Skip to content
Earley Parser in Python 3
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


Earley Parser in Python 3

Copyright (C) 2013 tobyp. Licensed under the GNU General Public License v3+. See the code and the LICENSE file.


Jay Earley's parsing algorithm can parse pretty much all context-free languages (though they might need to be processed in some way first -- see Advanced Grammars below). The algorithm is easy to understand and follow, simple enough to implement, and (in my opinion) simply beautiful. This parser is intended more as a proof-of-concept parser than a performance-optimized, practically feasible parser.

Implementation description

Everything needed for simple parsing is included in the file In this implementation, tokens are scanned greedily using regular expressions. Each terminal type can contain a function to produce a more useful/desirable form of value for continued parsing than a simple substring of the input, by transforming a re.MatchObject. (The special None token is used to swallow whitespace/comments/whatever you don't need.)

Each production rule is also provided with a function to transform its parsed parts into a more useful/processed form, to be passed up as arguments into rule-functions of other productions containing it in their right-hand side. This is especially useful for creating trees like ASTs ( contains a bit of code that might help for this sort of thing).


The file contains a simple calculator for mathematical expressions (read line-by-line from standard input), supporting basic arithmetic operators, a handful of functions, and a few constants.

Advanced Grammars

Grammars with epsilon rules (i.e. productions with no symbols on the right-hand side) can be used with the EpsilonGrammar class ( This works by converting the grammar into a form without epsilon-rules. CAUTION: grammars with an epsilon rule for the start nonterminal are not supported. Check for this trivial case manually before you parse.

Complex Grammars

To save some work with frequently used constructs like optional symbols, alternatives, or repetition of symbols, the ComplexGrammar class ( will process a grammar with special syntax for the right-hand sides of the production rules, and produce an EpsilonGrammar (that will in turn produce a normal Grammar) that works equivalently. The rule functions of this grammar are a bit more complicated. A small example is contained at the bottom of

Complex Grammar syntax

  • term2 (tightest binding):
    • (term)
      • description: group symbols.
      • passed as: list
      • example: A (LPAREN RPAREN) B parsing A LPAREN RPAREN B would pass of 'A', ['LPAREN', 'RPAREN'], 'B'
    • [term]
      • description: optional term.
      • passed as: None or list
      • example: A [LPAREN RPAREN] B parsing A LPAREN RPAREN B would pass 'A', ['LPAREN', 'RPAREN'], 'B'
      • example: A [LPAREN RPAREN] B parsing A B would pass 'A', None, 'B'
    • {term}
      • description: repeat term
      • passed as: list
      • example: A {B} C parsing A B B B C would pass 'A', ['B', 'B', 'B'], 'C'
    • {term:token}
      • description: list with seperator (note that the seperator can only be a token, not a term)
      • passed as: list (without seperators, those are dropped)
      • example: A {B:C} D parsing A B C B D would pass 'A', ['B', 'B'], 'D'
  • term1 (medium binding)
    • term1 | term2
      • description: alternative. Groups left to right.
      • passed as: however the alternative that was picked is passed
      • example: A | B parsing A would pass 'A'
      • example: A | (B | C) parsing B would pass ['B']
    • term2
      • description: anything that can be a term2 can also be a term1 - for obvious reasons.
  • term (loostest binding)
    • term term1
      • description: concatenation. Groups left to right.
      • passed as: seperate arguments.
      • example: A B C parsing A B C would simply pass 'A', 'B', 'C' (not 'A', ['B', 'C'] and not ['A', 'B', 'C']!)

A good way to check out how this works would be to wrap the lambda function creation in ComplexGrammar.simplify_term in a way that gives you useful information (stack dumps are rather unhelpful if there are a lot of lambdas involved), and then look at the the terms member of the ComplexGrammar object (which will have the rules as simplified by ComplexGrammar and EpsilonGrammar).

You can’t perform that action at this time.