// Twitter Cards // Prexisting Head The Biologist Is In: April 2018

Thursday, April 5, 2018

The Naming of Things

If you've been following me here for a bit, you've probably noticed I'm interested in plant breeding (especially garden veggies). My main goals are to have healthy plants that grow and produce well for me with minimal inputs in my short-season climate. The measure of, "tasty" I go by is what tastes good to me and my family, with what other people consider tasty (when I occasionally do taste-tests) held to a lesser significance.

Two large cherry sized, blocky, white tomatoes. They're sitting on a notebook with a sketched map of the garden, showing where all the plants are and which plants were grown from the same batches of seed. There is a blue pen pointing at the specific plant which produced the fruit.
From 2017, with garden notes.
I've been working with tomatoes for several years and have developed some more fine-tuned ideas about what I want the plants to become. One of my lines, seen at right, is approaching stability. That is to say, most plants from one year to the next produce very similar fruit. The fruit are blocky, large-cherry sized, white (well, paler than yellow) in color, and have a very thick outer fruit-wall (not the skin). They've tested well with people in and outside my immediate family, so I've been thinking about the possibility of distributing their seed in the future.

A few dozen of the large white cherry tomatoes sitting on a white plastic cutting board.
From 2016.
In my personal notes, I've been using the rather uncreative name of, "Abbey White" for these tomatoes. It is sufficiently descriptive to let me know what I'm talking about in my notes, but it isn't a name I expect to attach to the variety when/if I start distributing it. I could easily adjust it to, "Abbey's White", but I'm not sure I want to go with that either.

In the forground is a ceramic bowl filled with diced white tomatoes. In the background is a large wooden cutting board covered in white, yellow, and orange tomatoes (as well as a few green tomatilloes).
From 2016.
"Wait. Tomatoes are red, right!?" A white tomato might seem kinda unusual, but it's just one of a very wide spectrum of colors that tomatoes can be found in. (Check out these companies I have no affiliation with: Artisan Seeds, Baker Creek Heriloom Seeds, TomatoEden, and SeedSavers Exchange. There's so much more diversity in color and taste available if you're willing to grow tomatoes from seed.) My tomatoes tend to be any color but red. Red fruit that have turned up in my garden have tended to have a taste I didn't favor, so over a few years I stopped growing as many red tomatoes. I expect I'll need to bring in some new genetics before I can grow red tomatoes that will taste good to me.



While I was thinking about how to go about naming this variety (and others in the future), I came across twitter user @JanelleCShane. She's been playing with Recurrent Neural Networks (basically a type of AI (specifically a type of machine learning)) trained on diverse datasets, like fruit names (or knitting patterns (or Irish melodies)), so I tweeted:

(I only later noticed my garbled grammar.) I was somewhat surprised when she responded back, asking if I had a list of tomato variety names she could train her AI with. I didn't, but I was pretty sure I could pull one together pretty quickly from online resources. After some looking, I found several sources ([1], [2], [3], [4], [5], [6]) with large lists of tomato variety names. To avoid spending too much time gathering the names, I wrote web scrapers to process each source and output text files with lists of names. In total, across the six sources, I collected 11,719 distinct tomato variety name strings. Some may represent extinct varieties. Some are in other languages. Some are numerical codes. There's also capitalization and spelling variations. I threw them all into a file that Janelle could use to train her AI.

Have a look at her blog post on the tomato name trained AI at: http://aiweirdness.com/post/172622965862/tomatonames



So. What did the trained AI come up with? Well, at first the AI got overly fascinated with the numerical code names in the training dataset. It produced lots of new "names" that would be quite not useful for naming a new variety. Janelle stripped out most of the code names from the list and trained the AI again.

This time there were some really good results, some really wrong results, and all sorts of weirdness in between. I've highlighted some of my favorites from each category.

The Good,the Weird,and the Wrong.
  • Floranta
  • Sweet Lightning
  • Speckled Boy
  • Flavelle
  • Market Days
  • Fancy Bell
  • Pinkery Plum
  • Mountain Gem
  • Garden Sunrise
  • Honey Basket
  • Cold Brandy
  • Sun Heart
  • Flaminga
  • Sunberry
  • Special Baby
  • Golden Pow
  • Birdabee
  • Sandwoot
  • Bear Plum
  • The Bango
  • Grannywine
  • Sun Burger
  • Bungersine
  • First No.4
  • Smoll Pineapple
  • The Ball
  • Golden Cherry Striped Rock
  • Eggs
  • Old German Baby
  • Frankster Black
  • Bumbertime
  • Adoly Pepp Of The Wonder
  • Cherry, End Students
  • Small Of The Elf
  • Champ German Ponder
  • Pearly Pemper
  • Green Zebra Pleaser
  • Flute First
  • Speckled Garfech
  • Green Dork
  • Cluster Gall
  • Shirve’s Gigant Bullburk
  • Giant Ballsteak
  • Black Crape
  • Brandywine, True Grub
  • Caraball
  • Ranny Blue Ribber
  • Roma Wasting Star
  • Scar Giant
  • Bug Beauty
  • Banana Placente
  • Bananana
  • Stoner
  • Speckled Bake
  • Ruck
  • Green Boor
  • Wonder Bagg
  • Sun Bung
  • Bellende
  • Shart Delight
  • Solad Piss


There were also a collection that would fit perfectly among the real tomato names, though they'd be kinda strange in other contexts.
  • Matt's Sandwich
  • Indigo Tree
  • Striped Hollow Potato Leaf
  • Lelly's Yellow Stuffers
  • Terra Pink Strain
  • Greek Boar
  • Ton's Oxheart
  • Babla's German Paste
  • Mortgage Lifter, Honey Blues

I really like when the AI tried to name a tomato after a person. It didn't have enough examples for real human names, but it gave it a good solid try.
  • Matt's Sandwich
  • Lelly's Yellow Stuffers
  • Ton's Oxheart
  • Babla's German Paste
  • Shirve’s Gigant Bullburk

Amusingly, the AI came up with an existing name that wasn't in the training dataset. "Sunberry" is the name of another fruit. It's a close relative of the tomato, so I think I'll call that a positive score for the AI.



Do any of these names fit my tomato? I'm not sure. I do rather like, "Flavelle" and, "Mountain Gem". I'll probably have to let the ideas ferment a while before I come to a decision.

I have recently seen a tomato that the name, "Speckled Garfech" would be perfect for. It came out of someone else's breeding program, so I won't share a photo. Imagine a yellow/orange striped tomato covered in green spots.
Two photos combined. The top half is a photo of a large yellow ceramic bowl filled with small cherry tomatoes. The cherry tomatoes area a mix of white and pale orange with a pink blush on one end. The bottom half is a photo of a closeup of a single larger tomato that is white with pale dark stripes. There are smaller red tomatoes and other items in the background.
From 2017.

I've got a couple more tomato lines that I'd like to stabilize (photographs at right). The upper photo shows a mix of small, very sweet cherries in pale-yellow/white with a pink blush on the bottom end of some. I'll be growing seeds from the ones with the blush. I expect the same phenotype will turn up next year, but I'm also sure there are lots of recessive alleles still hiding in them (for larger fruit, other tastes, and not having the blush).

The lower photo is of a larger, meaty white with pale stripes. This one is a bit further along already thanks to some lucky genetics, even though this phenotype only appeared in the last year. The fruit color, size, and shape are all due to recessive alleles, so those traits should already be stable. The stripes, flavor, and plant growth details probably won't be stable yet. I'll be growing several seeds from this fruit this year to find out.


References: