Member Avatar for Goldfinch

Hi, I'm looking into implementing a program that can translate words into another language (english to name the language) and I am looking for an easily readable language. I posted in this forum because its not really a coding issue, and its closer to the Natural Language Processing that is a computer science field.

  • 2 Contributors
  • 2 Replies
  • 209 Views
  • 1 Week Discussion Span
  • Latest Post Latest Post by Goldfinch

Dani AI

Generated

A practical first decision is whether the project will map isolated words (a bilingual lexicon) or handle full sentences. Word-only systems can work well with a modest dictionary plus simple normalization; sentence-level translation requires tokenization, part-of-speech and morphological analysis, handling word-order differences and context-dependent senses. The original post from names English as the source; that direction often simplifies morphology handling on the source side but still requires care on the target side.

Pick a target language that reduces engineering overhead. Favor languages with consistent spelling and few inflectional forms so rule-based or small-data approaches pay off quickly. Examples to consider are Malay/Indonesian or simplified creoles and auxiliary languages (which trade expressive subtlety for regularity). Languages with heavy agglutination, many irregular verbs, or mandatory tone/diacritics increase preprocessing work and data requirements. Availability of parallel text (even small phrasebooks) is a practical constraint: an easy grammar plus no data is still hard.

Suggested minimal roadmap and common pitfalls:

  • Start with a seed bilingual dictionary and a normalizer (lowercase, strip punctuation, canonicalize diacritics).
  • Add a tokenizer and simple lemmatizer or affix stripper to collapse forms to dictionary lemmas.
  • Implement rule-based reordering for the most common sentence patterns before trying ML.
  • When parallel sentences are available, extract phrase pairs and expand coverage; evaluate with small held-out sets and manual error analysis focusing on named entities and idioms.
  • Watch for polysemy and multiword expressions: they cause most early errors.

A tiny pseudocode sketch for a word-first pipeline:

# pseudo-code
tokenize(sentence) -> tokens
for t in tokens:
  t_norm = normalize(t)
  lemma = lemmatize(t_norm)
  out = lexicon.get(lemma, handle_oov(t))
assemble output with simple reordering rules

As noted, language regularity matters. For ’s early prototype, iterate: dictionary → rules → small ML models, and only scale to full neural systems after collecting parallel data.

Recommended Answers

All 2 Replies

Member Avatar for Hiroshe

Are you just translating 'words', ie, making a dictionary? Or are you translating whole sentences?

I'm assuming you want to translate sentences.

(1) Inflected languages would be nice here. Inflected means that the ending of the word indicates what part of speach it is (aswell as tense). This means that you don't need to worry about the sentence structure as much.

(2) Consitency is also important; the more consitant it is, the less you need to spend time on exceptions.

(3) Easy to learn, and simply syntax.

I think that Esperanto is a good start for you! It's a constructed language that was designed to be simple. It is also fairly well known. You can learn it with this.

In your title I think you mentioned you were also looking for a library to help you (try to include all of the information in the body of the post as well). nltk might be a good start. You may also find that scikit-learn will help depending on how you want to approach this.

Edited by Hiroshe because: spelling
Member Avatar for Goldfinch

Thank You, BTW, I like the Umbreon.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.