Demo Page:
Synthetic Data Domain Adaptation for ASR via LLM-based Text and Phonetic Respelling Augmentation
[Paper] [Filtering code]
Natsuo Yamashita, Koichi Nagatsuka, Hiroaki Kokubo, Kota Dohi, Tuan Vu Ho
Hitachi, Ltd.
Abstract End-to-end automatic speech recognition often degrades on domain-specific data due to scarce in-domain resources. We propose a synthetic-data-based domain adaptation framework with two contributions: (1) a large language model (LLM)-based text augmentation pipeline with a filtering strategy that balances lexical diversity, perplexity, and domain-term coverage, and (2) phonetic respelling augmentation (PRA), a novel method that introduces pronunciation variability through LLM-generated orthographic pseudo-spellings. Unlike conventional acoustic-level methods such as SpecAugment, PRA provides phonetic diversity before speech synthesis, enabling synthetic speech to better approximate real-world variability. Experimental results across four domain-specific datasets demonstrate consistent reductions in word error rate, confirming that combining domain-specific lexical coverage with realistic pronunciation variation significantly improves ASR robustness.
Usage Notice
This demo page is provided for research and academic demonstration purposes only.
All contents on this page, including prompt examples, generated texts, synthesized audio, and code snippets, are not intended for commercial use.
Any commercial use of the materials presented here is strictly prohibited without prior written permission from the authors.
Overview
LLM-based text augmentation pipeline first over-generates a large pool of candidate sentences (steps 1-5: domain seed, context seed, multilingual prompting, paraphrasing, multiple LLMs) and then applies filtering (step 6) which optimizes TTR, perplexity, and domain-specific terms weighting.
LLM-based phonetic respelling rewrites text with LLMs into alternative spellings that approximate natural spoken pronunciation, introducing realistic pronunciation variability without relying on phonetic symbols.
LLM-based text augmentation pipeline
This section shows sample prompts and generated texts on each dataset. Results in three languages are also presented to demonstrate multilingual prompting.
1. Domain seed
"Generate {n} situations/scenarios in English where air traffic controllers are speaking in air traffic control.
Note that you should generate the scenarios or situations in which utterances take place, not the utterances themselves.
No annotations or explanations are needed; output only {n} scenarios or situations in English.
Please generate as diverse a set of scenarios and situations as possible.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed."
- • "Instructing an aircraft to taxi to a runway"
- • "Denying pushback due to ramp congestion"
- • "Granting clearance to cross active runway"
- • "Assigning runway for departure"
- • "Issuing landing clearance"
2.1. Context seed: from scenarios
"Generate 10 English sentences according to following scenario in air traffic control in Europe.
Scenario: Instructing an aircraft to taxi to a runway
Make sure each sentence include as diverse names, locations, aircraft names, and technical terms as possible.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Lufthansa 307, taxi via Juliet and Alpha to holding point Runway 27R, contact Ground on 121.8."
- • "Alitalia 452, proceed to Runway 04L via taxiways Bravo and Kilo, hold short at Sierra intersection."
- • "KLM 263, proceed via Golf and Uniform to holding point Runway 36C, give way to the arriving Embraer."
- • "Swiss 118, taxi to Runway 18R via Lima, Oscar, cross Runway 14, advise when ready for departure."
- • "Air France 214, taxi via Charlie and Mike to Runway 09, hold short of Whiskey, expect departure shortly."
2.2. Context seed: from domain-specific terms
Generate 10 English sentences in the different scenarios of air traffic control in Europe, including the term of 'Belstar'.
Make sure that each sentence is a single sentence and use as different names, locations, aircraft names, technical terms, numbers, and contexts as possible.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is at most 200 characters long.
The main language is English.
Use the terms of Belstar itself, do not change the language for the term of Belstar.
Annotations, explanations, or repetitions are not needed.
- • Belstar 214, descend to flight level 110 and contact Zurich Approach on 119.7.
- • "Eurojet 650, maintain present heading and expect direct Belstar waypoint in ten miles. "
- • "Munich Tower, Lufthansa 327 passing Belstar, requesting climb to FL340. "
- • "Ryanair 982, after Belstar, proceed direct to Krakow VOR, flight planned route approved. "
- • "ATC to Swiss 87, radar contact established five miles south of Belstar, continue as cleared. "
4. Paraphrasing
"Rewrite and paraphrase the sentence of Belstar 214, descend to flight level 110 and contact Zurich Approach on 119.7. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Belstar.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • Belstar two one four, descend to altitude one one zero and switch to Geneva Approach frequency one one nine point seven.
- • "Approach clearance for Belstar two one four: reduce your altitude to flight level one one zero and contact Vienna Control at one one nine point seven."
- • "Belstar two one four, you are cleared to descend to one one zero flight level and reach out to Munich Approach on one one nine decimal seven."
- • "Descend to flight level one one zero, Belstar two one four, and proceed to communicate with Marseille Approach on frequency one one nine point seven."
- • "Belstar two one four, adjust your altitude to one one zero, then call Amsterdam Approach on one one nine decimal seven."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 34648 | Trasadingen handles Three helicopters. |
| 31386 | Manchester Tower issued routing about olben to Pilot Sarah flying Echo November Five Zero. |
| 31282 | Delta Airbus Three Fifty leaves Atlanta on Proceed. |
| 29750 | The Airbus Three Twenty exits Athens on Papa Oscar. |
| 28367 | Ryanair navig Oslo airport using advanced navig. |
1. Domain seed
"请生成 {n} 个关于空中交通管制中管制员发话的情境/场景，内容为中文。
请注意，生成的应是发言发生的情境或场景，而不是发言内容本身。
无需注释或解释，仅输出 {n} 个中文情境或场景。
请尽可能多样化地生成这些情境或场景。
每个句子的长度不少于5个汉字且不超过100个汉字。
不需要添加注释、说明或重复。"
- • "允许起飞"
- • "允许降落"
- • "允许滑行"
- • "进入跑道等待"
- • "跑道进入许可"
2.1. Context seed: from scenarios
"请根据以下欧洲航空管制场景，生成10个中文句子。
场景：允许起飞
请确保每个句子尽可能包含多样的人名、地点、飞机名称和技术术语。
请确保每个句子的长度不少于5个字符且不超过100个字符。
不需要注释、解释或重复内容。
每个句子应单独占一行，并用换行符分隔。"
- • "巴黎塔台呼叫奥地利航空OS423，允许使用27L跑道起飞，风速210度8节。"
- • "伦敦塔台通知汉莎航班LH789，空中客车A320可从09R跑道按仪表程序起飞。"
- • "马德里地面呼叫伊比利亚航空IB520，波音737已获准滑行至36L等待起飞指令。"
- • "阿姆斯特丹塔台，荷兰皇家航空KL1023，你已经获得起飞许可，请由24跑道起飞。"
- • "柏林塔台呼叫瑞士国际LX987，你的A220飞机已批准从08R跑道起飞。"
2.2. Context seed: from domain-specific terms
请在欧洲航空管制的各种场景下，生成10个包含belstar术语的中文句子。
要求：
- 每个句子仅限一句，并尽可能使用不同的人名、地点、飞机名称、技术术语、数字和语境。
- 每个句子应单独占一行，并用换行符分隔。
- 每个句子的长度不少于5个字符且不超过100个字符。
- 主要语言为中文。
- 请直接使用belstar这一术语，不要更改该术语的语言。"
- • "地面管制通知belstar 125，准备前往巴黎戴高乐机场起飞许可已批准。"
- • "机长安德烈通过无线电报告，belstar 452目前即将抵达法兰克福011号滑行道。"
- • "belstar 301，请在跑道27左等待下一步起飞通知。"
- • "由于布鲁塞尔上空突发雷雨，belstar 789请求绕道飞行并等待新指令。"
- • "航路管制员询问belstar 522，目前是否已经达到了预定的巡航高度FL390。"
3. Multilingual Prompting: Translation
"Translate the Chinese sentence of sentence in air traffic control to English.
