parlarina.com

I have never found a language-learning programme that really worked for me. With Parlarina, I wanted to bring together different methods that I could switch between according to how I felt: having a conversation, reviewing words or working through a text, while the app retained what I had learnt and where I had struggled.

Conversation at the centre

The Smart Chat is at the centre of my idea. I believe that speaking is a particularly good way to learn a language and overcome the hesitation to use it in everyday life. The other exercises should support these conversations and give them new starting points.

For that to work, the chat needs information about what has happened during learning. Which words are still missing? Which mistakes recur? What is becoming easier? It should use these observations to shape suitable conversation exercises rather than start from scratch every time.

Shared knowledge about learning

I called the central analysis the Central Scrutinizer. It brings together information from the different learning modules. Progress, vocabulary and recurring mistakes are recorded in a shared system so that they remain available when the learner changes methods.

In the existing prototype, conversations are analysed and new words are stored with their context sentences. A learner profile retains observations about strengths, difficulties and interests. The chat receives this information along with words due for review, allowing a word from an exercise to reappear later in a conversation.

Alongside this, I built structured learning paths, grammar, listening and speaking exercises, games and text work based on the Birkenbihl method. The choice was intended to offer different ways of working on the same language. Over time, though, it grew into a scope I had not fully appreciated at the beginning.

Too much at once

I wanted too much in the first attempt. Enthusiasm for the possibilities led me to keep adding features and languages until my ideas exceeded what I could manage. I am happy with the quality. It was the number of languages, exercises and modules that became too large. I misjudged that scope and want to limit it in the rebuild.

I still believe in the idea behind Parlarina and intend to rebuild the project in the near future. I will start with fewer features and languages and develop the application gradually. A smaller scope should let me check more carefully whether the individual parts work and genuinely complement one another in learning before I add more.

The application shown here is the existing prototype. The main learning app is not yet publicly accessible; only the Hanzi trainer can currently be used without signing in. The initial features and languages for the rebuild have not yet been decided.

The existing technical implementation

The prototype runs on PHP, MariaDB and JavaScript without a frontend framework. The learning modules are separate and access shared learning information. Before each chat response, instructions are assembled from the conversation mode, correction style, learner profile and vocabulary due for review.

A common interface connects the application to different AI providers. If a response takes too long, a second provider can be called in parallel. The application supplies the conversation history and learning information, allowing providers to be switched.

Generated exercises are checked for structure, how their answers are evaluated in the interface, and content. These checks are intended to identify problems such as gap-fill exercises with several plausible answers.