The Current
by Hunter WorlandJul 17, 2024
In the late 19th century, an ophthalmologist from a polyglottic outpost of the Russian Empire invented Esperanto – a common second language that would eliminate linguistic barriers. Its community of esperantists dreamed their language could facilitate international commerce and forge a “a common brotherhood” between nations. The language gained millions of speakers but ultimately did little to remove the linguistic friction between markets. Can AI apps?
To examine this, we leveraged our consumer survey panel to investigate LLM translation efficacy, challenges among current language learners, and relative value between applications. Our research was deliberately outside the Anglosphere. English’s role as the global language makes it a less interesting market since the need for translation or language learning is inherently lower. As a case study, we used Germany – not only the largest market in Europe but also one with one of the highest incidence of English-language professional and educational requirements.
I. Language exam
We pitted the reigning champion of consumer application translation, Google Translate (which leverages its own form of AI, neural machine translation), against OpenAI's GPT-4o and Anthropic's Claude 3.5 Sonnet. Our judges were bilingual consumers fluent in both German and English.
The respondents selected the most natural translation across three passages: an idiomatic text, a technical text, and a conversational text.
The two LLMs essentially matched Google Translate across the passages with no statistically significant performance gap. Yet, a three-way tie for Google Translate is really a loss. If underlying translation technology is held equal, generative AI platforms certainly win on product experience as they leverage contextual understanding to provide more nuanced, coherent, and adaptable translations than Google Translate's primarily statistical neural machine translation. approach.
II. Pain Points
Our second survey targeted German learners of any foreign language. Surprisingly, the most significant pain point for language learners wasn't traditional pedagogical challenges like spelling or reading comprehension, but speech, including accent and pronunciation. Similarly, the largest unlock they envisioned was increased interaction with native speakers.
This presents a natural opportunity for generative AI disintermediation, both in product experience and underlying technology. The models can dynamically generate contextually appropriate language samples, simulating the diversity of real-world interactions. This capability allows for the coverage of a broader range of scenarios, including niche and informal use cases that often elude traditional pedagogical frameworks.
Pursuit of the opportunity will require:
Leveraging granular user data to create hyper-personalized learning pathways. For instance, optimizing instruction for a German B2 speaker requiring fluency in Canadian English accounting vocabulary for a short-term internship with frequent C-suite (i.e., formal language) presentations
Building and continually updating specialized language datasets for professional and regional dialects, say Canadian English accounting
Developing intelligent systems to dynamically route translation and learning tasks to the most proficient LLMs for specific language pairs and domains, addressing performance disparities only exacerbated by geofencing regulations
III. Applications
The application of underlying technology, by my account, falls into two (over-generalized) buckets: enhanced learning versus enhanced translation. It's a choice between augmenting human capability or outsourcing it entirely.
Our data, however, reveals a surprising twist. Contrary to the path of least resistance, consumers favor better learning over automation and more frictionless communication. An overwhelming majority of our respondents expressed a greater willingness to pay a subscription for a learning platform with 2x efficacy rather than a 2x translation platform.
Here are a few early-stage businesses at the frontier of AI language learning:
Our findings underscore the importance of hyper-personalized learning experiences. Promova addresses this need by offering a customizable AI-enhanced platform that goes beyond conventional adaptivity. By tailoring its approach to diverse learning styles and specific challenges like dyslexia, Promova creates an inclusive learning environment that resonates with our survey respondents' desire for personalized instruction
Speak harnesses AI to create an immersive tutoring experience, addressing crucial challenges like accent and pronunciation while fostering the native speaker interactions that our data shows learners value.
Our research emphasized the demand for speech. Univerbal rises to this challenge with its AI-powered tutor, facilitating unscripted dialogues across 21 languages and offering a flexible, accessible alternative to traditional learning methods.
Praktika's innovative use of AI avatars demonstrates the potential for technology to create more natural learning environments. By incorporating nuanced tones and emotional inflections, Praktika has developed a highly engaging approach to language acquisition, as evidenced by its rapidly growing user base of 1.2 million monthly active users.
For younger learners, personalized and interactive experiences are particularly crucial. Buddy AI caters to this demographic with its AI-powered English tutor, blending expert-crafted lesson plans with playful, voice-based interactions.
Reach out to hworland@nea.com continue the conversation.
Notes
This research represents two consumer surveys. The first survey evaluating translation efficacywas conducted among a representative sample of 150 adults living in the Germany. The second survey was conducted among a representative sample of 200 adults also living in Germany. The survey was fielded using the Pollfish platform during June 2024. Pollfish evaluates language fluency. Pollfish partners directly with app developers; the developer defines an appropriate and specific non-cash incentive in exchange for completed surveys that benefit real consumers but doesn’t motivate them to become career panelists. Please note that as with all survey research, there is a potential for sampling error and other forms of bias. Results should be interpreted as an indication of sentiment among the target population rather than an exact measure.
The information provided in this blog post is for educational and informational purposes only and is not intended to be investment advice, or recommendation, or as an offer to sell or a solicitation of an offer to buy an interest in any fund or investment vehicle managed by NEA or any other NEA entity. New Enterprise Associates (NEA) is a registered investment adviser with the Securities and Exchange Commission (SEC). However, nothing in this post should be interpreted to suggest that the SEC has endorsed or approved the contents of this post. NEA has no obligation to update, modify, or amend the contents of this post nor to notify readers in the event that any information, opinion, forecast or estimate changes or subsequently becomes inaccurate or outdated. In addition, certain information contained herein has been obtained from third-party sources and has not been independently verified by NEA. Any statements made by founders, investors, portfolio companies, or others in the post or on other third-party websites referencing this post are their own, and are not intended to be an endorsement of the investment advisory services offered by NEA.
NEA makes no assurance that investment results obtained historically can be obtained in the future, or that any investments managed by NEA will be profitable. To the extent the content in this post discusses hypotheticals, projections, or forecasts to illustrate a view, such views may not have been verified or adopted by NEA, nor has NEA tested the validity of the assumptions that underlie such opinions. Readers of the information contained herein should consult their own legal, tax, and financial advisers because the contents are not intended by NEA to be used as part of the investment decision making process related to any investment managed by NEA.