Skip to content


Repository files navigation

Global Names Parser: GNparser written in Go


Try GNparser online.

Try GNparser with OpenRefine

GNparser splits scientific names into their semantic elements with an associated meta information. Parsing is indispensable for matching names from different data sources, because it can normalize different lexical variants of names to the same canonical form.

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. Scala project is in an archived state, biodiversity now uses Go code for parsing. 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. On a Mac you might also need to go to System Preferences and security panel select Allow from other developers. Then, after running gnparser, click 'Yes' in a dialog box allowing to run a program from an "unregistered developer".

tar xvf gnparser-v1.0.0-linux.tar.gz
sudo cp gnparser /usr/local/bin
# for CSV output
gnparser "Homo sapiens Linnaeus"
# for TSV output
gnparser -f tsv "Homo sapiens Linnaeus"
# for JSON output
gnparser -f compact "Homo sapiens Linnaeus"
gnparser -f compact "Homo sapiens Linnaeus" | jq
# or
gnparser -f pretty "Homo sapiens Linnaeus"
gnparser -h


Zenodo DOI can be used to cite GNparser


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, TSV 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.

GNparser reached a stable v1. Differences between v1 and v0


Number of names parsed per second on an AMD Ryzen 7 5800H CPU (8 cores, 16 threads), GNparser v1.3.0:

gnparser 1_000_000_names.txt -j 200 > /dev/null
Threads names/sec
1 9,000
2 19,000
4 35,000
8 56,000
16 82,000
100 107,000
200 111,000

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


  • 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 RESTful service and interactive web interface.
  • Can run as a command line application.
  • Can be used as a library in Go projects.
  • Can be scaled to many CPUs and computers (if 250 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 to 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 lexical variants of names. Three versions of canonical forms are included:

Canonical Example Use
- Spiraea alba var. alba Du Roi Best for disambiguation, but has many lexical variants
Full Spiraea alba var. alba Presentation, infraspecies disambiguation
Simple Spiraea alba alba Name matching, presentation
Stem Spiraea alb alb Best for matching fem./masc. inconsistencies

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. It allows to match names with inconsistent gender suffixes in specific epithets (for example alba vs. albus). 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 mostly about canonical form of a name you can use default --format csv flag with command line tool.

CSV/TSV 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.
CanonicalStem Simplest canonical form with removed suffixes
CanonicalSimple Simplest canonical form
CanonicalFull Canonical form with hybrid sign and ranks
Authors Authorship 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 according to the 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 authorship from the middle of the name

Often data administrators spit name-strings 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 the default --format csv flag for gnparser 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 are some significant problems with parsing.
  • "quality": 4 - 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 failed.

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 authorship. The words 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 words contains 3 parts:

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

The words section belongs to additional details. To use it enable --details flag for the command line application.

gnparser -d "Pardosa moesta Banks, 1892"



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.

Install with Homebrew (Mac OS X, Linux)

Homebrew is a packaging system originally made for Mac OS X. You can use it now for Mac, Linux, or Windows X WSL (Windows susbsystem for Linux).

  1. Install Homebrew according to their instructions.

  2. Install gnparser with:

    brew tap gnames/gn
    brew install gnparser

Linux or Mac OS X

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

sudo mv path_to/gnparser /usr/local/bin


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 user and/or system environment variables.

It is also possible to install Windows Subsystem for Linux on Windows (v10 or v11), and use gnparser as a Linux executable.

Install with Go

If you have Go installed on your computer use

go get -u

For development install gnu make and use the following:

git clone
cd gnparser
make tools
make install

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


Command Line

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

Relevant flags:

--help -h : help information about flags.

--batch_size -b : Sets a maximum number of names collected into a batch before processing. This flag is ignored if parsing mode is set to streaming with -s flag.

--cultivar -C : Adds support for botanical cultivars like Sarracenia flava 'Maxima' and graft-chimaeras like + Crataegomespilus

--capitalize -c : Capitalizes the first letter of name-strings.

--details -d : Return more details for a parsed name. This flag is ignored for CSV/TSV formatting.

--diaereses -D : Preserves diaereses within names, e.g. Leptochloöpsis virgata. The stemmed canonical name will be generated without diaereses.

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

CSV and TSV formats return a header row and the CSV/TSV-compatible parsed result.

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

--ignore_tags -i : 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.

--port -p : set a port to run web-interface and RESTful API.

--nsqd-tcp : requires --port. Allows to redirect web-service log output to NSQ messaging server's TCP-based endpoint. It is handy for aggregations of logs from GNparser web-services running inside of Docker containers or in Kubernetes pods.