Write all numbers in spelled-out English words instead of using digits.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "Ground control informs Belstar one two five that clearance for departure to Paris Charles de Gaulle Airport has been granted."
- • "Captain Andre reports via radio that Belstar four five two is currently approaching Frankfurt taxiway zero one one."
- • "Belstar three zero one, please hold short at runway two seven left and await further departure instructions."
- • "Due to a sudden thunderstorm over Brussels, Belstar seven eight nine requests diversion and awaits new instructions."
- • "The enroute controller asks Belstar five two two whether the assigned cruising altitude flight level three niner zero has been reached."
4. Paraphrasing
"Rewrite and paraphrase the sentence of Ground control informs Belstar one two five that clearance for departure to Paris Charles de Gaulle Airport has been granted. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Belstar.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Moscow Ramp to Belstar Uniform Eight, clearance received, ready for departure for JFK, permission to take off."
- • "Control tower to Belstar Alpha Six, permission granted for departure to Heathrow, you’re clear for takeoff."
- • "Madrid Ground to Belstar Three Zero Eight, takeoff clearance given, proceed toward Frankfurt Airport."
- • "Belstar Juliet One Seven, you’re authorized for runway departure, onward to Tokyo Narita, have a safe takeoff."
- • "Tango Control calling Belstar Bravo Niner, runway is yours to depart for Dubai International, cleared for takeoff."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 34648 | Trasadingen handles Three helicopters. |
| 31386 | Manchester Tower issued routing about olben to Pilot Sarah flying Echo November Five Zero. |
| 31282 | Delta Airbus Three Fifty leaves Atlanta on Proceed. |
| 29750 | The Airbus Three Twenty exits Athens on Papa Oscar. |
| 28367 | Ryanair navig Oslo airport using advanced navig. |
1. Domain seed
"航空管制における，管制官が発話する状況・シナリオを日本語で{n}個生成して。
発話そのものではなく，発話の状況・シナリオであることに注意して。
本文の注釈は不要で，{n}個の状況・シナリオのみを日本語で出力して。
できるだけ多様な状況・シナリオを生成してください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。"
- • "出発クリアランスの発出"
- • "出発クリアランスの再確認"
- • "出発クリアランスの変更指示"
- • "離陸許可の発出"
- • "離陸許可のキャンセル"
2.1. Context seed: from scenarios
"ヨーロッパの航空管制における以下のシナリオに従って、10個の日本語の文を生成してください。
シナリオ: 出発クリアランスの発出
各文にはできるだけ多様な名前、場所、航空機名、専門用語を含めてください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。
各文は1行ずつで出力し，それぞれ改行してください。"
- • "フランクフルトグランド、ルフトハンザ410、滑走路25Cから出発クリアランスをお願いします。"
- • "チューリッヒ出発、スイスエア312、コペンハーゲン行き、FL350でのクリアランスを受領しました。"
- • "パリグランド、エールフランス75、ターミナル2F、エンジンスタート許可を申請します。"
- • "ミラノタワー、アリタリア722、ルートLIMC via DCT、出発指示を確認しました。"
- • "アムステルダム地上、KLM436、ビーコン点灯、ALPHA9へタキシーします。"
2.2. Context seed: from domain-specific terms
ヨーロッパの航空管制のさまざまなシナリオにおいて、『Belstar』という用語を含む10個の日本語の文を生成してください。
各文は1文のみとし、できるだけ異なる名前、場所、航空機名、専門用語、数字、文脈を使ってください。
各文は1行ずつで出力し，それぞれ改行してください。
各文は5文字以上100文字以下であることを必ず守ってください。
主な言語は日本語です。
『Belstar』という用語自体を使用し、その用語の言語は変更しないでください。
注釈、説明、繰り返しは不要です。
- • "東京タワー管制へ、belstar450、現在マドリード上空FL320、天候良好。"
- • "モナコ空港でbelstar機の再着陸指示が発令された。"
- • "belstar123、次のウェイポイントまでの距離を教えてください。"
- • "ロンドンからパリへ向かうbelstar777便に乱気流警報が出ました。"
- • "ザルツブルク上空でbelstarの通信トラブルが確認されました。"
3. Multilingual Prompting: Translation
"Translate the Japanese sentence of sentence in air traffic control to English.
Write all numbers in spelled-out English words instead of using digits.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "Tokyo Tower Control, belstar four five zero, currently over Madrid at flight level three two zero, weather is good."
- • "A go-around was ordered for the Belstar flight at Monaco airport."
- • "Belstar one two three, please advise the distance to the next waypoint."
- • "A turbulence advisory has been issued for Belstar triple seven flight bound from London to Paris."
- • "A communication problem with Belstar was confirmed over Salzburg."
4. Paraphrasing
"Rewrite and paraphrase the sentence of Tokyo Tower Control, belstar four five zero, currently over Madrid at flight level three two zero, weather is good. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Belstar.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "London Radar, belstar four five zero, passing over Rome at altitude flight level two nine zero, experiencing clear skies."
- • "New York Approach, belstar four five zero, maintaining flight level three five zero over Paris, visibility excellent."
- • "Sydney Center, belstar four five zero, currently cruising above Johannesburg at level three three zero, weather fair."
- • "Dubai Control, belstar four five zero, over Berlin at flight level three one zero, with no weather issues reported."
- • "Honolulu Tower, belstar four five zero, at this time above Lisbon at cruising altitude three two zero, conditions calm."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 34648 | Trasadingen handles Three helicopters. |
| 31386 | Manchester Tower issued routing about olben to Pilot Sarah flying Echo November Five Zero. |
| 31282 | Delta Airbus Three Fifty leaves Atlanta on Proceed. |
| 29750 | The Airbus Three Twenty exits Athens on Papa Oscar. |
| 28367 | Ryanair navig Oslo airport using advanced navig. |
1. Domain seed
"Generate {n} situations/scenarios in English where pilots are speaking in air traffic control.
Note that you should generate the scenarios or situations in which utterances take place, not the utterances themselves.
No annotations or explanations are needed; output only {n} scenarios or situations in English.
Please generate as diverse a set of scenarios and situations as possible.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed."
- • "A pilot descends early at ATC’s request for traffic separation."
- • "A pilot relays a passenger disturbance requiring security assistance on arrival."
- • "A pilot updates ATC with revised estimates after a change in groundspeed."
- • "A pilot asks for a climb to avoid reported icing conditions."
- • "A pilot declares a medical emergency and requests immediate descent."
2.1. Context seed: from scenarios
"Generate 10 English sentences according to following scenario in air traffic control in Europe.
Scenario: A pilot descends early at ATC’s request for traffic separation.
Make sure each sentence include as diverse names, locations, aircraft names, and technical terms as possible.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Captain Smith, this is London Center, descend your Airbus A320 to flight level 150 for traffic separation from the approaching Boeing 737. "
- • "Lufthansa 472, reduce altitude to 3,500 feet as requested by Frankfurt Tower due to nearby Ryanair 738 climbing through your altitude. "
- • "EasyJet 2945, initiate descent now to 7,000 feet, Paris Control advises traffic conflict with a descending Cessna Citation XLS. "
- • "Turkish Airlines 1904, descend early to 12,000 feet for radar separation with the preceding Emirates A380, Brussels Center instructs. "
- • "Delta 867, start your early descent to 4,500 feet; Amsterdam ATC managing separation from the climbing Bombardier CRJ900. "
2.2. Context seed: from domain-specific terms
Generate 10 English sentences in the different scenarios of air traffic control in Europe, including the term of 'QNH'.
Make sure that each sentence is a single sentence and use as different names, locations, aircraft names, technical terms, numbers, and contexts as possible.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is at most 200 characters long.
The main language is English.
Use the terms of Belstar itself, do not change the language for the term of QNH.
Annotations, explanations, or repetitions are not needed.
- • "EasyJet 243, descend to 5,000 feet and set QNH 1013 on your altimeter for the approach into Heathrow."
