c d John T(pres,[3,pl])
vunerg think T(pres,[3,pl]) ce
d he T(pres,[3,pl]) vbe smart
(frontier)
John -s think that he -s be smart
(after morpheme realization)
John think -s that he be -s smart
(after affix-hop)
John think -s that he be -s smart
(after morpheme realization, stage 2)
John thinks that he is smart
(spellout)
See full derivation for the example here.
Note: strikethrough, e.g. selfhedJohn, is used to indicate unpronounced copies.
Example 2
There are likely to be several prizes awarded
(expletive passive with leftwards TH/EX) (Chomsky 2001)
Stream: [prizes, several, award, prt, 'v~', there, 'Tinf',
likely, v_be, 'T', c] derives:
c there T(pres,[3,pl]) vbe likely to v~
several prizes prt award
(frontier)
there -re be likely to be several prizes -en award
(after morpheme realization)
there be -re likely to be several prizes award -en
(after affix-hop)
there be -re likely to be several prizes award -en
(after morpheme realization, stage 2)
there are likely to be several prizes awarded
(spellout)
See full derivation for the example here.
Note: strikethrough, e.g. severalprizes,
is used to indicate unpronounced copies.
Core architecture highlights:
A small set of primitive operations may apply to
a current syntactic object (SO) and a stream of heads. If no
matching operation is available, the derivation crashes.
Structure is recursively built by Merge in two possible
configurations: (1) current SO + head of stream = external
merge. (2) current SO + sub-SO = internal merge.
Heads may probe (within the current SO) for matching features on
a goal at the time of external merge. A list of available goals are
maintained on a stack. (There is no explicit search of the
current SO.)
A probe may agree with multiple goals, e.g. phi-complete T
may value nom Case for prt and a sentential object.
In turn, a goal may value uPhi on multiple probes, eg. prt and
T.
A probe may simultanously probe for multiple uFeatures,
e.g. cQ has both uWh and uT.
Multiple uFeatures may be simultanously valued by corresponding
iFeatures on one goal (economy).
Alternatively, multiple uFeatures may be valued by iFeatures on
different goals.
A spell-out component is included for English. Supports
verbal inflection, do-support, and
that-insertion.
Merge takes place repeatedly to derive a single syntactic object
(SO) in a Workspace (WS).
Generally, there may be several WSs involved in building a
parse. For example, the External Argument (EA) is formed in its own
sub-WS, then Merged with the main WS containing the Internal
Argument (IA) and verbal structure (R, v*, T and C).
FormSet may be used construct a set containing a finite number
of matching syntactic objects formed independently in sub-WSs.
ef(theta/wh): edge position is restricted to
theta-role bearing or wh-bearing syntactic objects.
Local Extent (LE) head. (Similar idea to a phase head.)
When fully
projected, the phrase headed by the LE head defines a LE
boundary.
The current stack position will be marked (with a b = boundary) and
further stack operations will not be able to search beyond the mark.
Derivations
For full step-by-step derivations, visit the following links.
(These webpages are automatically generated by my system.)
(Also, please visit Jason Ginsburg's page
for Jason's derivations.)
On the computational modeling of English relative
clauses. Fong, S. and J. Ginsburg.
Appendix: containing a list of
Minimalist Machine operations and full derivations for all examples in the paper.
Combining linguistic theories in a Minimalist
Machine. Ginsburg, J. and S. Fong. In
Minimalist Parsing Stabler, E.J. and R.C. Berwick
(eds.). 39-68. Oxford University Press.
A New Approach to Tough-Constructions. Fong, S. and J. Ginsburg. In
Proceedings of the 31st West Coast Conference on Formal
Linguistics (WCCFL 31) Santana-LaBarge, R.E.
(ed.). 180-188.Cascadilla Proceedings Project.
The Minimalist Machine runs on macOS, Linux and
Windows. Pre-compiled binaries are available.
It has a Graphical User Interface (GUI) built in Javascript/HTML
that runs in any modern browser.
The theory is implemented in Prolog. We rely on the freely
available SWI-Prolog, currently version 9, which you should also
install. This code runs in a Terminal window.
The browser and the Prolog code communicate via a Websocket (WS)
interface. Nothing extra needs to be installed for this.
Interested?
Download Machine.zip (3,178,802 bytes, uploaded
8/7/2023, the .zip file is
a compressed directory).
Unzip this into a directory called Machine.
The Instructions tab has detailed step-by-step instructions on
how to run the code.
Download the Software first. Then go here for easy, step-by-step
instructions.
