pszemraj's picture
fix demo link
c837f47
|
raw
history blame
6.78 kB
metadata
license:
  - cc-by-nc-sa-4.0
  - apache-2.0
tags:
  - grammar
  - spelling
  - punctuation
  - error-correction
  - grammar synthesis
datasets:
  - jfleg
widget:
  - text: i can has cheezburger
    example_title: cheezburger
  - text: There car broke down so their hitching a ride to they're class.
    example_title: compound-1
  - text: >-
      so em if we have an now so with fito ringina know how to estimate the tren
      given the ereafte mylite trend we can also em an estimate is nod s i again
      tort watfettering an we have estimated the trend an called wot to be
      called sthat of exty right now we can and look at wy this should not hare
      a trend i becan we just remove the trend an and we can we now estimate
      tesees ona effect of them exty
    example_title: Transcribed Audio Example 2
  - text: >-
      My coworker said he used a financial planner to help choose his stocks so
      he wouldn't loose money.
    example_title: incorrect word choice (context)
  - text: >-
      good so hve on an tadley i'm not able to make it to the exla session on
      monday this week e which is why i am e recording pre recording an this
      excelleision and so to day i want e to talk about two things and first of
      all em i wont em wene give a summary er about ta ohow to remove trents in
      these nalitives from time series
    example_title: lowercased audio transcription output
  - text: Frustrated, the chairs took me forever to set up.
    example_title: dangling modifier
  - text: I would like a peice of pie.
    example_title: miss-spelling
  - text: >-
      Which part of Zurich was you going to go hiking in when we were there for
      the first time together? ! ?
    example_title: chatbot on Zurich
  - text: >-
      Most of the course is about semantic or  content of language but there are
      also interesting topics to be learned from the servicefeatures except
      statistics in characters in documents. At this point, Elvthos introduces
      himself as his native English speaker and goes on to say that if you
      continue to work on social scnce,
    example_title: social science ASR summary output
  - text: >-
      they are somewhat nearby right yes please i'm not sure how the innish is
      tepen thut mayyouselect one that istatte lo variants in their property e
      ere interested and anyone basical e may be applyind reaching the browing
      approach were
  - medical course audio transcription
parameters:
  max_length: 128
  min_length: 4
  num_beams: 4
  repetition_penalty: 1.21
  length_penalty: 1
  early_stopping: true

grammar-synthesis-large - beta

A fine-tuned version of google/t5-v1_1-large for grammar correction on an expanded version of the JFLEG dataset.

usage in Python (after pip install transformers):

from transformers import pipeline

corrector = pipeline(
              'text2text-generation',
              'pszemraj/grammar-synthesis-large',
              )
raw_text = 'i can has cheezburger'
results = corrector(raw_text)
print(results)

give it a spin in Colab at this notebook.

Model description

The intent is to create a text2text language model that successfully completes "single-shot grammar correction" on a potentially grammatically incorrect text that could have a lot of mistakes with the important qualifier of it does not semantically change text/information that IS grammatically correct.

Compare some of the heavier-error examples on other grammar correction models to see the difference :)

Limitations

  • dataset: cc-by-nc-sa-4.0
  • model: apache-2.0
  • this is still a work-in-progress and while probably useful for "single-shot grammar correction" in a lot of cases, give the outputs a glance for correctness ok?

Use Cases

Obviously, this section is quite general as there are many things one can use "general single-shot grammar correction" for. Some ideas or use cases:

  1. Correcting highly error-prone LM outputs. Some examples would be audio transcription (ASR) (this is literally some of the examples) or something like handwriting OCR.
    • To be investigated further, depending on what model/system is used it might be worth it to apply this after OCR on typed characters.
  2. Correcting/infilling text generated by text generation models to be cohesive/remove obvious errors that break the conversation immersion. I use this on the outputs of this OPT 2.7B chatbot-esque model of myself.

    An example of this model running on CPU with beam search:

original response:
                ive heard it attributed to a bunch of different philosophical schools, including stoicism, pragmatism, existentialism and even some forms of post-structuralism. i think one of the most interesting (and most difficult) philosophical problems is trying to let dogs (or other animals) out of cages. the reason why this is a difficult problem is because it seems to go against our grain (so to
synthesizing took 306.12 seconds
Final response in 1294.857 s:
        I've heard it attributed to a bunch of different philosophical schools, including solipsism, pragmatism, existentialism and even some forms of post-structuralism. i think one of the most interesting (and most difficult) philosophical problems is trying to let dogs (or other animals) out of cages. the reason why this is a difficult problem is because it seems to go against our grain (so to speak)

Note: that I have some other logic that removes any periods at the end of the final sentence in this chatbot setting to avoid coming off as passive aggressive

  1. Somewhat related to #2 above, fixing/correcting so-called tortured-phrases that are dead giveaways text was generated by a language model. Note that SOME of these are not fixed, especially as they venture into domain-specific terminology (i.e. irregular timberland instead of Random Forest).

Training and evaluation data

More information needed 😉

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 8e-05
  • train_batch_size: 4
  • eval_batch_size: 1
  • seed: 42
  • distributed_type: multi-GPU
  • gradient_accumulation_steps: 32
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.02
  • num_epochs: 1

Framework versions

  • Transformers 4.20.1
  • Pytorch 1.11.0+cu113
  • Datasets 2.3.2
  • Tokenizers 0.12.1