The 4% Untranslate Cluster in a Live Patent Deposition
You are the realtime reporter for a two-attorney patent deposition in a hotel conference room, feeding both counsel's laptops. At the 90-minute mark, a rough count of your last ten minutes shows about 25 untranslated words per 600 written - roughly 4% - clustered in a 12-minute stretch of assay names, gene symbols, and 'phosphodiester.' One attorney has already asked you, off the record, to 'clean it up.' You cannot stop writing; the witness is still testifying. Your personal dictionary holds about 4,200 entries, none biochemical. The next scheduled break is in 35 minutes, and the job is expected to run another four hours.
1. What do you actually change in your writing during the next 35 minutes, and what do you deliberately leave alone?
Read the worked answer
Leave the writing alone in the sense that matters: you keep writing every word and you do not stop mid-sentence to build multi-stroke entries. Stopping to define a four-stroke gene symbol costs you the next two sentences, which is a worse defect than an untranslate.
What you do change is consistency. When the witness repeats a term, hit the identical strokes each time so a single later definition clears every instance in the file at once. Inventing a fresh fingering on each occurrence multiplies the cleanup work.
Two narrow exceptions justify an on-the-fly definition: the term is short (one or two strokes) and it recurs at a natural pause inside a break in the Q&A. Otherwise, mark it - a scratch list or a marker stroke in the file - rather than trying to hold a dozen terms in memory.
Also do not adopt untested briefs mid-job. A cold brief that conflicts with a common entry turns a 4% untranslate cluster into silent wrong words, which is a materially worse problem.
2. The attorney has asked you to 'clean it up.' What do you say, what do you commit to, and what do you refuse to promise?
Read the worked answer
First, draw the distinction the request is blurring: untranslate versus mistranslate. An untranslate is visible as an untranslate; a wrong word reads as testimony and can be quoted. The cluster is almost entirely untranslates, which is the less dangerous of the two failures.
Then be specific about the fix and the timing: at the break I will define the recurring stems, which should clear the bulk of the cluster within the hour; the one-off species names I will resolve during scoping against the audio and the exhibits.
Do not promise a fully clean live feed by a stated time. Do not agree to slow down or to start skipping material to keep the feed tidy - that trades a visible defect for an invisible one in the certified record.
Ask for the fastest real fix available: their expert's report and exhibit list. A word list lifted from documents the witness will read beats anything you can build from hearing the terms.
3. At the break you have 35 minutes and roughly 30 candidate terms. How do you decide what to define and in what order?
Read the worked answer
Sort by three filters, applied in this order:
- Frequency in the file so far. A term appearing nine times is worth the effort; a species name appearing once is a scoping item, not a dictionary item.
- Stroke cost. One- and two-stroke entries get built now. A six-stroke binomial gets a short abbreviation only if it is frequent enough to repay the memory load.
- Conflict risk. Every new entry gets checked against the existing 4,200 before it goes in. Under time pressure this is the step people skip, and it is the one that can quietly corrupt a common word.
After adding, write each new entry into a scratch file inside a sentence and read the output - isolated strokes translate fine, but sentences expose spacing and boundary problems.
Finally, save a dated backup of the dictionary as it stood before this session, so a bad batch can be rolled back instead of debugged live.