Segment a corpus into new documents of roughly equal sized text chunks, with the possibility of overlapping the chunks.
corpus_chunk(
x,
size,
truncate = FALSE,
use_docvars = TRUE,
verbose = quanteda_options("verbose")
)tokens object whose token elements will be segmented into chunks
integer; the (approximate) token length of the chunks. See Details.
logical; if TRUE, truncate the text after size
if TRUE, repeat the docvar values for each chunk;
if FALSE, drop the docvars in the chunked tokens
if TRUE print the number of tokens and documents before and
after the function is applied. The number of tokens does not include paddings.
The token length is estimated using stringi::stri_length(txt) / stringi::stri_count_boundaries(txt) to avoid needing to tokenize and rejoin
the corpus from the tokens.
Note that when used for chunking texts prior to sending to large language
models (LLMs) with limited input token lengths, size should typically be set
to approximately 0.75-0.80 of the LLM's token limit. This is because
tokenizers (such as LLaMA's SentencePiece Byte-Pair Encoding tokenizer)
require more tokens than the linguistically defined grammatically-based
tokenizer that is the quanteda default. Note also that because of the
use of stringi::stri_count_boundaries(txt) to approximate token length
(efficiently), the exact token length for chunking will be approximate.
data_corpus_inaugural[1] |>
corpus_chunk(size = 10)
#> Corpus consisting of 152 documents and 4 docvars.
#> 1789-Washington.1 :
#> "Fellow-Citizens of the Senate and of the House of"
#>
#> 1789-Washington.2 :
#> "Representatives: Among the vicissitudes incident to life"
#>
#> 1789-Washington.3 :
#> "no event could have filled me with greater anxieties than"
#>
#> 1789-Washington.4 :
#> "that of which the notification was transmitted by your"
#>
#> 1789-Washington.5 :
#> "order, and received on the 14th day of the present month."
#>
#> 1789-Washington.6 :
#> "On the one hand, I was summoned by my Country, whose voice"
#>
#> [ reached max_ndoc ... 146 more documents ]