Global Names Parser: gnparser written in Go

Try gnparser online.

gnparser splits scientific names into their semantic elements with an associated meta information. For example, "Homo sapiens Linnaeus" is parsed into:

Element	Meaning	Position
Homo	genus	(0,4)
sapiens	specificEpithet	(5,12)
Linnaeus	author	(13,21)

This parser, written in Go, is the 3rd iteration of the project. The first, biodiversity had been written in Ruby, the second, also gnparser, had been written in Scala. This project is now a substitution for the other two. It will be the only one that is maintained further. All three projects were developed as a part of Global Names Architecture Project.

To use gnparser as a command line tool under Windows, Mac or Linux, download the latest release, uncompress it, and copy gnparser binary somewhere in your PATH.

wget https://github.com/gnames/gnparser/uploads/55d247b8fbade60116c7e3b650dd978c/gnparser-v0.9.0-linux.tar.gz
tar xvf gnparser-v0.9.0-linux.tar.gz
sudo cp gnparser /usr/local/bin
# for CSV output
gnparser "Homo sapiens Linnaeus"
# for JSON output
gnparser -f compact "Homo sapiens Linnaeus"
# or
gnparser -f pretty "Homo sapiens Linnaeus"
gnparser -h

• Introduction
• Speed
• Features
• Use Cases
  • Getting the simplest possible canonical form
  • Quickly partition names by the type
  • Normalizing name-strings
  • Removing authorships from the middle of the name
  • Figuring out if names are well-formed
  • Creating stable GUIDs for name-strings
  • Assembling canonical forms etc. from original spelling
• Installation
  • Linux or OS X
  • Windows
  • Install with Go
• Usage
  • Command Line
  • Pipes
  • gRPC server
  • Usage as a REST API Interface
  • Use as a Docker image
  • Use as a library in Go
  • Use as a shared C library
• Parsing ambiguities
  • Names with filius (ICN code)
  • Names with subgenus (ICZN code) and genus author (ICN code)
• Authors
• Contributors
• References
• License

Introduction

Global Names Parser or gnparser is a program written in Go for breaking up scientific names into their elements. It uses peg -- a Parsing Expression Grammar (PEG) tool.

Many other parsing algorithms for scientific names use regular expressions. This approach works well for extracting canonical forms in simple cases. However, for complex scientific names and to parse scientific names into all semantic elements, regular expressions often fail, unable to overcome the recursive nature of data embedded in names. By contrast, gnparser is able to deal with the most complex scientific name-strings.

gnparser takes a name-string like Drosophila (Sophophora) melanogaster Meigen, 1830 and returns parsed components in CSV or JSON format. The parsing of scientific names might become surprisingly complex and the gnparser's test file is a good source of information about the parser's capabilities, its input and output.

Speed

Number of names parsed per hour on a i7-8750H CPU (6 cores, 12 threads, at 2.20 GHz), parser v0.5.1

Threads	names/hr
1	48,000,000
2	63,000,000
4	128,000,000
8	202,000,000
16	248,000,000
100	293,000,000

For simplest output Go gnparser is roughly 2 times faster than Scala gnparser and about 100 times faster than Ruby biodiversity parser. For JSON formats the parser is approximately 8 times faster than Scala one, due to more efficient JSON conversion.

Features

• Fastest parser ever.
• Very easy to install, just placing executable somewhere in the PATH is sufficient.
• Extracts all elements from a name, not only canonical forms.
• Works with very complex scientific names, including hybrid formulas.
• Includes gRPC server that can be used as if a native method call from C++,
• C#, Java, Python, Ruby, PHP, JavaScript, Objective C, Dart.
• Use as a native library from Go projects.
• Can run as a command line application.
• Can be scaled to many CPUs and computers (if 300 millions names an hour is not enough).
• Calculates a stable UUID version 5 ID from the content of a string.
• Provides C-binding to incorporate parser into other languages.