- • "Lufthansa 482, climb flight level 120 and report when passing QNH 1009 near Frankfurt airspace. "
- • "Ryanair 512, maintain current heading, adjust QNH to 1015 for Paris Charles de Gaulle transition."
- • "British Airways 789, holding at waypoint ALPHA, set QNH 1017 and await further clearance from London Control."
- • "Air France 320, cleared to land runway 27, verify QNH 1020 and report established on final approach."
4. Paraphrasing
"Rewrite and paraphrase the sentence of EasyJet 243, descend to 5,000 feet and set QNH 1013 on your altimeter for the approach into Heathrow. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of QNH.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Delta three forty-seven, descend now to fourteen thousand feet and dial QNH one-zero-one-three on your altimeter for your arrival into JFK."
- • "Emirates five twenty-one, reduce altitude to eight thousand feet and adjust QNH one-zero-one-three on your altimeter for approach at Dubai International."
- • "British Airways one ninety-eight, lower altitude to six thousand feet and set QNH one-zero-one-three on your altimeter as you prepare to land at Gatwick."
- • "Lufthansa three twelve, descend to seven thousand feet, and input QNH one-zero-one-three on your altimeter for the approach into Frankfurt."
- • "Qantas four sixty, drop to ten thousand feet and tune QNH one-zero-one-three on your altimeter for final approach at Sydney Airport."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 249567 | Air Europa Four Five Six maintain Two Hundred Forty level. |
| 22736 | Vlasim Delivery clears Austrian Arrows 321 for engine start. |
| 19033 | Germany control point Munich Trans passed captain is Marx Fernaler. |
| 17124 | Sierra Victor Six One Zero informs Rome Radar immediately about critical icing. |
| 16464 | Tango Yankee Zero Four Seven notifies Sydney Centre of rapid ice accretion on airframe. |
1. Domain seed
"生成 {n} 个飞行员在空中交通管制中讲话的情景或场景（用中文）。 请注意，您应该生成的是发生话语的情景或场景，而不是话语本身。 不需要注释或解释，仅输出 {n} 个中文情景或场景。 请尽量生成多样化的情景和场景。 确保每句话至少有 5 个字符，且不超过 100 个字符。 不需要注释、解释或重复。"
- • "飞行员请求冲出跑道撤离。"
- • "飞行员报告有医疗紧急情况。"
- • "飞行员通报飞机客舱压力异常。"
- • "飞行员要求终止当前进近，执行复飞。"
- • "飞行员报告液压系统报警。"
2.1. Context seed: from scenarios
"请根据以下欧洲航空管制场景，生成10个中文句子。
场景：飞行员请求冲出跑道撤离。
请确保每个句子尽可能包含多样的人名、地点、飞机名称和技术术语。
请确保每个句子的长度不少于5个字符且不超过100个字符。
不需要注释、解释或重复内容。
每个句子应单独占一行，并用换行符分隔。"
- • "上海飞行员张伟请求空管允许波音737紧急冲出巴黎戴高乐机场跑道撤离。"
- • "伦敦希思罗机场塔台指挥官刘洋指示空中巴士A320机组立即准备跑道紧急撤离程序。"
- • "法兰克福机场控制员陈晨确认俄罗斯苏霍伊Superjet100飞行员米哈伊的撤离请求。"
- • "马德里机场地勤人员与飞行员王强协调空客A350跑道冲出后的消防及救援工作。"
- • "米兰马尔彭萨机场塔台指挥王丽指示波音777飞行员约翰执行跑道紧急撤离程序。"
2.2. Context seed: from domain-specific terms
请在欧洲航空管制的各种场景下，生成10个包含QNH术语的中文句子。
要求：
- 每个句子仅限一句，并尽可能使用不同的人名、地点、飞机名称、技术术语、数字和语境。
- 每个句子应单独占一行，并用换行符分隔。
- 每个句子的长度不少于5个字符且不超过100个字符。
- 主要语言为中文。
- 请直接使用QNH这一术语，不要更改该术语的语言。"
- • "飞行员张明，请报告MH370当前qnh设置。"
- • "伦敦塔台通知空客A320，qnh已调整至1013百帕。"
- • "柏林空管指示波音747调整qnh至1020以适应降落。"
- • "飞行员安娜，飞越阿姆斯特丹，请确认你收到的qnh数值。"
- • "苏黎世控制中心要求空中巴士A380更新qnh为1008。"
3. Multilingual Prompting: Translation
"Translate the Chinese sentence of sentence in air traffic control to English.
Write all numbers in spelled-out English words instead of using digits.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "Pilot Zhang Ming, please report the current QNH setting of MH Three Seven Zero."
- • "London Tower informs Airbus A three two zero that the QNH has been adjusted to one zero one three hectopascals."
- • "Berlin air traffic control instructed the Boeing seven four seven to adjust the QNH to one zero two zero to accommodate landing."
- • "Pilot Anna, flying over Amsterdam, please confirm the QNH value you have received."
- • "Zurich Control Center requested Airbus A380 to update QNH to one thousand and eight."
4. Paraphrasing
"Rewrite and paraphrase the sentence of Pilot Zhang Ming, please report the current QNH setting of MH Three Seven Zero. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of QNH.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Captain Li Wei, kindly provide the present QNH configuration for flight Delta Four Five Eight at Beijing Approach."
- • "Flight Engineer Anna Smith, please confirm the QNH setting currently assigned to Jetstream Eight Two Three over London Center."
- • "Commander Raj Patel, report the existing QNH value for aircraft Bravo Seven Zero One near Mumbai Tower."
- • "Pilot James Brown, advise the air pressure QNH reading for United Flight Echo Nine Six heading into New York Control."
- • "Captain Maria Garcia, state the active QNH setting for Skyhawk Twelve Four Six inbound to Madrid Radar."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 249567 | Air Europa Four Five Six maintain Two Hundred Forty level. |
| 22736 | Vlasim Delivery clears Austrian Arrows 321 for engine start. |
| 19033 | Germany control point Munich Trans passed captain is Marx Fernaler. |
| 17124 | Sierra Victor Six One Zero informs Rome Radar immediately about critical icing. |
| 16464 | Tango Yankee Zero Four Seven notifies Sydney Centre of rapid ice accretion on airframe. |
1. Domain seed
"航空管制における，パイロットが発話する状況・シナリオを日本語で{n}個生成して。
発話そのものではなく，発話の状況・シナリオであることに注意して。
本文の注釈は不要で，{n}個の状況・シナリオのみを日本語で出力して。
できるだけ多様な状況・シナリオを生成してください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。"
- • "上昇高度の変更を要請する状況"
- • "通信異常を報告する場面"
- • "航空身体検査証明書の確認依頼場面"
- • "他機とのニアミスを報告する場面"
- • "滑走路上の異物を発見したことを伝える状況"
2.1. Context seed: from scenarios
"ヨーロッパの航空管制における以下のシナリオに従って、10個の日本語の文を生成してください。
シナリオ: 上昇高度の変更を要請する状況
各文にはできるだけ多様な名前、場所、航空機名、専門用語を含めてください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。
各文は1行ずつで出力し，それぞれ改行してください。"
- • "ブリュッセルコントロール、エアバスA320、FL350へ上昇許可を要請します。"
- • "ルフトハンザ123便、モナコ上空でFL370までの上昇をリクエストします。"
- • "ロンドン航空管制、ボーイング777、現在のFL320からFL390へ上がりたいです。"
- • "パリCDGアプローチ、エミレーツA380、上昇によりFL410まで移行希望。"
- • "アムステルダムCTRL、エアバスA321、FL340からFL360へ上昇許可をお願いします。"
2.2. Context seed: from domain-specific terms
ヨーロッパの航空管制のさまざまなシナリオにおいて、『QNH』という用語を含む10個の日本語の文を生成してください。
各文は1文のみとし、できるだけ異なる名前、場所、航空機名、専門用語、数字、文脈を使ってください。
各文は1行ずつで出力し，それぞれ改行してください。
各文は5文字以上100文字以下であることを必ず守ってください。
主な言語は日本語です。
『QNH』という用語自体を使用し、その用語の言語は変更しないでください。
注釈、説明、繰り返しは不要です。
- • "ロンドンヒースロー空港では、離陸前にパイロットが最新のqnhを確認した。"
- • "パリのATCは乱気流情報とともにqnhの更新を航空機に伝達した。"
- • "エアバスA320の機長はフランクフルト着陸時に正確なqnhを設定した。"
- • "オーロラ観測ツアーのヘリコプターはノルウェーの空域でqnhを頻繁に調整した。"
- • "エジンバラ管制塔は霧が濃いため、到着便にqnhの再確認を指示した。"
3. Multilingual Prompting: Translation
"Translate the Japanese sentence of sentence in air traffic control to English.
