SentencePiece

SentencePiece is an unsupervised text tokenizer and detokenizer mainly for Neural Network-based text generation systems where the vocabulary size is predetermined prior to the neural model training. SentencePiece implements subword units (e.g., byte-pair-encoding (BPE) [Sennrich et al.]) and unigram language model [Kudo.]) with the extension of direct training from raw sentences. SentencePiece allows us to make a purely end-to-end system that does not depend on language-specific pre/postprocessing.

This is not an official Google product.

Technical highlights

• Purely data driven: SentencePiece trains tokenization and detokenization models from sentences. Pre-tokenization (Moses tokenizer/MeCab/KyTea) is not always required.
• Language independent: SentencePiece treats the sentences just as sequences of Unicode characters. There is no language-dependent logic.
• Multiple subword algorithms: BPE [Sennrich et al.] and unigram language model [Kudo.] are supported.
• Subword regularization: SentencePiece implements subword sampling for subword regularization and BPE-dropout which help to improve the robustness and accuracy of NMT models.
• Fast and lightweight: Segmentation speed is around 50k sentences/sec, and memory footprint is around 6MB.
• Self-contained: The same tokenization/detokenization is obtained as long as the same model file is used.
• Direct vocabulary id generation: SentencePiece manages vocabulary to id mapping and can directly generate vocabulary id sequences from raw sentences.
• NFKC-based normalization: SentencePiece performs NFKC-based text normalization.

For those unfamiliar with SentencePiece as a software/algorithm, one can read a gentle introduction here.

Comparisons with other implementations

| Feature | SentencePiece | subword-nmt | WordPiece | | :-------------------------------------- | :--------------------------------------------: | :-----------------------------------------------------: | :-----------------------------------------------: | | Supported algorithm | BPE, unigram, char, word | BPE | BPE* | | OSS? | Yes | Yes | Google internal | | Subword regularization | Yes | No | No | | Python Library (pip) | Yes | No | N/A | | C++ Library | Yes | No | N/A | | Pre-segmentation required? | No | Yes | Yes | | Customizable normalization (e.g., NFKC) | Yes | No | N/A | | Direct id generation | Yes | No | N/A |

Note that BPE algorithm used in WordPiece is slightly different from the original BPE.

Overview

What is SentencePiece?

SentencePiece is a re-implementation of sub-word units, an effective way to alleviate the open vocabulary problems in neural machine translation. SentencePiece supports two segmentation, byte-pair-encoding (BPE) [Sennrich et al.] and unigram language model [Kudo.]. Here are the high level differences from other implementations.

The number of unique tokens is predetermined

Neural Machine Translation models typically operate with a fixed vocabulary. Unlike most unsupervised word segmentation algorithms, which assume an infinite vocabulary, SentencePiece trains the segmentation model such that the final vocabulary size is fixed, e.g., 8k, 16k, or 32k.

Note that SentencePiece specifies the final vocabulary size for training, which is different from subword-nmt that uses the number of merge operations. The number of merge operations is a BPE-specific parameter and not applicable to other segmentation algorithms, including unigram, word and character.

Trains from raw sentences

Previous sub-word implementations assume that the input sentences are pre-tokenized. This constraint was required for efficient training, but makes the preprocessing complicated as we have to run language dependent tokenizers in advance. The implementation of SentencePiece is fast enough to train the model from raw sentences. This is useful for training the tokenizer and detokenizer for Chinese and Japanese where no explicit spaces exist between words.

Whitespace is treated as a basic symbol

The first step of Natural Language processing is text tokenization. For example, a standard English tokenizer would segment the text "Hello world." into the following three tokens.

[Hello] [World] [.]

One observation is that the original input and tokenized sequence are NOT reversibly convertible. For instance, the information that there is no space between “World” and “.” is dropped from the tokenized sequence, since e.g., Tokenize(“World.”) == Tokenize(“World .”)

SentencePiece treats the input text just as a sequence of Unicode characters. Whitespace is also handled as a normal symbol. To handle the whitespace as a basic token explicitly, SentencePiece first escapes the whitespace with a meta symbol "▁" (U+2581) as follows.

Hello▁World.

Then, this text is segmented into small pieces, for example:

[Hello] [▁Wor] [ld] [.]

Since the whitespace is preserved in the segmented text, we can detokenize the text without any ambiguities.

  detokenized = ''.join(pieces).replace('▁', ' ')

This feature makes it possible to perform detokenization without relying on language-specific resources.

Note that we cannot apply the same lossless conversions when splitting the sentence with standard word segmenters, since they treat the whitespace as a special symbol. Tokenized sequences do not preserve the necessary information to restore the original sentence.

• (en) Hello world. → [Hello] [World] [.] (A space between Hello and World)
• (ja) こんにちは世界。 → [こんにちは] [世界] [。] (No space between こんにちは and 世界)

Subword regularization and BPE-dropout

Subword regularization [Kudo.] and BPE-dropout Provilkov et al are simple regularization methods that virtually augment training data with on-the-fly subword sampling, which helps to improve the accuracy as well as robustness of NMT models.

To enable subword regularization, you would like to integrate SentencePiece library (C++/Python) into the NMT system to sample one segmentation for each parameter update, which is different from the standard off-line data preparations. Here's the example of Python library. You can find that 'New York' is segmented differently on each SampleEncode (C++) or encode with enable_sampling=True (Python) calls. The details of sampling parameters are found in sentencepiece_processor.h.

>>> import sentencepiece as spm
>>> s = spm.SentencePieceProcessor(model_file='spm.model')
>>> for n in range(5):
...     s.encode('New York', out_type=str, enable_sampling=True, alpha=0.1, nbest_size=-1)
...
['▁', 'N', 'e', 'w', '▁York']
['▁', 'New', '▁Yo