Use Cases

Getting the simplest possible canonical form

Canonical forms of a scientific name are the latinized components without annotations, authors or dates. They are great for matching names that differ in less stable parts. Use the canonicalName -> simple or canonicalName -> full fields from parsing results for this use case. Full version of canonical form includes infra-specific ranks and hybrid character for named hybrids.

The canonicalName -> full is good for presentation, as it keeps more details.

The canonicalName -> simple field is good for matching names from different sources, because sometimes dataset curators omit hybrid sign in named hybrids, or remove ranks for infraspecific epithets.

The canonicalName -> stem field normalizes simple canonical form even further. The normalization is done according to stemming rules for Latin language described in Schinke R et al (1996). For example letters j are converted to i, letters v are converted to u, and suffixes are removed from the specific and infraspecific epithets.

If you only care about canonical form of a name you can use --format csv flag with command line tool.

CSV output has the following fields:

Field	Meaning
Id	UUID v5 generated out of Verbatim
Verbatim	Input name-string without any changes
Cardinality	0 - N/A, 1 - Uninomial, 2 - Binomial etc.
CanonicalFull	Canonical form with hybrid sign and ranks
CanonicalSimple	Simplest canonical form
CanonicalStem	Simplest canonical form with removed suffixes
Authors	Author string of a name
Year	Year of the name (if given)
Quality	Parsing quality

Quickly partition names by the type

Usually scientific names can be broken into groups accoring by number of elements:

• Uninomial
• Binomial
• Trinomial
• Quadrinomial

The output of gnparser contains a Cardinality field that tells, when possible, how many elements are detected in the name.

Cardinality	Name Type
0	Undetermined
1	Uninomial
2	Binomial
3	Trinomial
4	Quadrinomial

For hybrid formulas, "approximate" names (with "sp.", "spp." etc.), unparsed names, as well as names from BOLD project cardinality is 0 (Undetermined)

Normalizing name-strings

There are many inconsistencies in how scientific names may be written. Use normalized field to bring them all to a common form (spelling, spacing, ranks).

Removing authorships from the middle of the name

Many data administrators store name-strings in two columns and split them into "name part" and "authorship part". This practice misses some information when dealing with names like "Prosthechea cochleata (L.) W.E.Higgins var. grandiflora (Mutel) Christenson". However, if this is the use case, a combination of canonicalName -> full with the authorship from the lowest taxon will do the job. You can also use --format csv flag for gnparse command line tool.

Figuring out if names are well-formed

If there are problems with parsing a name, parser generates qualityWarnings messages and lowers parsing quality of the name. Quality values mean the following:

• "quality": 1 - No problems were detected
• "quality": 2 - There were small problems, normalized result should still be good
• "quality": 3 - There were serious problems with the name, and the final result is rather doubtful
• "quality": 0 - A string could not be recognized as a scientific name and parsing fails

Creating stable GUIDs for name-strings

gnparser uses UUID version 5 to generate its id field. There is algorithmic 1:1 relationship between the name-string and the UUID. Moreover the same algorithm can be used in any popular language to generate the same UUID. Such IDs can be used to globally connect information about name-strings or information associated with name-strings.

More information about UUID version 5 can be found in the Global Names blog

Assembling canonical forms etc. from original spelling

gnparser tries to correct problems with spelling, but sometimes it is important to keep original spelling of the canonical forms or authorships. The positions field attaches semantic meaning to every word in the original name-string and allows users to create canonical forms or other combinations using the original verbatim spelling of the words. Each element in positions contains 3 parts:

1. semantic meaning of a word
2. start position of the word
3. end position of the word

For example ["specificEpithet", 6, 11] means that a specific epithet starts at 6th character and ends before 11th character of the string.

Installation

Compiled programs in Go are self-sufficient and small (gnparser is only a few megabytes). As a result the binary file of gnparser is all you need to make it work. You can install it by downloading the latest version of the binary for your operating system, and placing it in your PATH.

Linux or OS X

Move gnparser executable somewhere in your PATH (for example /usr/local/bin)

sudo mv path_to/gnparser /usr/local/bin

Windows

One possible way would be to create a default folder for executables and place gnparser there.

