ldrift

Quantify the strength of selection and drift in linguistic timeseries.

Dependencies

Software

The software requires a UNIX environment such as Linux or Mac OS X with the following software, and required libraries:

• java
• R with libraries ggplot2, dplyr, grid, gridExtra
• tsinfer available at https://github.com/mnewberry/tsinfer
• sr available at: https://github.com/Schraiber/selection

The rhyming analysis additionally requires:

• The OCaml compiler (verified to work with ocaml 4.02.3)
• OCaml libraries pcre (verified to work with 7.1.6) and batteries (verified to work with 2.3.1)

Corpora

• The Corpus of Historical American English (COHA), available from https://corpus.byu.edu/coha/
• The Penn Parsed Corpora of Historical English, available from http://www.ling.upenn.edu/hist-corpora/

Execution

Generating random drift

To generate random timeseries under drift, edit simulate-bridge.sh to inform it of the location of sr, then run it (see comments in that file for usage).

The specific data used for Figure 1 can be generated using the seed:

./sr -R -e 1329052268 -b 0.1,0.8,0,2

Past-tense verbs

Symlink the directory containing the COHA distribution to data/, or edit make-past-tense.sh (and pastrhymes.ml for rhyming analysis) to inform it of the location of the directory containing COHA's part-of-speech-tagged files, i.e. COHA/Word_lemma_PoS/.

Run make-past-tense.sh to generate local/data/vvd-COHA. (~10m)

Edit the first line of past-tense.R to specify the location of tsinfer, then run it (as Rscript past-tense.R) to generate the local/out/past-tense-binned.csv and local/out/past-tense-results.csv used to produce Figure 2. (~1m)

To conduct the rhyming analysis, run make-past-tense.sh to retreive the CMU rhyming dictionary. Run ocamlbuild -pkgs pcre,batteries pastrhymes.native to compile pastrhymes.native. Run pastrhymes.native to generate the output, local/out/rhyme.schemes.tsv, local/out/rhyme.timeseries.vvd.tsv, and local/out/rhyme.timeseries.tsv. (~60m)

do-support

Follow the instructions in make-do-support.README to generate local/data/do-PPCHE, which involve cloning another repository and linking it to corpus data.

As with past-tense verbs, edit the first line of do-support.R to specify the location of tsinfer, then run it (as Rscript do-support.R) to generate the local/out/do-support-binned.csv and local/out/do-support-results.csv used to produce Figure 3. (~1m)

Negation

Follow the instructions in make-negation.README to generate local/data/neg-data.csv, which involve cloning another repository and linking it to corpus data.

As with past-tense verbs, edit the first line of negation.R to specify the location of tsinfer, then run it (as Rscript negation.R) to generate the local/out/neg-binned.csv and local/out/neg-results.csv used to produce Figure 4. (<1m)