BRENNEN

Industries / Finance and business services / Legal and professional services / Court reporting

Stenography

Writing speech in shorthand, now mostly on a stenotype, with software that turns each chord of keys into English.

7

Milestones achieved, from 1879 to 2014.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1879

    A shorthand machine patented

    Miles M. Bartholomew of Belleville, Illinois, was granted a US patent for a stenographic machine on 20 May 1879. Stenograph, the main US maker of steno machines today, says Bartholomew invented "the first successful shorthand machine" and that it was used by official reporters as late as 1937. The keyboard site kbd.news points out that the patent calls the machine an "Improvement in Stenographic Machines" and doubts it was the first.

    stenograph.comkbd.news

    Credit

    Miles M. Bartholomew

  2. 1879

    Machine shorthand in Italian courts

    Antonio Michela Zucco's keyboard writes syllables by pressing several keys together. It gave the Turin city council an immediate verbatim record from January 1879 and was shown to the court in Naples that February and March. The Italian Senate adopted it from 1881 and still uses it, now linked to computers, though Il Post counted 23 Senate stenographers in 2023, under half the planned staff.

    senatoragazzi.itilpost.it

    Credit

    Antonio Michela Zucco

  3. 1911

    The stenotype keyboard

    Ward Stone Ireland's Stenotype, made by the Universal Stenotype Company of Owensboro, Kentucky, had keys that press all the way down. The Henry Ford museum holds one stamped "Pat. Jan. 3rd 1911", and Stenograph calls it "the first practical machine with a totally depressible keyboard." The company was forced into bankruptcy at the end of World War I by a defaulted government munitions contract.

    thehenryford.orgstenograph.com

    Credit

    Ward Stone Ireland; Universal Stenotype Company

  4. 1971

    Computers read steno notes

    Computer-aided transcription (CAT) looks up each steno stroke in a dictionary to produce English. IBM described the method in 1964, and in 1971 the US National Bureau of Standards tested an existing system in the Philadelphia Court of Common Pleas for the Federal Judicial Center. It found that a first computer transcript took about a tenth of the usual time, and called word boundaries, where one word ends and the next begins, "the most troublesome" problem. In Britain the National Physical Laboratory began CAT research on the Palantype shorthand machine in 1967.

    aclanthology.orgnvlpubs.nist.govfjc.govcollection.sciencemuseumgroup.org.uk

    Credit

    IBM; Ernest H. Short and Walter G. Leight (National Bureau of Standards); National Physical Laboratory

  5. 1978

    Computer transcripts become routine

    A 1978 National Center for State Courts guide reported that Philadelphia kept CAT running after the demonstration's funding ended, and that four companies sold it: Barons Data Systems, Stenocomp, Stentran and Stenograph Machines. A 1981 survey found that about 60% of federal court reporters using CAT "had found no appreciable improvement in their transcript efficiency."

    LessonBuying a CAT system did not by itself make transcripts faster.

    ojp.govojp.gov

    Credit

    J. Michael Greenwood and Jerry R. Tollar (National Center for State Courts)

  6. 2010

    Free steno software

    Plover, free software from the Open Steno Project, turns steno strokes into text on an ordinary computer and keyboard. In 2013 its founder Mirabai Knight said it had been around for about three years and cut the hardware cost "from $4,000 down to $50." QMK, the open-source firmware used in many hobby keyboards, supports Plover's protocols. Knight said she was not aiming at the court reporting market.

    openstenoproject.stenoknight.comdocs.qmk.fm

    Credit

    Mirabai Knight, Joshua Lifton, Hesky Fisher

  7. 2014

    A shortage forecast

    A study by Ducker Worldwide for NCRA, the US court reporters' association, counted about 32,000 stenographic court reporters and projected a shortfall of nearly 5,500 positions by 2018. In 2026 a study for the Texas Office of Court Administration found that steno programs nationwide fell by 27% between 2022 and 2026, leaving 19 NCRA-approved programs, and that active certified reporters in Texas fell by 14% between 2005 and 2025.