Use Windows+R keys combination and type "cmd". In the appeared terminal window type:

mkdir C:\bin
copy path_to\gnparser.exe C:\bin

Add C:\bin directory to your PATH environment variable.

Install with Go

If you have Go installed on your computer use

go get -u github.com/gnames/gnparser
cd $GOPATH/srs/github.com/gnames/gnparser
make install

You do need your PATH to include $HOME/go/bin

Usage

Command Line

gnparser -f pretty "Quadrella steyermarkii (Standl.) Iltis & Cornejo"

Relevant flags:

--help -h : help information about flags

--format -f : output format. Can be compact, pretty, csv, or debug. Default is csv.

CSV format returns a header row and the CSV-compatible parsed result.

--jobs -j : number of jobs running concurrently.

--nocleanup -n : keeps HTML entities and tags if they are present in a name-string. If your data is clean from HTML tags or entities, you can use this flag to increase performance.

To parse one name:

# CSV ouput (default)
gnparser "Parus major Linnaeus, 1788"
# or
gnparser -f csv "Parus major Linnaeus, 1788"

# JSON compact format
gnparser "Parus major Linnaeus, 1788" -f compact

# pretty format
gnparser -f pretty "Parus major Linnaeus, 1788"

# to parse a name from the standard input
echo "Parus major Linnaeus, 1788" | gnparser

To parse a file:

There is no flag for parsing a file. If parser finds the given file path on your computer, it will parse the content of the file, assuming that every line is a new scientific name. If the file path is not found, gnparser will try to parse the "path" as a scientific name.

Parsed results will stream to STDOUT, while progress of the parsing will be directed to STDERR.

gnparser -j 200 names.txt > names_parsed.txt

# to parse files using pipes
cat names.txt | gnparser -f csv -j 200 > names_parsed.txt

# to keep html tags and entities during parsing. You gain a bit of performance
# with this option if your data does not contain HTML tags or entities.
gnparser "<i>Pomatomus</i>&nbsp;<i>saltator</i>"
gnparser -n "<i>Pomatomus</i>&nbsp;<i>saltator</i>"
gnparser -n "Pomatomus saltator"

To parse a file returning results in the same order as they are given (slower):

gnparser -j 1 names.txt > names_parsed.txt

Potentially the input file might contain millions of names, therefore creating one properly formatted JSON output might be prohibitively expensive. Therefore the parser creates one JSON line per name (when compact format is used)

You can use up to 20 times more "threads" than the number of your CPU cores to reach maximum speed of parsing (--jobs 200 flag). It is practical because additional threads are very cheap in Go and they try to fill out every idle gap in the CPU usage.

Pipes

About any language has an ability to use pipes of the underlying operating system. From the inside of your program you can make the CLI executable gnparser to listen on a STDIN pipe and produce output into STDOUT pipe. Here is an example in Ruby:

def self.start_gnparser
  io = {}

  ['compact', 'csv'].each do |format|
    stdin, stdout, stderr = Open3.popen3("./gnparser -j 200 --format #{format}")
    io[format.to_sym] = { stdin: stdin, stdout: stdout, stderr: stderr }
  end
end

Such arrangement would give you a nearly native performance for large datasets.

gRPC server

Relevant flags:

--help -h : help information about flags

--grpc -g : sets a port to run gRPC server, and starts gnparser in gRPC mode.

--jobs -j : number or workers allocated per gRPC request. Default corresponds to the number of CPU threads. If you have a full control over gRPC server of gnparser, set this option to 100-300 jobs.

gnparser -g 8989 -j 200

For an example how to use gRPC server check gnparser Ruby gem as well as gRPC documentation.

It also helps to read gnparser.proto file to understand how to deal with inputs and outputs of gRPC server.