Write all numbers in spelled-out English words instead of using digits.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "At London Heathrow Airport, the pilot confirmed the latest QNH before takeoff."
- • "Paris ATC transmitted the QNH update to the aircraft along with turbulence information."
- • "The captain of the Airbus A three twenty set the correct QNH during the Frankfurt landing."
- • "The helicopter for the aurora observation tour frequently adjusted its QNH in Norwegian airspace."
- • "Edinburgh control tower instructed arriving flights to reconfirm the qnh due to dense fog."
4. Paraphrasing
"Rewrite and paraphrase the sentence of At London Heathrow Airport, the pilot confirmed the latest QNH before takeoff. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of QNH.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "At New York JFK Airport, the captain of a Boeing seven seventy-seven verified the current QNH prior to departure."
- • "Before takeoff from Sydney Kingsford Smith Airport, the Airbus A three hundred pilot checked the latest QNH reading."
- • "The commander of a Bombardier Global Express at Paris Charles de Gaulle Airport confirmed the applicable QNH before leaving the stand."
- • "At Dubai International Airport, the flight crew of an Embraer E two hundred ensured the QNH was up to date before taxiing for takeoff."
- • "Prior to departure at Tokyo Haneda Airport, the pilot flying the Boeing seven eighty-seven validated the most recent QNH information."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 249567 | Air Europa Four Five Six maintain Two Hundred Forty level. |
| 22736 | Vlasim Delivery clears Austrian Arrows 321 for engine start. |
| 19033 | Germany control point Munich Trans passed captain is Marx Fernaler. |
| 17124 | Sierra Victor Six One Zero informs Rome Radar immediately about critical icing. |
| 16464 | Tango Yankee Zero Four Seven notifies Sydney Centre of rapid ice accretion on airframe. |
1. Domain seed
"Please generate {n} scenarios or situations in which utterances occur at the Supreme Court of India, in English.
Note that you should generate the scenarios or situations in which utterances take place, not the utterances themselves.
No annotations or explanations are needed; output only {n} scenarios or situations in English.
Please generate as diverse a set of scenarios and situations as possible."
- • "Hearing a constitutional amendment case regarding reservation policies."
- • "Reviewing a public interest litigation related to environmental protection."
- • "Adjudicating a dispute between two state governments over water sharing."
- • "Deciding on the validity of a new electoral reform law."
- • "Addressing an appeal in a high-profile criminal case."
2.1. Context seed: from scenarios
"Generate 10 English sentences according to following scenario in supreme court in India.
Scenario: Hearing a constitutional amendment case regarding reservation policies.
Make sure each sentence include as diverse named entities and technical terms as possible.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "The Supreme Court bench, led by Chief Justice Dhananjaya, examined the constitutional validity of the 103rd Amendment impacting reservation policies for Economically Weaker Sections."
- • "Justice Chandrachud referenced landmark cases such as Indra Sawhney and Minerva Mills to assess the balance between affirmative action and the basic structure doctrine."
- • "The Union Government’s counsel submitted affidavits justifying the amendment under Articles 368 and 46, emphasizing socio-economic inclusion and social justice principles."
- • "An NGO representing Dalit communities highlighted disparity data from the National Sample Survey Office supporting expanded reservations in educational institutions and public employment sectors."
- • "Solicitor General Mehta cited the Kesavananda Bharati verdict to assert that reservation policies must not alter the Constitution’s fundamental framework including equality clauses under Article 14."
2.2. Context seed: from domain-specific terms
"Generate 10 English sentences in the different scenarios of supreme court in India, including the term of 'Sibal'. Make sure that each sentence is a single sentence and use as different named entities, technical terms, numbers, and contexts as possible. Each sentence should be output on its own line, separated by line breaks. Ensure that each sentence is at least 5 characters and at most 200 characters long. The main language is English. Use the terms of 'Sibal' itself, do not change the language for the term of 'Sibal'. Annotations, explanations, or repetitions are not needed. "
- • "In the Supreme Court of India, Sibal argued the constitutional validity of the Aadhaar Act before a five-judge bench."
- • "During the landmark 2012 hearing on internet censorship, Sibal represented the government in front of the Supreme Court. "
- • "Sibal submitted extensive documentation on digital privacy rights during the Public Interest Litigation (PIL) at the Supreme Court."
- • "The Supreme Court referred to Sibal's brief while deliberating the admissibility of electronic evidence in criminal trials."
- • "In a 2015 Supreme Court case concerning freedom of speech, Sibal contended the limits of Section 66A of the IT Act."
4. Paraphrasing
"Rewrite and paraphrase the sentence of In the Supreme Court of India, Sibal argued the constitutional validity of the Aadhaar Act before a five-judge bench. in air traffic control into 9 different versions.
Make sure that the sentences use as different names, locations, aircraft names, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Sibal.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "In the Supreme Court of India, Sibal presented arguments defending the Aadhaar Act’s constitutionality before a panel of five judges."
- • "Sibal appeared in India’s highest judiciary forum to contest the Aadhaar Act’s legal standing before a bench comprising five justices."
- • "Arguing the Aadhaar Act’s constitutional validity, Sibal addressed a quintet of judges at the supreme judicial forum in India."
- • "Before a cluster of five judges at India’s top court, Sibal upheld the Aadhaar Act’s constitutional legitimacy."
- • "At India’s apex court, Sibal challenged the legal framework of the Aadhaar Act before a distinguished five-member bench."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 4743 | Mafatlal judgement affects Indian corporate tax strategies. |
| 3621 | The Indian judiciary's highest authority debated rebia protection legislation. |
| 2776 | Dehors reports compiled by advocacy NGOs guided the Court in assessing toxic discharge ramifications on suburban populations. |
| 2624 | The Indian Supreme Court oversaw a placitum involving extensive labor disagreement settlements. |
| 2420 | Opposing labor factions advocated before India’s Supreme Court for reinforced statutory actions targeting illicit economic protests. |
1. Domain seed
"请用中文生成 {n} 个在印度最高法院中发生发言的情境或场景。
请注意，生成的应是发言发生的情境或场景，而不是发言内容本身。
无需注释或解释，仅输出 {n} 个中文情境或场景。
请尽可能多样化地生成这些情境或场景。
- • "审理宪法修正案合法性的法庭。"
- • "听取有关环境保护公益诉讼的案件。"
- • "审理两个邦之间的水资源分配纠纷"
- • "审查新的选举改革法的有效性。"
- • "处理一起备受关注的刑事案件上诉。"
2.1. Context seed: from scenarios
"请根据印度最高法院中的以下情境，生成10个中文句子。
情境：审理宪法修正案合法性的法庭。
请确保每个句子都是单句，并尽可能多地使用不同的专有名词、专业术语、数字和情境。
请确保每个句子的长度不少于5个字符且不超过100个字符。
无需注释、解释或重复。
每个句子请单独占一行，并用换行符分隔。"
- • "印度最高法院于2024年3月15日开始审理第103号宪法修正案的合法性问题。"
- • "首席大法官主持此次审判，涉及《基本权利平等原则》与修正案内容的冲突。"
- • "法庭重点关注修正案是否违反了印度宪法第14条的平等保护条款。"
- • "辩护方提出第103号修正案旨在促进社会正义，保护边缘群体权益。"
- • "公诉方质疑修正案是否侵犯了宪法赋予公民的基本自由权利。"
2.2. Context seed: from domain-specific terms
"请在印度最高法院的不同场景下，生成 10 个包含“Sibal”术语的中文句子。 请确保每个句子都是单句，并尽可能多地使用不同的专有名词、专业术语、数字和情境。 每个句子应单独占一行，并用换行符分隔。 请确保每个句子的长度不少于5个字符且不超过100个字符。 主要语言为中文。 请直接使用“Sibal”这一术语，不要更改该术语的语言。 不需要注释、解释或重复。"
- • "印度最高法院庭审中，律师sibal提请法官关注宪法第21条的问题。"
- • "sibal在最高法院辩论印度信息技术法案第66条修订案的合宪性。"
- • "最高法院第3号法庭由sibal代理案件，该案涉及土地征用补偿纠纷。"
- • "sibal提交的诉状援引1973年最高法院判例，主张言论自由应获得保护。"
- • "在最高法院调解会议上，sibal强调保护未成年人隐私权的重要性。"
3. Multilingual Prompting: Translation
"Translate the Chinese sentence of sentence in supreme court in India to English.
