How I Translated a Novel with ChatGPT
Xavi Arderiu
Business Manager | CEO | CMO | Mediapro | Danone | Unilever | ESADE | Harvard CS50 | GOOGLE | AWS | AZURE | Leadership | Digital Transformation | Global Marketing & Local Implementation | Full P&L | MBI | StartUp
In the summer of 2023, I proudly unveiled "The Sagrada Familia Hunt," a thrilling translation of my Spanish novel, originally published as "Enigma Gaudi" in 2018.
This article delves into the journey of how I accomplished this feat in record time, thanks to the knowledge I've accumulated over the years through extensive experimentation with various AI tools, with a significant boost from ChatGPT.
Early Ventures into AI Translation
It all began in 2018 when I self-published a thrilling Spanish novel titled "Enigma Gaudí." Intrigued by the potential of expanding its reach to English-speaking audiences, I decided to embark on the challenging task of translating it.
Fast forward to 2019, I experimented with two distinct Artificial Intelligence tools - Amazon Web Services' Translator and DeepL, a European-based company renowned for its linguistic prowess. While both exhibited exceptional performance, I opted against Google Cloud and Microsoft Azure due to the lack of feasible free or scalable options.
Translating a novel is a complex process. It involves thousands of words intricately weaving characters and evolving environments into a cohesive narrative. To maintain the overall story's coherence, I realized that translating it piecemeal was a flawed strategy. Instead, I aimed for a holistic approach.
The initial attempts yielded promising results. Armed with two translations, I began refining the text delivered by AWS, leveraging DeepL to rectify any unnatural expressions. However, a substantial gap remained between a proper translation and the literal word-for-word translation. This arose from two core challenges.
First, the translations often retained Spanish idioms that didn't align with English expressions. For instance, phrases like "my eyes went white" in Spanish needed to be meticulously transformed into native English expressions like "my eyes rolled back." This task was both time-consuming and required extensive internet research, as I, a non-native English speaker, sought the most suitable translations.
The second issue stemmed from the inherent nature of the Spanish language, which often lacks determinants associated with verbs. This resulted in inconsistencies that rendered the text repetitive and monotonous, necessitating extensive revisions.
These dual challenges demanded significant manual intervention, slowing down progress. By the end of 2019, having translated only a quarter of the story, I temporarily halted the project, as it failed to meet the standards I envisioned for this creative endeavor.
2022: Emergence of Large Language and Diffusion Models
Over the following years, I engaged in other projects and training, while the Spanish version of my book garnered a modest following reaching a thousand readers. However, the breakthrough came in the summer of 2022.
The AI community was abuzz with generative AI's potential, primarily focused on visual creations through Diffusion Models. My exploration initially revolved around tools such as Midjourney, Dall-e, and Stable Diffusion, predominantly in the realm of visual arts.
However, a pivotal moment arrived in November when ChatGPT entered the scene. While my initial interactions with ChatGPT involved rudimentary Python code to solve the Harvard's CS50 final project course which I was taking, I swiftly recognized the tool's untapped potential at that time, even as a novice coder.
Subsequently, I delved into the world of ChatGPT APIs, utilizing them to craft my final project itself: a "voice to blog" web application, seamlessly integrating AssemblyAI and OpenAI APIs. This innovative platform allowed users to record their voice, transcribe content with AssemblyAI, enhance it using OpenAI's capabilities, and generate a concise summary and captivating headline. To complement the blog post, I enlisted Dall?e, another OpenAI tool, to create descriptive images. The results astounded me, considering it was my initial foray into Python and JavaScript. I successfully created a platform that enabled individuals without writing skills to contribute to local newspapers at a hyper-local level (i.e. at a Zip Code level).
This "Voice2Blog" experiment proved pivotal, as it provided valuable insights into harnessing ChatGPT's power effectively. In December 2022, equipped with newfound proficiency, I self-published "Powerland" without AI assistance, adhering to my belief in preserving the integrity of human-created content.