Usage as a REST API Interface

Use web-server REST API as a slower, but a more wide-spread alternative to gRPC server. Web-based user interface and API are invoked by --web-port or -w flag. To start web server on http://0.0.0.0:9000

gnparser -w 9000

Opening a browser with this address will now show an interactive interface to parser. API calls would be accessibe on http://0.0.0.0:9000/api.

Make sure to CGI-escape name-strings for GET requests. An '&' character needs to be converted to '%26'

• GET /api?q=Aus+bus|Aus+bus+D.+%26+M.,+1870
• POST /api with request body of JSON array of strings

require 'json'
require 'net/http'

uri = URI('https://parser.globalnames.org/api')
http = Net::HTTP.new(uri.host, uri.port)
http.use_ssl = true
request = Net::HTTP::Post.new(uri, 'Content-Type' => 'application/json',
                                   'accept' => 'json')
request.body = ['Solanum mariae Särkinen & S.Knapp',
                'Ahmadiago Vánky 2004'].to_json
response = http.request(request)

Use as a Docker image

You need to have docker runtime installed on your computer for these examples to work.

# run as a gRPC server on port 7777
docker run -p 0.0.0.0:7777:7777 gnames/gognparser -g 7777
# run grpc on 'default' 8778 port
docker run -p 0.0.0.0:8778:8778 gnames/gognparser
# to run as a daemon with 50 workers
docker run -d gnames/gognparser -g 7777 -j 50

# run as a website and a RESTful service
docker run -p 0.0.0.0:80:8080 gnames/gognparser -w 8080

# just parse something
docker run gnames/gognparser "Amaurorhinus bewichianus (Wollaston,1860) (s.str.)"

Use as a library in Go

package main

import (
  "fmt"

  "github.com/gnames/gnparser"
)

func main() {
  opts := []gnparser.Option{
    gnparser.Format("csv"),
    gnparser.WorkersNum(100),
  }
  gnp := gnparser.NewGNparser(opts...)
  res, err := gnp.ParseAndFormat("Bubo bubo")
  if err != nil {
    fmt.Println(err)
  }
  fmt.Println(res)
}

To avoid JSON format we provide gnp.ParseToObject function. Use gnparser.proto file as a reference of the available object fields.

gnp := NewGNparser()
o := gnp.ParseToObject("Homo sapiens")

fmt.Println(o.Canonical.Simple)
switch d := o.Details.(type) {
case *pb.Parsed_Species:
  fmt.Println(d.Species.Genus)
case *pb.Parsed_Uninomial:
  fmt.Println(d.Uninomial.Value)
...
}

Use as a shared C library

It is possible to bind gnparser functionality with languages that can use C Application Binary Interface. For example such languages include Python, Ruby, Rust, C, C++, Java (via JNI).

To compile gnparser shared library for your platform/operating system of choice you need GNU make and GNU gcc compiler installed:

make clib
cd binding
cp libgnparser* /path/to/some/project

As an example how to use the shared library check this StackOverflow question and biodiversity Ruby gem. You can find shared functions at their export file.

Parsing ambiguities

Some name-strings cannot be parsed unambiguously without some additional data.

Names with filius (ICN code)

For names like Aus bus Linn. f. cus the f. is ambiguous. It might mean that species were described by a son of (filius) Linn., or it might mean that cus is forma of bus. We provide a warning "Ambiguous f. (filius or forma)" for such cases.

Names with subgenus (ICZN code) and genus author (ICN code)

For names like Aus (Bus) L. or Aus (Bus) cus L. the (Bus) token would mean the name of subgenus for ICZN names, but for ICN names it would be an author of genus Aus. We created a list of ICN generic authors using data from IRMNG to distinguish such names from each other. For detected ICN names we provide a warning "Possible ICN author instead of subgenus".

Authors

• Dmitry Mozzherin

Contributors

• Geoff Ower
• Hernan Lucas Pereira

If you want to submit a bug or add a feature read CONTRIBUTING file.

References

Rees, T. (compiler) (2019). The Interim Register of Marine and Nonmarine Genera. Available from http://www.irmng.org at VLIZ. Accessed 2019-04-10

License

Released under MIT license