Write all numbers in spelled-out English words instead of using digits.
Output only the results with one line per conversation without line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "During the Supreme Court trial in India, lawyer Sibal urged the judge to focus on the issue of Article Twenty-one of the Constitution."
- • "Sibal argued about the constitutionality of the amendment to Section sixty-six of the Indian Information Technology Act in the Supreme Court of India."
- • "The Supreme Court's courtroom number three is represented by Sibal in a case involving land acquisition compensation disputes."
- • "The complaint submitted by Sibal cites the nineteen seventy-three Supreme Court precedent, asserting that freedom of speech should be protected."
- • "At the Supreme Court mediation meeting, Sibal emphasized the importance of protecting the privacy rights of minors."
4. Paraphrasing
"Rewrite and paraphrase the sentence of During the Supreme Court trial in India, lawyer Sibal urged the judge to focus on the issue of Article Twenty-one of the Constitution. in supreme court in India into 9 different versions.
Make sure that the sentences use as different named entities, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Sibal.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "During the landmark case at India's highest judiciary, advocate Sibal emphasized to the bench the critical importance of Article twenty-one of the Indian Constitution."
- • "In the constitutional bench hearing in New Delhi, lawyer Sibal implored the justices to prioritize matters concerning Article twenty-one protections."
- • "At the apex court session held in the Republic of India, legal expert Sibal stressed the significance of examining Article twenty-one rights."
- • "In the supreme judicial forum of India, counsel Sibal urged the presiding judge to concentrate on constitutional Article twenty-one issues."
- • "During the judiciary panel's review in India’s capital, Sibal the lawyer pressed for attention toward Article twenty-one under the Constitution."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 4743 | Mafatlal judgement affects Indian corporate tax strategies. |
| 3621 | The Indian judiciary's highest authority debated rebia protection legislation. |
| 2776 | Dehors reports compiled by advocacy NGOs guided the Court in assessing toxic discharge ramifications on suburban populations. |
| 2624 | The Indian Supreme Court oversaw a placitum involving extensive labor disagreement settlements. |
| 2420 | Opposing labor factions advocated before India’s Supreme Court for reinforced statutory actions targeting illicit economic protests. |
1. Domain seed
"インドの最高裁判所における，発話する状況・シナリオを日本語で10個生成して。
発話そのものではなく，発話の状況・シナリオであることに注意して。
本文の注釈は不要で，10個の状況・シナリオのみを日本語で出力して。
できるだけ多様な状況・シナリオを生成してください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。
- • "憲法改正の合憲性を巡る審理が行われている"
- • "主要政党間で選挙法に関する意見陳述が交わされている"
- • "少数民族の権利保護を求める訴訟の判決発表"
- • "死刑判決の上告審理が進行している"
- • "環境保護団体が産業開発計画に異議を申し立てる公聴会"
2.1. Context seed: from scenarios
"インドの最高裁判所における以下のシナリオに従って、10個の日本語の文を生成してください。
シナリオ: 憲法改正の合憲性を巡る審理が行われている
各文にはできるだけ多様な固有表現や専門用語を含めてください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。
各文は1行ずつで出力し，それぞれ改行してください。"
- • "インド最高裁は憲法改正の基本構造理論を検討している。"
- • "憲法裁判所は議会の改正権限の範囲を詳細に審査した。"
- • "最高裁判所の判事たちは連邦制維持の原則を重視している。"
- • "憲法制定者の意図解釈が審理の中心課題となっている。"
- • "改正条項と基本権保障の整合性が争点として浮上した。"
2.2. Context seed: from domain-specific terms
"インドの最高裁判所のさまざまなシナリオにおいて、『Sibal』という用語を含む10個の日本語の文を生成してください。
各文は1文のみとし、できるだけ異なる固有表現や専門用語、数字、文脈を使ってください。
各文は1行ずつで出力し，それぞれ改行してください。
各文は5文字以上100文字以下であることを必ず守ってください。
主な言語は日本語です。
『Sibal』という用語自体を使用し、その用語の言語は変更しないでください。
注釈、説明、繰り返しは不要です。"
- • "インド最高裁判所の判決で、sibal弁護士が重要な意見を述べた。"
- • "sibalは憲法改正に関する訴訟で注目された弁護士である。"
- • "最高裁でのsibalの主張は人権保障の新基準を提示した。"
- • "sibalが担当したケースは環境法に関する重要判例となった。"
- • "インド最高裁の判事がsibalの法的論点を評価した。"
3. Multilingual Prompting: Translation
"Translate the Japanese sentence of sentence in supreme court in India to English.
Write all numbers in spelled-out English words instead of using digits.
Output only the results with one line per conversation without line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "In the Supreme Court of India, Lawyer Sibal expressed an important opinion in the verdict."
- • "Sibal is a lawyer who gained attention for lawsuits concerning constitutional amendments in the Supreme Court of India."
- • "Sibal's argument in the Supreme Court presented a new standard for the protection of human rights."
- • "The case handled by Sibal became an important precedent in environmental law in the Supreme Court of India."
- • "A judge of the Supreme Court in India evaluated Sibal’s legal issues."
4. Paraphrasing
"Rewrite and paraphrase the sentence of In the Supreme Court of India, Lawyer Sibal expressed an important opinion in the verdict. in supreme court in India into 9 different versions.
Make sure that the sentences use as different named entities, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Sibal.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "In a landmark ruling at the Supreme Court of India, Lawyer Sibal delivered a pivotal perspective during the decision announcement."
- • "During the twenty-third session of the Supreme Court, Lawyer Sibal articulated a crucial viewpoint influencing the final judgment."
- • "At India's top judicial tribunal, Lawyer Sibal offered a significant argument that shaped the court's verdict."
- • "Lawyer Sibal contributed an essential insight in the Supreme Court's deliberation of a high-profile case within India’s highest legal forum."
- • "In the January hearing held in New Delhi’s Supreme Court, Lawyer Sibal pronounced a vital opinion impacting the ruling."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 4743 | Mafatlal judgement affects Indian corporate tax strategies. |
| 3621 | The Indian judiciary's highest authority debated rebia protection legislation. |
| 2776 | Dehors reports compiled by advocacy NGOs guided the Court in assessing toxic discharge ramifications on suburban populations. |
| 2624 | The Indian Supreme Court oversaw a placitum involving extensive labor disagreement settlements. |
| 2420 | Opposing labor factions advocated before India’s Supreme Court for reinforced statutory actions targeting illicit economic protests. |
1. Domain seed
"Generate {n} situations or scenarios describing pharmaceuticals or health supplements.
Please note that you should generate situations or scenarios of speaking, not the actual utterances.
Do not include any explanatory notes or annotations in the text; output only the {n} situations or scenarios in English.
Please generate as diverse a set of situations or scenarios as possible."
- • "A supplement store employee discussing product origin and testing."
