diff --git a/Changes b/Changes
index 9f2fdf9..f963676 100644
--- a/Changes
+++ b/Changes
@@ -1,3 +1,8 @@
+0.7.0 2025-06-03
+        - conllu2korapxml:
+            - Add automatic offset computation from 'text' comments
+              (implemented with AI assistance)
+
 0.6.3 2024-06-04
         - Trim filenames to fix double space after filename metadata
         - Set permissions of zip contents to 666
diff --git a/script/conllu2korapxml b/script/conllu2korapxml
index c8e4ec0..a68aced 100755
--- a/script/conllu2korapxml
+++ b/script/conllu2korapxml
@@ -14,7 +14,7 @@
 my %opts;
 my %processedFilenames;
 
-our $VERSION = '0.6.3';
+our $VERSION = '0.7.0';
 our $VERSION_MSG = "\nconllu2korapxml - v$VERSION\n";
 
 use constant {
@@ -59,6 +59,12 @@
 my @spansTo;
 my $current;
 my ($unknown, $known) = (0, 0);
+my $sentence_text = '';
+my $doc_offset = 0;
+my $text_pos = 0;
+my $compute_offsets_mode = 0;
+# Buffer dependency arcs until all token offsets in a sentence are known
+my @dep_buffer;
 
 my ($write_morpho, $write_syntax, $base) = (1, 0, 0);
 my $filename;
@@ -95,6 +101,9 @@
         }
         $processedFilenames{$filename}=1;
         $i=0;
+        $doc_offset = 0;
+        $sentence_text = '';
+        $compute_offsets_mode = 0;
       } elsif(/^#\s*foundry\s*[:=]\s*(.*)/) {
         if(!$foundry_name) {
           $dependency_foundry_name = $foundry_name = $1;
@@ -134,8 +143,43 @@
         @spansFrom = split(/\s+/, $1);
       }  elsif (/^(?:#|0\.4)\s+(?:end_offsets|to)\s+[:=]\s*(.*)/) {
         @spansTo = split(/\s+/, $1);
+      }  elsif (/^#\s*text\s*=\s*(.*)/) {
+        # Store sentence text for automatic offset computation
+        $sentence_text = decode('UTF-8', $1);
       }
     } elsif ( !/^\s*$/ ) {
+      # Pre-split columns before offset computation needs the raw form
+      my @raw_cols = split('\t');
+      chomp $raw_cols[$#raw_cols] if @raw_cols;
+
+      # Enter offset computation mode when no explicit offsets given
+      if (!$compute_offsets_mode && scalar @spansFrom == 0
+          && $sentence_text && $docid) {
+        $compute_offsets_mode = 1;
+        @spansFrom = ();
+        @spansTo = ();
+        # Sentence-level span covers the full sentence text
+        $spansFrom[0] = $doc_offset;
+        $spansTo[0] = $doc_offset + length($sentence_text);
+        $text_pos = 0;
+      }
+
+      # Locate each token form in sentence text via index()
+      if ($compute_offsets_mode && @raw_cols >= 2 && $raw_cols[0] =~ /^\d+$/) {
+        my $raw_form = decode('UTF-8', $raw_cols[1]);
+        my $t_num = $raw_cols[0];
+        # Find token starting from current position (handles SpaceAfter)
+        my $pos_in_text = index($sentence_text, $raw_form, $text_pos);
+        if ($pos_in_text >= 0) {
+          $spansFrom[$t_num] = $doc_offset + $pos_in_text;
+          $spansTo[$t_num] = $spansFrom[$t_num] + length($raw_form);
+          # Advance past this token for the next search
+          $text_pos = $pos_in_text + length($raw_form);
+        } else {
+          $log->warn("WARNING: Token form not found in sentence text in $conllu_file line $.");
+        }
+      }
+
       if ( !$docid || scalar @spansTo == 0 || scalar @spansFrom == 0 ) {
         if ( !$docid ) {
           $log->warn("WARNING: Invalid input in $conllu_file: text_id (e.g. '# text_id = GOE_AGA.00000') missing in line $. when writing to $outh");
@@ -155,8 +199,7 @@
         $s =~ s/</&lt;/g;
         $s =~ s/>/&gt;/g;
         $s;
-      } split('\t');
-      chomp  $parsed[9];
+      } @raw_cols;
       if (@parsed != 10) {
         $log->warn("WARNING: skipping strange parser output line in $docid");
         $i++;
@@ -169,14 +212,8 @@
       }
       if($parsed[6] =~ /\d+/ && $parsed[7] !~ /_/) {
         $write_syntax=1;
-        my $from=$spansFrom[$parsed[6]];
-        my $to=$spansTo[$parsed[6]];
-          $parse .= qq@<span id="s${s}_n$t" from="$spansFrom[$t]" to="$spansTo[$t]">
-<rel label="$parsed[7]">
-<span from="$from" to="$to"/>
-</rel>
-</span>
-@;
+        # Buffer dep arcs; flushed at end of sentence when offsets are ready
+        push @dep_buffer, [$s, $t, $parsed[6], $parsed[7]];
       }
       my $pos = $parsed[4];
       my $upos = $parsed[3];
@@ -208,6 +245,17 @@
   </span>
 );
         $i++;
+    } else {
+      # Empty line = end of sentence
+      flush_dep_buffer();
+      if ($compute_offsets_mode) {
+        # Advance doc offset past sentence + 1-char space separator
+        $doc_offset += length($sentence_text) + 1;
+        @spansFrom = ();
+        @spansTo = ();
+        $compute_offsets_mode = 0;
+      }
+      $sentence_text = '';
     }
   }
   $current .= "\n";
@@ -230,7 +278,22 @@
   }
 }
 