    vernoncourtreporters.comtxcourts.gov

    Credit

    Ducker Worldwide; Texas Office of Court Administration with Texas A&M University's Public Policy Research Institute

Watch for

  1. Free training to refill schools

    Lexitas, a court reporting firm that funds Project Steno, a nonprofit that recruits steno students, said Project Steno had enrolled more than 1,500 students in its free six-week Basic Training program, that 40% of participants went on to accredited steno schools, and that its partner schools' graduation rates were five times the national average. The release does not say who measured these figures or over what period.

    Achieved whenan outside count, such as NCRA's school reports or a state study, shows more steno graduates coming from these students, against a Texas study that found program numbers still falling in 2026.

    lexitaslegal.comtxcourts.gov

    Claimed by

    Lexitas, about Project Steno, June 2025

Open

  1. 01

    Open dictation at test speeds

    An openly licensed set of five-minute dictation recordings at the speeds of NCRA's Registered Professional Reporter (RPR) test: literary text at 180 words per minute, a jury charge (a judge's instructions to a jury) at 200, and two-voice question and answer at 225. Each comes with its exact text, the time of every word and a scoring script, so that students, teachers and software makers can be scored the same way.

    The brief
    Credit · unclaimed
  2. 02

    Guessing untranslated strokes

    A Plover plugin that suggests the most likely word when a stroke is missing from the writer's dictionary, using the sentence around it, and an open test that scores any tool that does this.

    The brief
    Credit · unclaimed

Open 01 · the brief

Open dictation at test speeds

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An openly licensed set of five-minute dictation recordings at the speeds of NCRA's Registered Professional Reporter (RPR) test: literary text at 180 words per minute, a jury charge (a judge's instructions to a jury) at 200, and two-voice question and answer at 225. Each comes with its exact text, the time of every word and a scoring script, so that students, teachers and software makers can be scored the same way.

Why it is still open

NCRA sells its practice dictation through its Learning Center. Typey Type, a free steno practice site, has only one openly licensed sound, a metronome, and the free practice channel that the keyboard maker Platinum Steno recommends is on YouTube with no licence stated. I found no open dataset of this kind on GitHub or Hugging Face.

thejcr.comraw.githubusercontent.comdocs.stenokeyboards.comncra.org

Built with what exists

public-domain court text, such as US federal court opinions; volunteer readers; Montreal Forced Aligner, an open-source tool that finds the time of each word in a recording; Plover to capture the strokes of volunteer steno writers.

  1. First step

    Write the three scripts from public court text, record one take of each, and publish them with word times.

  2. Done looks like

    Three takes of each of the three legs released under CC BY or CC0. Each take is within 2% of its target speed, measured from the word times, and a second person gets the same speeds and pass or fail results with the published script.

Open 02 · the brief

Guessing untranslated strokes

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A Plover plugin that suggests the most likely word when a stroke is missing from the writer's dictionary, using the sentence around it, and an open test that scores any tool that does this.

Why it is still open

Plover's documentation says a stroke with no dictionary entry "results in an untranslate", left on the screen as raw steno. Stenograph says its CATalyst software "identifies your mistranslated steno and offers accurate translation suggestions" but publishes no test of it. I found no Plover plugin that uses a language model and no research paper on the task.

plover.readthedocs.iostenograph.com

Built with what exists

Plover's reverse lookup, which turns English into the strokes that would write it, to make test sentences; public court text; a small open-weight language model.

plover.readthedocs.io

  1. First step

    Check the licence of Plover's default dictionary, build the test sentences from it, and measure Plover's starting score.

  2. Done looks like

    On at least 1,000 sentences with 5% of dictionary entries held out, the share of untranslated strokes the plugin gets right as its first guess and within its first five. Code and data are published, and a second person's rerun lands within one percentage point.