--species-group-cut : Changes stemmed canonical for autonym or species group names (e.g. Aus bus bus). It cuts infraspecific epithet, leaving only genus and specific epithet. All other data stays the same. This feature might be useful to match names like Aus bus and Aus bus bus.

--stream -s : GNparser can be used from any language using pipe-in/pipe-out of the command line application. This approach requires sending 1 name at a time to GNparser instead of sending names in batches. Streaming allows to achieve that.

--unordered -u : does not restore the order of output according to the order of input.

--version -V : shows the version number of GNparser.

--web-logs : requires --port. Enables output of logs for web-services.

To parse one name:

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

# TSV output
gnparser -f tsv "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 botanical cultivar name
gnparser "Anthurium 'Ace of Spades'" --cultivar
gnparser "Phyllostachys vivax cv aureocaulis" -c

# to parse name that is all in low-case
gnparser "parus major" --capitalize
gnparser "parus major" -c

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.

# to parse with 200 parallel processes
gnparser -j 200 names.txt > names_parsed.csv

# to parse file with more detailed output
gnparser names.txt -d -f compact > names_parsed.txt

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

# to parse using `stream` method instead of `batch` method.
cat names.txt | gnparser -s > names_parsed.csv

# to not remove 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 -i "<i>Pomatomus</i>&nbsp;<i>saltator</i>"
gnparser -i "Pomatomus saltator"

If jobs number is set to more than 1, parsing uses several concurrent processes. This approach increases speed of parsing on multi-CPU computers. The results are returned in some random order, and reassembled into the order of input transparently for a user.

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.


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 -s --format #{format}")
    io[format.to_sym] = { stdin: stdin, stdout: stdout, stderr: stderr }

@marcobrt kindly provided an example in PHP.

Note that you have to use --stream -s flag for this approach to work.

R language package

For R language it is possible to use rgnparser package. It implements mentioned above pipes method. It does require gnparser app be installed.

Ruby Gem

Ruby developers can use GNparser functionality via biodiversity gem. It uses C-binding and does not require an installed gnparser app.


@tobymarsden created a wrapper for node.js. It uses C-binding and does not require an installed gnparser app.

Usage as a REST API Interface or Web-based User Graphical Interface

Web-based user interface and API are invoked by --port or -p flag. To start web server on

gnparser -p 9000

Opening a browser with this address will now show an interactive interface to parser. API calls would be accessible on

The api is and schema are described fully using OpenAPI specification.

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('')
http =, uri.port)
http.use_ssl = true
request =, 'Content-Type' => 'application/json',
                                   'accept' => 'json')
request.body = ['Solanum mariae Särkinen & S.Knapp',
                'Ahmadiago Vánky 2004'].to_json
response = http.request(request)

Enabling logs for GNparser's web-service

There are several ways to enable logging from a web service.

The following enables web-access logs to be printed to STDERR

gnparser -p 80 --web-logs

This next settings allows to send logs to a NSQ messaging service. This option allows aggregation logs from several instances of GNparser together. It is a great way for log aggregation and analysis if the instances run inside Docker containers or as Kubernetes Pods.

gnparser -p 80 --nsqd-tcp=

An important note: the address must point to the TCP service of nsqd.

To enable logs to be sent to STDERR and NSQ run

gnparser -p 80 --web-logs --nsqd-tcp=

Use as a Docker image

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

# run as a website and a RESTful service
docker run -p gnames/gognparser -p 8080 --web-logs

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

Use as a library in Go

import (


func Example() {
  names := []string{"Pardosa moesta Banks, 1892", "Bubo bubo"}
  cfg := gnparser.NewConfig()
  gnp := gnparser.New(cfg)
  res := gnp.ParseNames(names)
  // Output:
  // Banks 1892
  // Bubo bubo
  // Id,Verbatim,Cardinality,CanonicalStem,CanonicalSimple,CanonicalFull,Authorship,Year,Quality
  // e2fdf10b-6a36-5cc7-b6ca-be4d3b34b21f,"Pardosa moesta Banks, 1892",2,Pardosa moest,Pardosa moesta,Pardosa moesta,Banks 1892,1892,1

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.

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".



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


Mozzherin, D.Y., Myltsev, A.A. & Patterson, D.J. “gnparser”: a powerful parser for scientific names based on Parsing Expression Grammar. BMC Bioinformatics 18, 279 (2017).

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


Released under MIT license