- • "A patient discussing over-the-counter painkiller choices for arthritis."
- • "A transplant patient discussing medication adherence challenges online."
- • "A nurse warning about taking expired medications."
- • "A diabetic patient confirming with a pharmacist how to alternate injections."
2.1. Context seed: from scenarios
"Generate 10 English sentences according to following scenario describing pharmaceuticals or health supplements.
Scenario: A supplement store employee discussing product origin and testing.
Make sure each sentence include as diverse drug names, supplement names, named entities and technical terms as possible.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Our CoQ10 capsules, sourced from Canadian farms, undergo rigorous HPLC testing to ensure purity and potency."
- • "The Omega-3 softgels from Nordic Naturals are third-party tested for heavy metals at SGS laboratories in Switzerland."
- • "Each batch of our patented Curcumin C3 Complex is validated using LC-MS/MS methods for authenticity and concentration."
- • "We source Glucosamine Sulfate raw materials exclusively from European suppliers who comply with GMP certification standards."
- • "Our Vitamin D3 supplements are derived from lanolin, verified through UV spectrophotometry to meet label claims accurately."
2.2. Context seed: from domain-specific terms
"Please generate 10 English sentences describing pharmaceuticals or health supplements, each containing the term "Ofloxacin".
Make sure that each sentence is a single sentence and use as different drug names, supplement names, named entities, technical terms, numbers, and contexts as possible.
Each sentence should be output on its own line, separated by line breaks.
Ensure that each sentence is at least 5 characters and at most 200 characters long.
The main language is English.
Use the terms of Ofloxacin itself, do not change the language for the term of Ofloxacin.
Annotations, explanations, or repetitions are not needed. "
- • Ofloxacin is commonly prescribed to treat bacterial infections such as urinary tract infections and respiratory tract infections.
- • "The pharmacokinetics of Ofloxacin show rapid absorption with peak plasma concentrations occurring within one to two hours."
- • "Doctors recommend Ofloxacin 200 mg twice daily for seven days to effectively manage acute bacterial exacerbations of chronic bronchitis. "
- • "Ofloxacin ophthalmic solution 0.3% is used to treat bacterial conjunctivitis and other eye infections caused by susceptible strains."
- • "In a clinical trial, Ofloxacin demonstrated a 90% success rate in eradicating Helicobacter pylori when combined with other antibiotics."
4. Paraphrasing
"Rewrite and paraphrase the sentence of Ofloxacin is commonly prescribed to treat bacterial infections such as urinary tract infections and respiratory tract infections., describing pharmaceuticals or health supplements, into 9 different versions.
Make sure that the sentences use as different drug names, supplement names, named entities, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Ofloxacin.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Ofloxacin is frequently utilized in pharmaceutical treatments for respiratory ailments and infections of the urinary system."
- • "For various bacterial infections like infections of the kidneys and lungs, Ofloxacin is a commonly recommended antibiotic."
- • "In the realm of antibiotics, Ofloxacin is a popular choice to combat bacterial illnesses affecting the urinary and respiratory tracts."
- • "Physicians often prescribe Ofloxacin to address bacterial infections, particularly those impacting the respiratory and urinary systems."
- • "Commonly, Ofloxacin is administered to patients suffering from respiratory tract infections as well as urinary bladder infections caused by bacteria."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 14799 | Has HealthyBend joint supplement manufacturing ceased? |
| 11924 | Will stopping Lovaza after extended usage trigger withdrawal symptoms? |
| 11541 | Does Nature's Bounty Omega Three supplement contain gluten? |
| 11258 | Red yeast rice capsules sold as Hypocol often mandate membership registration. |
| 10908 | Can Hu Gan Ning Granules cause stomach discomfort? |
1. Domain seed
"请生成{n}个描述药品或保健品的情境或场景。
请注意，你应生成的是发言的情境或场景，而不是实际的发言内容。
请不要在正文中包含任何解释性说明或注释，只需用中文输出这{n}个情境或场景。
请尽可能生成多样化的情境或场景。"
- • "奶茶店推荐胶原蛋白饮品"
- • "腰椎病患者尝试新型止痛贴。"
- • "爱心公益活动为灾区捐赠药品"
- • "中老年夫妻学习保健品服用顺序。"
- • "环保志愿者户外必备防中暑药品"
2.1. Context seed: from scenarios
"请根据以下情境生成10个描述医药品或健康补充剂的中文句子。
情境：奶茶店推荐胶原蛋白饮品
请确保每个句子都尽可能多地包含不同的药品名称、保健品名称、专有名词和专业术语。
请确保每个句子的长度不少于5个字符且不超过100个字符。
无需注释、解释或重复。
每个句子请单独占一行，并用换行符分隔。"
- • "本店推荐富含水解胶原蛋白肽的美容饮品，搭配维生素C促进胶原合成。"
- • "饮品中特含鱼胶原蛋白和透明质酸，助力皮肤保湿与弹性修复，适合日常补充。"
- • "结合葡萄籽提取物和肽类复合物的胶原蛋白奶茶，有助于抗氧化与延缓衰老。"
- • "本款胶原蛋白饮料含有多肽胶原蛋白、辅酶Q10以及核酸，支持皮肤细胞再生。"
- • "富含猪皮胶原蛋白和氨基葡萄糖的健康饮品，改善关节健康同时增强肌肤活力。"
2.2. Context seed: from domain-specific terms
"请生成10个包含术语Ofloxacin的中文句子，每个句子用于描述医药品或健康补充剂。
请确保每个句子都是单句，并尽可能使用不同的药品名称、保健品名称、专有名词、专业术语、数字和语境。
每个句子请单独占一行，并用换行分隔。
每个句子的长度不少于5个汉字且不超过100个汉字。
主要语言为中文。
Ofloxacin一词请保持原语言，不要翻译。
不需要添加注释、说明或重复。"
- • "Ofloxacin是一种广谱抗菌药物，常用于治疗呼吸道感染和尿路感染。"
- • "临床研究表明，Ofloxacin对革兰阴性菌具有显著抑制作用。"
- • "患者在服用Ofloxacin期间应避免同时使用含镁或铝的抗酸药。"
- • "Ofloxacin胶囊的推荐剂量通常为每天200毫克至400毫克。"
- • "药品说明书建议孕妇慎用Ofloxacin，以防止对胎儿产生潜在风险。"
3. Multilingual Prompting: Translation
"Translate to English, the Chinese sentence of {sentence}, describing pharmaceuticals or health supplement.
Write all numbers in spelled-out English words instead of using digits.
Output only the results with one line per conversation without line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "Ofloxacin is a broad-spectrum antibiotic commonly used to treat respiratory infections and urinary tract infections."
- • "Clinical studies have shown that Ofloxacin has a significant inhibitory effect on Gram-negative bacteria."
- • "Patients should avoid using antacids containing magnesium or aluminum simultaneously while taking Ofloxacin."
- • "The recommended dosage of Ofloxacin capsules is usually two hundred to four hundred milligrams per day."
- • "The pharmaceutical instructions recommend that pregnant women use Ofloxacin with caution to prevent potential risks to the fetus."
4. Paraphrasing
"Rewrite and paraphrase the sentence of Ofloxacin is a broad-spectrum antibiotic commonly used to treat respiratory infections and urinary tract infections., describing pharmaceuticals or health supplements, into 9 different versions.
Make sure that the sentences use as different drug names, supplement names, named entities, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Ofloxacin.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Dispertab formulations improve the absorption rate of ibuprofen in the digestive system while minimizing adverse reactions."
- • "The use of Dispertab technology in curcumin supplements can greatly accelerate drug dissolution in the GI tract and lessen negative effects."
- • "Dispertab tablets significantly boost the solubility of metformin in gastrointestinal fluids and help in decreasing unwanted side effects."
- • "By employing Dispertab for vitamin D supplements, dissolution in the stomach is enhanced and potential side effects are reduced."
