19th century.

We were very interested in language models and language modeling as a technique, firstly, because when we started the project,

especially now, it's a very hotly debated topic in data science but now also widely beyond it.

A language model, a neural language model, is a probability distribution over the words, over language.

Language models help you identifying what is the most likely word you are expected to see.

A language model tries to predict what comes next in a sequence.

These language models tend to represent the language used as it is used in a certain society or time.

So most language models are trained on contemporary data, data today.

They don't actually know anything about the past.

It was very important to have language models trained on 19th century data

because 19th century language and 19th century society is so different from today's society and today's way of speaking.

And there's a lot of discussion in natural language processing

and in general in the whole community about language modeling due to the polarization of chat GPT

and this is a specific example in which we show you

that these things rely on the data that have been trained on.

So in many ways what we're interested in is actually

related to the issue of model bias and so what the model predicts reflects often societies or

prejudices but also in general not always negative but also just like attitudes in general in a society.

The overarching question was, is there a relation between the deep changes in society in the time following

or during the so -called

Industrial Revolution in Great Britain and the way language was changing?

So we started this sub -project called Atypical Animacy,

where we wanted to see to what extent and machines were talked about as if they were animate beings.

So we devised this computational method that basically asks the computer, given this sentence containing the word machine, we mask the word machine,

and we ask the computer, can you guess what the word that is missing is?

And what we found out is that if we hide the word machine in a sentence, sometimes a language model will predict a living entity.

So actually where this model made like a wrong prediction, this was actually very interesting for us.

Because it's assumed there was a human, but actually there was a machine in this sentence.

For example, some professions like suing, so if you would have like, there were like many suing machines.

If you would hide the word machine in this sentence, a Victorian language model would count there were many suing girls

or women because this work was mostly done by these groups at this time.

There is a kind of interesting irony in this work

because we're using a tool which many people now take to be as

if it were alive to investigate this particular phenomenon from history in which people in the past also made the same leap,

which is imagining that the machines around them were also alive.

But a language model is no more alive than a steam engine was alive in the 19th century.

So for example, ChatGPT is a very hot topic at the moment

and everyone talks about it and some people have started questioning whether it has intelligence and there's a whole debate around this.

And if we look back in time,

people were having similar conversations and similar feelings of threat even towards machines.

And I think these kind of discoveries were only possible thanks to large -scale computational methods instead of like closed reading approaches.

And from the point of view of analysing the language,

we have a lot to learn from the amount of historical texts that we have from the long 19th century.

Better understanding the past can really help us be better prepared to understand the present and the future.

Computer scientists have a particular sense of humour and in the ongoing work around the language model we have used,

which is based on an architecture called BERT,

it's been an ongoing joke to change the name of the language model according to what it's been trained on.

The French version of BERT is called Camembert and so on.

And the language model

that we trained as part of the project was named by a colleague of mine, Blurt out

which I think is a good thing because it just shows you in a nice,

succinct way that this language model was trained on the BL collections, and that's what it can be used for to understand those books.

And it reminds you all the time of the specific relation between the model and the material it's trained on.

アプリをダウンロード

流暢に話すために必要なすべて

AI発音採点文ごとに採点

IPA練習すべての音をマスター

単語学習自分の単語帳を作る

単語ゲーム遊びながら覚える

QRをスキャン

Download on theApp Store Get it onGoogle Play

FOR YOU

このレッスンの語彙とスピーキングのポイント

このスピーキングレッスンは、動画「Living with Machines: The language of mechanisation」を教材にしています。繰り返し出てくる語は次のとおりです：language, model, machine, word, trained。この動画には、シャドーイング用の文が54文、単語が833語あります。音声の長さは5:14です。話す速さは1分あたり約159語で、日常会話に近い自然なペースです。単語の86%は英語の頻出3,000語に含まれます。残りは練習の前に確認しておきましょう。

この動画の重要語彙

動画に出てくる覚えておきたい単語15語を、発音と意味つきで紹介します。

単語	発音	意味
sue 動詞	/sjuː/	訴える
predict 動詞	/pɹɪˈdɪkt/	予言する
topic 名詞	/ˈtɑpɪk/	話題, テーマ
relation 名詞	/ɹɪˈleɪʃ(ə)n/	関係, 関連
extent 名詞	/ɪkˈstɛnt/	程度
technique 名詞	/tɛkˈniːk/	技術
discovery 名詞	/dɪˈskʌv.ɹi/	発見
architecture 名詞	/ˈɑː.kɪˌtɛk.t͡ʃə/	建築, 建築学
reflect 動詞	/ɹɪˈflɛkt/	跳ね返す, 反射する
remind 動詞	/ɹəˈmaɪnd/	思い出させる
sequence 名詞	/ˈsiː.kwəns/	順序
investigate 動詞	/ɪnˈvɛs.tə.ɡeɪt/	調べる, 調査する
mask 名詞	/mæsk/	仮面, 覆面
rely 動詞	/ɹɪˈlaɪ/	頼る, あてにする
scientist 名詞	/ˈsaɪəntəst/	科学者

動画に出てくる句動詞

単語	意味
find out 動詞	見出す, 調べる

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	are expected · are trained · have been trained
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	have been trained · have started · it's been

注意したい発音

話し手はwe're, don'tなど、短縮形や弱形を4回使っています。聞こえたとおりの短い形で発音しましょう。

「sh」と「zh」の音: computational /ˌkɑmpjʊˈteɪʃənəl/, relation /ɹɪˈleɪʃ(ə)n/, collections /kəˈlɛkʃənz/, profession /pɹəˈfɛʃ.ən/, prediction /pɹɪˈdɪkʃən/
長い単語（アクセントの位置に注意）: computational /ˌkɑmpjʊˈteɪʃənəl/, architecture /ˈɑː.kɪˌtɛk.t͡ʃə/, investigate /ɪnˈvɛs.tə.ɡeɪt/, phenomenon /fəˈnɒm.ɪ.nən/, entity /ˈɛn.tə.ti/ [-tʰi]

日本語話者が苦手な音：

/r/ と /l/ の区別 — /l/ は舌先を歯茎につけ、/r/ はどこにもつけない: relation /ɹɪˈleɪʃ(ə)n/, reflect /ɹɪˈflɛkt/, rely /ɹɪˈlaɪ/, probability /ˌpɹɑ.bəˈbɪl.ə.ti/, neural /ˈnʊɹəl/
/v/ — /b/ にならないように、上の歯を下唇に当てる: discovery /dɪˈskʌv.ɹi/, investigate /ɪnˈvɛs.tə.ɡeɪt/, devise /dɪˈvaɪz/, overarching /ˌoʊvəɹˈɑɹt͡ʃɪŋ/
/f/ — 「フ」ではなく、上の歯と下唇で出す: reflect /ɹɪˈflɛkt/, phenomenon /fəˈnɒm.ɪ.nən/, profession /pɹəˈfɛʃ.ən/, firstly /ˈfɜɹstli/

この動画での練習方法

まず声を出さずに動画を最後まで聞き、知らない単語をメモします。
まず0.75倍速で一文ずつシャドーイングし、慣れてきたら通常の速度に戻します。
自分の声を録音して元の音声と比べます。sue, predict, topicなどの単語に特に注意しましょう。

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: Living with Machines: The language of mechanisation - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

アプリをダウンロード

おすすめ動画

When AIs act emotional

John Berger / Ways of Seeing , Episode 1 (1972)

The Problem with ANGLISH (the de-Frenchified Version of English)

The Engineering that Runs the Digital World 🛠️⚙️💻 How do CPUs Work?

The Great A.I. Debate: Should We Fear the Digital Revolution? [2015] | Intelligence Squared

MIT Introduction to Deep Learning | 6.S191

How do some languages not have tense?

Introduction to statistical prediction models

Harnessing generative AI in architecture, engineering & construction

Noam Chomsky on Language Aquisition

LangChain in 5 Minutes (Explained Clearly)

What AI can (and can't) currently do in a real lab | Jennifer Doudna