← News Topics

Speaking Across Languages

Day 1: Will AI Take the Interpreter’s Job?

言語を越えて話す ― 第1日:AIは通訳者の仕事を奪うのか

Headsets on a conference table facing a screen of blurred translated subtitles and a laptop with a soft translation waveform

An interpreter in Indonesia was preparing for an assignment in Vietnam when the agency called with unexpected news. The client had decided to use AI instead, and her trip was cancelled. She described the experience on the BBC’s Asia Specific. For her, the effect of AI on interpreting was already more than a prediction about the future.

インドネシアの通訳者は、ベトナムでの仕事の準備をしていたとき、代理店から予期しない知らせを受けました。依頼主が代わりにAIを使うことに決め、彼女の出張はキャンセルされたのです。彼女はその経験をBBCのAsia Specificで語りました。彼女にとって、通訳へのAIの影響は、すでに将来についての予測以上のものでした。

At some conferences, AI now turns speech into translated subtitles on a screen. An attendee who understands most of a presentation may only need the gist of the unfamiliar parts. In the past, an organization might have hired an interpreter to whisper those parts to that person. A screen can now be considered good enough for that particular task, even if it does not reproduce everything an interpreter would convey.

一部の会議では、AIが発話を画面上の翻訳字幕に変えるようになっています。発表のほとんどを理解している参加者が必要とするのは、知らない部分の要旨だけかもしれません。以前なら、組織はその人にその部分をささやき通訳する通訳者を雇ったかもしれません。画面は、通訳者が伝えるすべてを再現しなくても、その特定の仕事には十分だとみなされるようになっています。

Another interpreter on the programme described a different problem. He had asked an AI tool to interpret between Japanese and English, but sometimes it answered a speaker’s question instead of interpreting it. At other times, it added explanations the speaker had never given. The translation might sound fluent while the tool’s behaviour remained unpredictable. In a meeting, participants could mistake its extra words for the speaker’s own.

番組に出演した別の通訳者は、別の問題を述べました。彼はAIツールに日本語と英語の間で通訳するよう頼んだのですが、ときどきそれは通訳する代わりに話者の質問に答えてしまいました。別のときには、話者が一度も言っていない説明を付け加えました。翻訳は流ちょうに聞こえても、ツールの挙動は予測できないままでした。会議では、参加者はその余分な言葉を、話者自身の言葉だと勘違いする可能性があります。

The choice of words can also affect relationships. The Indonesian interpreter recalled AI using kamu, a familiar form of “you,” when addressing a senior official; in that setting, anda would have been more appropriate. The difference concerns register—the level of formality a situation calls for. An interpreter must understand the people in the room and the protocol of the event, as well as the words being spoken.

言葉の選び方も、人間関係に影響しえます。インドネシアの通訳者は、AIが上級の公務員に呼びかけるとき、親しみのある「あなた」の形であるkamuを使ったことを思い出しました。その場では、andaのほうが適切だったでしょう。その違いはレジスター、つまり状況が求める丁寧さの水準に関わります。通訳者は、話されている言葉だけでなく、その場にいる人々と、行事の作法も理解しなければなりません。

These problems do not mean clients will always choose a person. The two interpreters disagreed about how quickly AI might improve and how much work it might take. They also recognized that a meeting where someone needs a rough understanding is different from a high-stakes legal or political discussion, where a mistaken statement could have serious consequences. The appropriate standard depends on what the interpretation is being used for.

これらの問題は、依頼主が常に人を選ぶという意味ではありません。二人の通訳者は、AIがどれだけ速く改善しうるか、どれだけの仕事を奪いうるかについて意見が分かれました。また、大まかな理解が必要な会議と、誤った発言が深刻な結果をもたらしうる、利害の大きい法律や政治の議論とは違うことも認めていました。適切な基準は、その通訳が何のために使われているかによって変わります。

AI therefore raises two questions at once: which interpreting tasks will clients stop paying people to do, and how will they manage the risks when they use a machine? At an important meeting, someone still needs oversight of what appears on the screen, and someone must be accountable if an error changes what participants understand. The cancelled assignment shows that AI does not have to interpret perfectly for clients to stop hiring interpreters for some meetings.