- • "Dispertab enhances the rate at which aspirin dissolves in the intestines, contributing to faster therapeutic effects and fewer side effects."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 14799 | Has HealthyBend joint supplement manufacturing ceased? |
| 11924 | Will stopping Lovaza after extended usage trigger withdrawal symptoms? |
| 11541 | Does Nature's Bounty Omega Three supplement contain gluten? |
| 11258 | Red yeast rice capsules sold as Hypocol often mandate membership registration. |
| 10908 | Can Hu Gan Ning Granules cause stomach discomfort? |
1. Domain seed
"医薬品や健康サプリメントの説明における，発話する状況・シナリオを日本語で{n}個生成して。
発話そのものではなく，発話の状況・シナリオであることに注意して。
本文の注釈は不要で，{n}個の状況・シナリオのみを日本語で出力して。
できるだけ多様な状況・シナリオを生成してください。"
- • "子育て中の母親がママ友にサプリメントの安全性について説明する場面"
- • "メール相談で薬剤師が市販風邪薬と処方薬の併用について説明する場面"
- • "テレビ番組の健康特集で、ゲスト医師が医薬品とサプリの違いを解説する場面"
- • "自律神経失調シンドローム患者向けサプリ説明会"
- • "健康食品店で販売員がオメガ3サプリの効能を紹介する場面"
2.1. Context seed: from scenarios
"医薬品や健康サプリメントの説明における以下のシナリオに従って、{n_generate}個の日本語の文を生成してください。
シナリオ: 子育て中の母親がママ友にサプリメントの安全性について説明する場面
各文にはできるだけ多様な医薬品名やサプリメント名、固有表現、専門用語を含めてください。
各文は5文字以上100文字以下であることを必ず守ってください。
注釈、説明、繰り返しは不要です。
各文は1行ずつで出力し，それぞれ改行してください。"
- • "私は毎日、鉄分とカルシウムが豊富な「ネイチャーメイド マルチビタミン」で栄養補給しています。"
- • "DHAやEPAを含む「ディアナチュラ スピルリナ」は脳の健康維持に効果的で安心です。"
- • "小児用ビオフェルミン」は子どもの腸内環境を整え、安全性も高くおすすめです。"
- • "免疫力を高めるために「アサヒ クリアフォーラム ビタミンC」を使っています。"
- • "「ミルクシスルエキス」は肝機能サポートに優れていて、副作用の心配がほとんどありません。"
2.2. Context seed: from domain-specific terms
"医薬品や健康サプリメントの説明において、Ofloxacinという用語を含む10個の日本語の文を生成してください。 各文は1文のみとし、できるだけ異なる医薬品名やサプリメント名，固有表現や専門用語、数字、文脈を使ってください。 各文は1行ずつで出力し，それぞれ改行してください。 各文は5文字以上100文字以下であることを必ず守ってください。 主な言語は日本語です。 Ofloxacinという用語自体を使用し、その用語の言語は変更しないでください。 注釈、説明、繰り返しは不要です。"
- • "オフロキサシン(Ofloxacin)は、一般細菌性副鼻腔炎の治療に使用されることがあります。"
- • "Ofloxacinはニューキノロン系抗菌薬の一種であり、幅広い抗菌スペクトルを持つことで知られています。"
- • "耳感染症の治療において、Ofloxacin点耳液は有効な選択肢となることがあります。"
- • "眼科領域では、Ofloxacin点眼液が細菌性結膜炎の治療に使用されます。"
- • "抗菌薬Ofloxacinは、呼吸器感染症の治療にも効果を発揮します。"
3. Multilingual Prompting: Translation
"Translate to English, the Japanese sentence of {sentence}, describing pharmaceuticals or health supplement.
Write all numbers in spelled-out English words instead of using digits.
Output only the results with one line per conversation without line breaks.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed. "
- • "Ofloxacin is sometimes used for the treatment of general bacterial sinusitis."
- • "Ofloxacin is a type of new quinolone antibacterial agent and is known to have a broad antibacterial spectrum."
- • "Ofloxacin otic solution may be a valid option in the treatment of ear infections."
- • "In the ophthalmology field, ofloxacin eye drops are used for the treatment of bacterial conjunctivitis."
- • "The antibacterial drug Ofloxacin is also effective in the treatment of respiratory infections."
4. Paraphrasing
"Rewrite and paraphrase the sentence of Ofloxacin is sometimes used for the treatment of general bacterial sinusitis.., describing pharmaceuticals or health supplements, into 9 different versions.
Make sure that the sentences use as different drug names, supplement names, named entities, technical terms, numbers, the order of words, and contexts as possible.
Write all numbers in spelled-out English words instead of using digits.
Do not change or replace the words of Ofloxacin.
Ensure that each sentence is a single sentence and is at least 5 characters and at most 200 characters long.
Annotations, explanations, or repetitions are not needed.
Each sentence should be output on its own line, separated by line breaks."
- • "Ofloxacin is occasionally prescribed to address bacterial sinus infections in adult patients."
- • "Ofloxacin may be utilized in treating mild bacterial sinusitis as part of outpatient therapy."
- • "Ofloxacin is sometimes recommended for managing bacterial sinus infections caused by common pathogens."
- • "Ofloxacin has limited use in combating bacterial sinus infections resistant to other antibiotics."
- • "Ofloxacin is prescribed in certain cases for addressing acute bacterial sinusitis symptoms."
6. Filtering
The code for filtering is available at [URL].
| Perplexity | Text |
|---|---|
| 14799 | Has HealthyBend joint supplement manufacturing ceased? |
| 11924 | Will stopping Lovaza after extended usage trigger withdrawal symptoms? |
| 11541 | Does Nature's Bounty Omega Three supplement contain gluten? |
| 11258 | Red yeast rice capsules sold as Hypocol often mandate membership registration. |
| 10908 | Can Hu Gan Ning Granules cause stomach discomfort? |
Phonetic Respelling Augmentation
This section shows example prompts and respelled texts with PRA on each dataset. Each text was synthesized by three TTS systems; kokoro-TTS, Azure TTS, Kyutai TTS.
"Please convert the following English sentence into a pronunciation respelling.
Reflect possible and various pronunciation errors, substitutions, or idiosyncrasies.
・Use English letters (not IPA) to represent each word as it would sound , and do not use hyphens.
・Separate words using spaces as in normal English.
・Do not use IPA, but use clear respelling so that a reader can pronounce it with the intended accent.
・Annotations, explanations, or repetitions are not needed.
・Output only the results with one line without line breaks.
