Language AI has reached an inflection point. As the latest AI advances break into the mainstream (case in point: ChatGPT), a new wave of startups have begun to build products and business models on top of language AI’s cutting-edge capabilities.
The Slator Language AI 50 Under 50 List features fifty of the most exciting language AI companies founded within the last fifty months. The list provides a cross-section of new entrants in the rapidly transforming language technology landscape. Selected companies range from early stage startups to well-funded platforms, and from companies with a single narrow use case, to fully integrated, end-to-end solutions.
The breakthrough advances in language AI promise to reshape the language services industry for years to come. AI will also reconfigure the wide range of industries that currently drive demand for language services and technology. The Slator Language AI 50 Under 50 List provides a snapshot of some of the earliest, notable language AI and technology companies in this space.
We will release the Slator Language AI 50 Under 50 List annually and listing is open to AI and technology companies with language and / or multilingual capabilities, with a founding date within the preceding 50 months. Companies who are interested in appearing in the 2024 edition may contact Slator.
The 2023 50-under-50 companies have been classified into the following eight categories:
- Al Agency
- Language Tech Platform
- Al Writing
- NLP Platform
- Al Dubbing
- Voice Al Toolkit
- Synthetic Video
A description of each category is included in the tables below. Each card expands to display more information about each 50-under-50 company.
AI Agencies offer AI-powered translation and speech recognition (ASR) technology via expert-in-the-loop platforms. Companies such as XL8 and Bering Lab, focus on specific sectors — media / entertainment and legal / financial content, respectively.
Many, like AI Communis, Devnagri, and Kanari AI, have built systems designed for a specific group of languages. Most operate a SaaS platform, while some also offer an on-premise solution and cloud-based APIs.
The AI agency model is one typified by the likes of Lilt and Unbabel, which were founded in 2015 and 2013, respectively.
Although it is possible for language service providers (LSPs) that pre-date the use of AI in translation workflows to pivot to an AI-first approach and transform into an AI agency, the latest AI agencies have been founded on an AI-first approach.
Language Tech Platforms
Language tech platforms offer technology-based solutions to support translation and interpreting. The category includes interpreting platforms, such as remote simultaneous interpreting (RSI) company Qonda and multilingual video call platform Viva Translate. Providers of translation management and workflow automation, such as AI-powered workflow automation tool, Blackbird, are also included.
The Language Tech category is the smallest category of the 50 Under 50. This is not because there is a scarcity of language technology companies but rather because it is a relatively “mature” category; The vast majority of potential candidates were founded prior to the 50 month cut-off.
AI Writing tools use generative AI capabilities to create text, based on a series of prompts. Companies typically provide writing for a range of content types, such as marketing and social media content, emails, and other business communications. Many, including Copy AI and WriteSonic, are built on top of large language models (LLMs) such as GPT-3, while others, like Typewise, provide customized language models.
The relevance of AI writing to the language industry is found in the fact that AI writers offer users the option to generate content in multiple languages — ranging from 25 languages (TextCortex AI) to “nearly any language” (HyperWrite). With translation companies now theoretically having the option of using AI writers to instantly generate content in their desired (target) languages — and thereby bypass translation —, we unpacked the question of whether AI writers represent a threat or an opportunity to LSPs in this September 2022 article.
Data-for-AI companies collect and annotate (or synthesize) data for model training. 50-under-50 data-for-AI companies all work with text and / or speech data and may also offer services for other types of data, such as video and image. Audio Bee, for example, builds speech training datasets that are used to build speech recognition (ASR) systems. Companies either collect and annotate real-world data or, as with Betterdata and Replica Analytics, produce synthetic data from limited amounts of organic data, which results in more data and removes privacy concerns.
These data-for-AI companies are often viewed as AI tooling companies — acting as “shovels-in-a-gold-rush” to collect, prepare, or create the data that underpins AI model training.
Natural language processing (NLP) platforms either develop and provide access to their own large language models (LLMs) or give users access to, and additional finetuning of, third-party LLMs. Some, like Gantry AI, aim to complement initial model development and training with a rigorous refinement process. Adept.ai wants to synthesize transformer models with existing software tools and APIs. NLP Cloud, meanwhile, bills itself as an “out-of-the-box” solution that lets companies access a range of open-source language models and use them in production.
NeuralSpace is a SaaS platform offering NLP APIs to engineers and specializes in traditionally low-resource languages spoken across Asia, the Middle East, and Africa. Here’s CEO Felix Laumann on SlatorPod. Germany’s Kern AI helps companies to embed NLP into their products and processes by enabling developers to prototype and implement full workflows that understand textual data. OpenAI rival Cohere, meanwhile, created its own LLMs which can be easily accessed via API across many cloud platforms, turning an expensive, multi-step process into an easy-to-use interface. Here’s Cohere Co-founder Nick Frosst on SlatorPod.
AI dubbing companies combine the latest advances in speech synthesis and translation. Not yet ready for prime time at major Disney+ or Netflix shows, AI dubbing enables access to content previously unavailable to hundreds of millions of users of the world’s social media and other content platforms. VOISEED, for example, offers a virtual studio for professional AI dubbing. India’s NeuralGarage says it complements the dubbing process, using AI to improve lip sync and facial movements of actors in dubbed content by removing “visual dissonance.” Indian rival Dubverse promises to allow users to dub videos into multiple languages in “almost real time.” Here’s Dubverse co-founder Anuja Dhawan on SlatorPod.
Fellow India-based startup Dubdub says it is an AI-based web tool that enables audio / video dubbing in over 50 languages without any language expertise. Actually taking a stab at servicing major broadcasters, media companies, and content distributors with AI dubbing is Israel’s Deepdub. And Canada’s Blanc builds deep video translation algorithms that enable people to speak any language on video.
Voice AI Toolkit
Voice AI Toolkit companies offer “virtual studios” for producing audio by integrating AI language technology with automated media production. Aflorithmic, for example, lets you turn text into “beautiful audio” in seconds. ElevenLabs is an AI voice synthesis software for creators and publishers — it is also planning to launch AI dubbing later in 2023. Sanas, meanwhile, is building a technology for real time accent translation.
Murf AI lets users manually adapt voiceover features, such as pitch, speed, timing, and emphasis. South Korea’s Supertone offers similar features, including text-to-speech, ASR, voice cloning, and NLP technology, alongside music-based AI such as singing voice synthesis and melody / lyrics transcription.
Language is also a key component of video, of course. Synthetic Video companies offer “virtual studios” for producing video by integrating AI language technology with automated media production. Elai is a synthetic video production platform and API for generating videos with digital presenters — apparently competing with Synthesia, which was founded in 2017 and, therefore, didn’t make the 50 months cut-off.
Fliki, too, integrates the full set of text-to-speech and text-to-video tech to integrate a range of capabilities into a single “virtual studio.” Hour One’s virtual presenters, meanwhile, speak in 60 different languages. Movio says its users can create their own custom avatar in just three minutes, and the platform offers text-to-speech in 200+ voices across 20+ languages. Wisecut leverages generative AI and speech recognition to edit videos automatically. Then there’s the multilingual AI presenters offered by Colossyan and DeepWord, while Inworld AI promises intelligent virtual characters for video games and NeuralLoom calls its presenters “digital employees.” UK-based Yepic is an AI video toolkit for creating, dubbing, and personalizing videos. Finally, Animatr AI lets you create short videos using English-language prompts.