したがってAIは、二つの問いを同時に投げかけます。依頼主はどの通訳の仕事に人への支払いをやめるのか、そして機械を使うとき、どうリスクを管理するのか。重要な会議では、画面に何が出ているかを監督する人が今も必要です。誤りが参加者の理解を変えてしまった場合に、説明責任を負う人もいなければなりません。キャンセルされた仕事は、一部の会議で依頼主が通訳者を雇うのをやめるために、AIが完璧に通訳する必要はないことを示しています。

Vocabulary

  1. assignment — a particular piece of work given to someone. Example: The interpreter accepted an assignment at an international conference.
  2. subtitles — written words shown on a screen to represent speech. Example: The audience followed the presentation through translated subtitles.
  3. gist — the main idea, without every detail. Example: I understood the gist of her speech, but I missed some important details.
  4. unpredictable — difficult to know or control in advance. Example: The tool’s unpredictable responses made it unsuitable for that meeting.
  5. register — the level of formality used in speaking or writing. Example: An interpreter must choose a suitable register when addressing an official.
  6. protocol — accepted rules for behaviour at an official event. Example: The interpreter learned the meeting’s protocol before it began.
  7. high-stakes — involving serious possible consequences. Example: A legal hearing is a high-stakes setting for interpretation.
  8. oversight — careful supervision or checking. Example: The organization arranged human oversight of the AI subtitles.
  9. accountable — responsible for explaining or answering for a decision or mistake. Example: The organizers must be accountable if an error affects the meeting.

Comprehension Questions

  1. What happened to the interpreter’s planned trip to Vietnam?
    1. She travelled there to test an AI system.
    2. The client cancelled the interpreting work and chose AI.
    3. The agency asked her to teach Vietnamese.
    4. She declined the assignment because of its cost.

    通訳者の予定されていたベトナムへの出張に、何が起きましたか?

  2. Why might an organization use AI subtitles for one attendee?
    1. The attendee only needs the main points of unfamiliar speech.
    2. Subtitles can always reproduce every nuance of speech.
    3. Human interpreters are not permitted at conferences.
    4. AI subtitles remove the need to listen to the speaker.

    組織が一人の参加者のためにAI字幕を使うことがあるのはなぜですか?

  3. What did the other interpreter say an AI tool sometimes did?
    1. It refused to translate Japanese words.
    2. It answered a question instead of interpreting it.
    3. It changed the meeting’s schedule.
    4. It identified every speaker incorrectly.

    別の通訳者は、AIツールがときどき何をしたと言いましたか?

  4. What does the example of kamu and anda illustrate?
    1. The two words refer to different countries.
    2. A word’s level of formality can matter in a particular relationship.
    3. AI cannot produce Indonesian text.
    4. Interpreters must always use the most familiar expression.

    kamuとandaの例は、何を示していますか?

  5. What is the article’s main point?
    1. AI must become perfect before it affects interpreting work.
    2. Every meeting requires the same standard of interpretation.
    3. AI is already changing some work, while its use still requires decisions about risk and responsibility.
    4. Human interpreters no longer work at international meetings.

    記事の主な主張は何ですか?

Discussion Questions

  1. In what kind of meeting might translated subtitles provide enough help? What information might still be missed?

    どのような会議なら、翻訳字幕で十分な助けになりうるでしょうか。また、どのような情報がまだ欠けうるでしょうか?

  2. Why is an AI tool answering a question instead of interpreting it a serious error?

    AIツールが通訳する代わりに質問に答えることが、深刻な誤りであるのはなぜでしょうか?

  3. How should organizers decide whether a meeting needs a professional interpreter?

    主催者は、会議にプロの通訳者が必要かどうかを、どのように判断すべきでしょうか?

  4. If an organization uses AI interpretation, what kind of human oversight would be useful?

    組織がAI通訳を使う場合、どのような人による監督が役立つでしょうか?

  5. How might AI change the skills that professional interpreters need?

    AIは、プロの通訳者に必要な技能をどのように変えうるでしょうか?

Speaking Practice

A company plans to use AI subtitles at every international meeting to save money. Explain when this might work and when the company should consider a professional interpreter. Speak for about one minute, using examples from the article.

ある会社は、費用を節約するために、すべての国際会議でAI字幕を使う計画です。これがうまくいきうる場合と、会社がプロの通訳者を検討すべき場合を説明してください。記事の例を使い、約1分間話してください。