Sentence: {sentence}"
kokoro-TTS
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The airport staff checked the Turkair schedule. | Ze airpot stahf chekt ze Turker skedjool. | ||
| Pilot Chen, Airbus Three Twenty, proceed to Paris, class Bravo airspace. | Pailot Chen, Airbus Sree Twenny, pruhseed to Paris, klass Bravoh airspays. | ||
| Air France One One Two, report over waypoint BALTR heading to Oslo. | Eh Frons wun wun too, repor ova waypoyn BALTR hedding tu Osslo. | ||
| In aviation management, European regulation is strict. | In ayveeay shun manaj ment yooro peean regyoo lay shun iz strik. | ||
| With devec tools, Paris airport reduces risks for Seven helicopter pilots. | Vit devec tools, Paris airport redooses risks for Sebben helicopter pailots. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| British Airways Seven Six Five keep level Two Four Zero. | British Airweys Sevem Six Fife keep levul Two For Zero | ||
| American Airlines flight five five zero contacted Dallas center regarding light rime ice encounters. | Amerikin Airlynz flite fife fife zero contacted Dallas sentur regarding lite rime ice encounters. | ||
| Can Captain Li Four land on Seven Charlie runway, Frankfurt Tower? | Kan Captn Lee For land on Seven Charlee runwye Frankfert Tower? | ||
| Airbus Three Twenty began approach following airclub instructions. | Airbus Three Twenny began approach follwing airclub instruckshuns. | ||
| KLM Four Triple Seven experiences cargo shift, announces Amsterdam control. | kay ell em for tripul sevin ekspeerienziz kargo shift, uhnaunsiz amsturdam control. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The supreme court justices expressed differing opinions about the "hansaria." | The supreem kort justisiz eksprest difering opinyuns about the hansaria. | ||
| A landmark decree by India's highest tribunal shaped legal views on potability limits concerning mercury levels. | A landmaark dikree by Indiyas highest trybunal sheypt leegal views on potability limts concerning mercury levls. | ||
| The diary presented as evidence reflects the defendant's inner transformation. | The daiyari prezented az evidens refleks the difandents inner transformayshan. | ||
| Before the Supreme Court, Chinnaiah's defence team made a passionate representation. | Befor the Supreme Kort, Chinnaiahz difens teem made a pashunet representeishun. | ||
| The psychological evaluation report raises concerns about a diagnosed impulse control issue. | Tha sykolojikul evualayshun repor rayzez konzerns about a diagnozed impulz control ishoo. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The supplement CalD3 supports bone health and density. | Tha supplemen Kal Dee Three supportz bone helth and dencity. | ||
| Can you describe how Norvir interacts with Sandimmune? | Can yu describe how Norveer interaks with Sandimune. | ||
| The health program featured multiple oral liq brands containing Axnil. | Tha helth program feechurd multiple oral lick brands containin Axnil. | ||
| The trend of consuming magnesium and calcium enriched vitamins persists. | Tha trend of konsummin magneezhum and kalshum enricht vitamins persists. | ||
| The patient's supplement protocol was enhanced with three distinct vitamin D compounds. | Tha payshent's supplemen protokol wuz enhanst with three distinkt vitamin Dee kompounds. |
Azure TTS
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The airport staff checked the Turkair schedule. | Ze airpot stahf chekt ze Turker skedjool. | ||
| Pilot Chen, Airbus Three Twenty, proceed to Paris, class Bravo airspace. | Pailot Chen, Airbus Sree Twenny, pruhseed to Paris, klass Bravoh airspays. | ||
| Air France One One Two, report over waypoint BALTR heading to Oslo. | Eh Frons wun wun too, repor ova waypoyn BALTR hedding tu Osslo. | ||
| In aviation management, European regulation is strict. | In ayveeay shun manaj ment yooro peean regyoo lay shun iz strik. | ||
| With devec tools, Paris airport reduces risks for Seven helicopter pilots. | Vit devec tools, Paris airport redooses risks for Sebben helicopter pailots. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| British Airways Seven Six Five keep level Two Four Zero. | British Airweys Sevem Six Fife keep levul Two For Zero | ||
| American Airlines flight five five zero contacted Dallas center regarding light rime ice encounters. | Amerikin Airlynz flite fife fife zero contacted Dallas sentur regarding lite rime ice encounters. | ||
| Can Captain Li Four land on Seven Charlie runway, Frankfurt Tower? | Kan Captn Lee For land on Seven Charlee runwye Frankfert Tower? | ||
| Airbus Three Twenty began approach following airclub instructions. | Airbus Three Twenny began approach follwing airclub instruckshuns. | ||
| KLM Four Triple Seven experiences cargo shift, announces Amsterdam control. | kay ell em for tripul sevin ekspeerienziz kargo shift, uhnaunsiz amsturdam control. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The supreme court justices expressed differing opinions about the "hansaria." | The supreem kort justisiz eksprest difering opinyuns about the hansaria. | ||
| A landmark decree by India's highest tribunal shaped legal views on potability limits concerning mercury levels. | A landmaark dikree by Indiyas highest trybunal sheypt leegal views on potability limts concerning mercury levls. | ||
| The diary presented as evidence reflects the defendant's inner transformation. | The daiyari prezented az evidens refleks the difandents inner transformayshan. | ||
| Before the Supreme Court, Chinnaiah's defence team made a passionate representation. | Befor the Supreme Kort, Chinnaiahz difens teem made a pashunet representeishun. | ||
| The psychological evaluation report raises concerns about a diagnosed impulse control issue. | Tha sykolojikul evualayshun repor rayzez konzerns about a diagnozed impulz control ishoo. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The supplement CalD3 supports bone health and density. | Tha supplemen Kal Dee Three supportz bone helth and dencity. | ||
| Can you describe how Norvir interacts with Sandimmune? | Can yu describe how Norveer interaks with Sandimune. | ||
| The health program featured multiple oral liq brands containing Axnil. | Tha helth program feechurd multiple oral lick brands containin Axnil. | ||
| The trend of consuming magnesium and calcium enriched vitamins persists. | Tha trend of konsummin magneezhum and kalshum enricht vitamins persists. | ||
| The patient's supplement protocol was enhanced with three distinct vitamin D compounds. | Tha payshent's supplemen protokol wuz enhanst with three distinkt vitamin Dee kompounds. |
Kyutai TTS
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The airport staff checked the Turkair schedule. | Ze airpot stahf chekt ze Turker skedjool. | ||
| Pilot Chen, Airbus Three Twenty, proceed to Paris, class Bravo airspace. | Pailot Chen, Airbus Sree Twenny, pruhseed to Paris, klass Bravoh airspays. | ||
| Air France One One Two, report over waypoint BALTR heading to Oslo. | Eh Frons wun wun too, repor ova waypoyn BALTR hedding tu Osslo. | ||
| In aviation management, European regulation is strict. | In ayveeay shun manaj ment yooro peean regyoo lay shun iz strik. | ||
| With devec tools, Paris airport reduces risks for Seven helicopter pilots. | Vit devec tools, Paris airport redooses risks for Sebben helicopter pailots. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| British Airways Seven Six Five keep level Two Four Zero. | British Airweys Sevem Six Fife keep levul Two For Zero | ||
| American Airlines flight five five zero contacted Dallas center regarding light rime ice encounters. | Amerikin Airlynz flite fife fife zero contacted Dallas sentur regarding lite rime ice encounters. | ||
| Can Captain Li Four land on Seven Charlie runway, Frankfurt Tower? | Kan Captn Lee For land on Seven Charlee runwye Frankfert Tower? | ||
| Airbus Three Twenty began approach following airclub instructions. | Airbus Three Twenny began approach follwing airclub instruckshuns. | ||
| KLM Four Triple Seven experiences cargo shift, announces Amsterdam control. | kay ell em for tripul sevin ekspeerienziz kargo shift, uhnaunsiz amsturdam control. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The supreme court justices expressed differing opinions about the "hansaria." | The supreem kort justisiz eksprest difering opinyuns about the hansaria. | ||
| A landmark decree by India's highest tribunal shaped legal views on potability limits concerning mercury levels. | A landmaark dikree by Indiyas highest trybunal sheypt leegal views on potability limts concerning mercury levls. | ||
| The diary presented as evidence reflects the defendant's inner transformation. | The daiyari prezented az evidens refleks the difandents inner transformayshan. | ||
| Before the Supreme Court, Chinnaiah's defence team made a passionate representation. | Befor the Supreme Kort, Chinnaiahz difens teem made a pashunet representeishun. | ||
| The psychological evaluation report raises concerns about a diagnosed impulse control issue. | Tha sykolojikul evualayshun repor rayzez konzerns about a diagnozed impulz control ishoo. |
| Text | Synthesized Audio | Respelling Text | Synthesized Respelling Audio |
|---|---|---|---|
| The supplement CalD3 supports bone health and density. | Tha supplemen Kal Dee Three supportz bone helth and dencity. | ||
| Can you describe how Norvir interacts with Sandimmune? | Can yu describe how Norveer interaks with Sandimune. | ||
| The health program featured multiple oral liq brands containing Axnil. | Tha helth program feechurd multiple oral lick brands containin Axnil. | ||
| The trend of consuming magnesium and calcium enriched vitamins persists. | Tha trend of konsummin magneezhum and kalshum enricht vitamins persists. | ||
| The patient's supplement protocol was enhanced with three distinct vitamin D compounds. | Tha payshent's supplemen protokol wuz enhanst with three distinkt vitamin Dee kompounds. |