Natural Language Processing

​

This is my first project. I was inspired to try this out because. I always wanted to understand how linguistics can be applied to the real world - and as AI Chat Bots have become more popular, I figured that this would be the perfect opportunity to understand  how to use this for the benefit of ESL writers / speakers. I would like to use this program to help ESL learners how to cultivate their own voice  (as opposed to a type of robotic / textbook AI type  narrative) by providing feedback on any writing / speaking tasks users submit to this bot. 

​

I'd like to create a fully developed sentiment-aware speaking assistant specifically designed to help L2 speakers communicate more naturally with L1 speakers and vice versa. 

​

In order to ease L2 speakers speaking to L1 speakers, or L1 to L2 to sound more natural or personable - this type of bot will be essential in creating a common barrier between people when needing to express, communicate, or delegate any tasks or every day conversation.

​

​Current:  .github/workflows

​

Factors to consider -  

 

1. Emotional Influence on L2 Speaking

• Anxiety & Hesitation: Many L2 speakers experience speech anxiety, leading to pauses, filler words, or avoidance of complex sentence structures.

• Confidence & Fluency: As proficiency grows, speakers use smoother intonation, varied sentence structures, and more expressive language.

• Frustration & Errors: Negative emotions (like frustration) can result in more self-corrections, broken syntax, or avoidance strategies.

​

2. Sentiment Analysis for L2 Speakers

Sentiment analysis can help detect and understand emotions in spoken L2 communication

• Analyzing Speech Patterns: Identifying positive, neutral, or negative sentiments based on tone, fluency, and word choice.

• Detecting Anxiety or Confidence Levels: Tracking pitch variations, pauses, and sentence complexity to assess emotional states and what words / phrases are particularly used then.

• Providing Feedback on Emotional Tone: Sentiment-aware AI can help learners adjust their speech by recognizing when they sound hesitant, meek, assertive, or enthusiastic.

​

​

Other Sentiment Models that can help with this bot:

​

https://aclanthology.org/2020.emnlp-main.567/

https://dl.acm.org/doi/fullHtml/10.1145/3629606.3629665

https://arxiv.org/html/2502.15367v1

​

Goals: :

• Collect ESL Speech Data: Use transcriptions and audio recordings of ESL learners speaking in different scenarios.

• Label Sentiments: Classify emotions like nervousness, confidence, frustration, and excitement.

• Train an NLP Model: Use a combination of speech recognition, sentiment analysis, and prosody detection to analyze emotional tone.

https://github.com/kateye1988/NaturalFlowAI/tree/main/.github/workflows

​

​

My README.md : 

​​​

​

This project provides analytical feedback for ESL (English as a Second Language) learners. 

 

The goal is to evaluate and provide actionable feedback to help writers and speakers cultivate their own style.

 

The program helps users understand their emotional tone and communication style by analyzing texts through large language models (LLMs) and sentiment classification. The sentiment analysis categorizes texts into positive, negative, neutral, informal, and formal classes, offering learners insights into their emotional tone and writing style. The feedback will include ESL-specific challenges such as common hypercorrection mistakes, tone adjustments, and cultural considerations. # ESL-Sentiment-Analysis-and-Modification

 

Within each specific class, the sentiment categories will the system detect: confidence, hesitation, frustration, enthusiasm. The program will provide insight on cultural considerations such as usage of puns, collocations, phrases, cultural references and the like.

 

Some Data sets that I will use: COCA - English Corpora - Contains native English language from various media platforms, including dated magazines and newspapers.
TOEFL11 Corpus – Contains non-native English essays. ICLE (International Corpus of Learner English) ICNALE (International Corpus Network of Asian Learners of English) – Learner essays with proficiency levels. Lang-8 Learner Corpus – Includes corrected texts from language learners

​

​

 

Blueprint / Architecture of my Project

​

Input: Text from ESL learners (e.g., essays, short responses, or spoken transcriptions)  

​

Processing: An LLM  with a  pre-trained model like BERT, RoBERTa - hugging face - building a custom LLM combined with sentiment classification and rule-based checks for ESL-specific errors.

- Pattern detection

- idiom cultural references 

​

Output: Detailed feedback- examples: 

“Your tone is neutral but shows hesitation (e.g., repeated use of ‘maybe’). Try replacing it with a confident phrase like ‘I believe’ to strengthen your style.”

​

“The phrase ‘break a leg’ is a cultural idiom meaning ‘good luck.’ It’s informal and may confuse non-native readers unfamiliar with it.”

 

Tools: Python, libraries like Hugging Face Transformers, NLTK, spaCy, or PyTorch/TensorFlow for the LLM; scikit-learn or TextBlob for sentiment analysis.

RAG for by passing authentication  / credential factors