Summer 2023: Applying AI to Translation without Sacrificing Coherence
As we approach 2026, a significant year for the Sagrada Familia, marking the centenary of Antoni Gaudí's mysterious passing, I contemplated again the task of translating my 40-chapter, +75k word book, "Enigma Gaudí". While it may be a relatively short novel, the translation presented a formidable challenge for ChatGPT.
Long story short, incorporating my expertise in prompt engineering, I devised a novel approach. I segmented the text into 300-word sections, meticulously refining the content using a secret formula, effectively doubling its size. I then initiated the translation process, closely monitoring the results on a dedicated website. To enhance reader engagement, I supplemented each chapter with an AI-generated image, courtesy of Stable Diffusion.
领英推荐
Moving to the details, the first step of text segmentation was straightforward, involving a simple word count tool within Word. The secret sauce lay in constructing a comprehensive prompt chain, facilitating consistent text enhancement across numerous iterations.
This prompt chain comprised three core elements, expertly woven to guide ChatGPT. The first prompt established context and delivered three vital instructions: expanding the text, imbuing it with the style of my literary idol, Graham Greene, and considering the insights from a professional review obtained during my initial self-publishing endeavor in 2018. To ensure clarity, the prompts were enclosed within triple brackets for the text, ensuring ChatGPT's comprehension.
The second prompt, executed after ChatGPT's initial response, focused on doubling the word count. ChatGPT excelled at enriching descriptions, adjectives, and scene transitions within paragraphs, without introducing new characters, plots, or scenarios.
Only in the third prompt did I task ChatGPT with translating the text into English. This iterative process yielded a consistently refined English translation, preserving the narrative's integrity.
Copiously employing these prompts, I achieved a consistent translation throughout the entire book.
Following the translation, I conducted a meticulous self-review, eventually publishing the novel on Wattpad. This platform, often dubbed "the YouTube for writers," allowed me to present the text chapter by chapter, encouraging reader feedback and facilitating fine-tuned corrections.
Intriguingly, I ventured into Facebook ads for niche market research, leading me to reconsider the book's title. It became evident that "The Sagrada Familia Hunt" held broader appeal among English-speaking readers.
In conclusion, the summer of 2023 was far from dull. This journey exemplifies the symbiotic relationship between human creativity and AI-driven productivity. While "The Sagrada Familia Hunt" may not be an entirely AI-authored book, AI played a pivotal role in enhancing productivity, ultimately enriching the translation and image creation processes.
Takeaways:
More than a century ago, a symbol began rising in the heart of Barcelona, its significance still unfolding. The role of investment funds in shaping the world's future and the potential emergence of Demopoly as a force in governance continue to fuel fascinating debates.
If you're curious to explore the result, the thriller is ready in Amazon.
Undoubtedly, AI is here to stay, revolutionizing art, knowledge, and content creation. Just as post-production retouching is seamlessly woven into various creative mediums, AI's influence is destined to become an inherent part of the creative process. As we embrace AI's potential across various domains, the lines between human and AI contributions will blur, and we, as creators, will continue pushing the boundaries of creativity.
Credits: Image: Stable Diffusion with the prompt I prepared with ChatGPT (Craft a hyperrealistic photography in 8k resolution with perfect proportions. A photograph taken with a professional CANON EOS that shows a woman's hand, realistically detailed, using a classical "fountain pen" on a modern "iPad screen". Blend photography and cinematic styles, adding Unreal Engine elements for impact. Achieve UHD quality.) Text: Mine but with asking ChatGPT to improve my English expressions to seem a native English speaker.
New book: "The AI Revolution in Book Publishing: A Concise Guide to Navigating Artificial Intelligence for Writers and Publishers"
7 个月Fascinating and important!
Director | Innovation | Web3 | XR | Digital Business & Transformation | Entrepreneur | Startups | Metaverse | Virtual Economy | NFTs | Blockchain | Corporate Venturing
1 年Wow, Xavi!! Such a great job exploring.and connecting technologies. Your article is a fantastic (and easy to understand for anyone) description of the process of AI integration in our daily lifes, not only as professionals but also as people enjoying their hobbies. Congrats! ??