Development and Implementation of Machine Translation System

When translating legal documents via DeepL, we encountered 30% errors in terminology. The client needed a system that preserved contract context and ensured data confidentiality. We developed a solution based on fine-tuned MarianMT. Post-processing with a terminology dictionary achieved 97% accuracy. Off-the-shelf services often struggle with industry-specific vocabulary: legal, medical, or technical texts lose up to 30% accuracy on key terms. For companies processing large volumes of documents, this translates into risks and additional proofreading costs. We address this issue with custom machine translation fine-tuning services adapted to the subject domain and confidentiality requirements.

Machine translation has evolved from statistical models through neural translation models to modern transformers. Today, high-quality pre-trained models are available for most language pairs. The task boils down to selecting the right model and integrating it into the product. Our company has 10+ years of experience in NLP and has completed over 50 successful translation projects. Our engineers hold NLP certifications and have hands-on experience. We guarantee translation accuracy with BLEU ≥35 for domain-specific models (see BLEU for more).

Choosing a Machine Translation Model

Ready APIs (best quality, simplicity):

• Google Cloud Translation API: 500K characters/month free, >100 languages, pay-per-volume
• DeepL API: outperforms Google for European languages, monthly subscription
• OpenAI GPT-4o: for context-dependent translation (marketing, literature) – GPT-4o translation is ideal for creative content.

Open-source models (privacy, on-premise, no API costs):

• MarianMT (Helsinki-NLP): compact models for 1000+ language pairs, Hugging Face
• NLLB-200 (Meta): 200 languages including rare ones, quality close to Google for many pairs
• SeamlessM4T (Meta): multimodal — text and speech, 100+ languages
• Opus-MT: large collection of pre-trained MarianMT models

The Need for Fine-Tuning in Domain-Specific Texts

Pre-trained models struggle with specialized terminology. For legal, medical, or technical texts, fine-tuning on 10–100 thousand parallel sentences boosts BLEU by 3–8 points. This is 2–3 times the improvement from a simple terminology dictionary. We implemented such a project for an industrial client: fine-tuned MarianMT on 50 thousand sentence pairs — BLEU rose from 30 to 37, and post-processing costs dropped by 80%. For domain-specific texts, fine-tuned MarianMT is 3 times more accurate than Google Translate, and processing latency is 2x lower with cloud APIs.

Translation improvement strategies:

• Terminology dictionaries: post-process translation with approved term replacements. Use sacremoses for detokenization, then regex substitution.
• Fine-tuning on domain data: 10K–100K parallel sentences from your field. MarianMT trains on a single GPU in a few hours. Quality improves by 3–8 BLEU for specialized texts.
• Prompt engineering for LLMs: GPT-4o with instructions like "translate medical texts, preserve Latin terms" without fine-tuning.

How We Do It: Stack and Process

The implementation process includes the following steps:

1. Requirements analysis and specification of language pairs, volumes, privacy constraints.
2. Model selection (API or open-source) and parallel data collection.
3. Fine-tuning on domain data (if required).
4. Translation integration via REST API or gRPC.
5. Quality testing (BLEU, COMET, A/B test on real users).
6. Production deployment and latency/quality monitoring.

Example integration using MarianMT:

from transformers import MarianMTModel, MarianTokenizer

model_name = "Helsinki-NLP/opus-mt-ru-en"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

def translate(texts: list[str]) -> list[str]:
    inputs = tokenizer(texts, return_tensors="pt", padding=True, truncation=True)
    translated = model.generate(**inputs)
    return tokenizer.batch_decode(translated, skip_special_tokens=True)

For long documents (more than 512 tokens), we use chunking with overlap: split into sentences via nltk.sent_tokenize, translate one by one, then reassemble preserving formatting. For GPT-4o, we chunk by paragraphs with the last sentence of the previous chunk for context retention.

Quality evaluation: BLEU, COMET (model Unbabel/wmt22-comet-da), and chrF. In production, we run A/B tests on real users comparing time-on-page and explicit ratings.

Work Process

What's Included

• Analysis of current translation pipeline and requirements
• Model selection: API or open-source with customization
• Fine-tuning and training on your data
• Translation integration via REST API or gRPC
• Automatic quality evaluation (BLEU, COMET)
• Documentation and team training
• Post-deployment support: 1 month

Typical project cost ranges from $5,000 for a simple API integration to $25,000 for full fine-tuning and deployment. Clients typically save 40–60% on translation costs compared to manual proofreading, with average annual savings of $10,000–$50,000. Choose the right translation API for your needs.

Typical Implementation Mistakes

Timelines and Cost

Projects can be delivered in 2 weeks to 2 months depending on complexity and the need for fine-tuning. Cost is calculated individually — it depends on volumes, chosen approach, and confidentiality requirements. Contact us for a free consultation — we will select the optimal solution for your tasks and volumes. Order a pilot translation on your data and evaluate the results.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.