+# Write buffered dependency arcs now that all token offsets are resolved
+sub flush_dep_buffer {
+  foreach my $dep (@dep_buffer) {
+    my ($ds, $dt, $dhead, $dlabel) = @$dep;
+    $parse .= qq@<span id="s${ds}_n$dt" from="$spansFrom[$dt]" to="$spansTo[$dt]">
+<rel label="$dlabel">
+<span from="$spansFrom[$dhead]" to="$spansTo[$dhead]"/>
+</rel>
+</span>
+@;
+  }
+  @dep_buffer = ();
+}
+
 sub closeDoc {
+  flush_dep_buffer();
   if ($write_morpho && $morpho_file) {
     newZipStream($morpho_file);
     $zip->print($morpho, qq( </spanList>\n</layer>\n));
diff --git a/t/ud_conllu.t b/t/ud_conllu.t
new file mode 100644
index 0000000..bec7f29
--- /dev/null
+++ b/t/ud_conllu.t
@@ -0,0 +1,370 @@
+use strict;
+use warnings;
+use Test::More;
+use Test::Script;
+use Test::TempDir::Tiny;
+
+my $UNZIP = `sh -c 'command -v unzip'`;
+chomp $UNZIP;
+
+if ($UNZIP eq '') {
+    plan skip_all => 'No unzip executable found in PATH.';
+}
+
+my $offset_data = <<'CONLLU';
+# filename = TEST/TEST/TEST_OFF_001/base/tokens.xml
+# text_id = TEST_TEST.TEST_OFF_001
+# text = Geras rytas.
+1	Geras	geras	ADJ	bdv.	Case=Nom	2	amod	_	_
+2	rytas	rytas	NOUN	dkt.	Case=Nom	0	root	_	_
+3	.	.	PUNCT	skyr.	_	2	punct	_	SpaceAfter=No
+
+# text = Kaip sekasi?
+1	Kaip	kaip	ADV	prv.	_	2	advmod	_	_
+2	sekasi	sektis	VERB	vksm.	_	0	root	_	SpaceAfter=No
+3	?	?	PUNCT	skyr.	_	2	punct	_	_
+
+CONLLU
+
+my $test_tempdir = tempdir();
+my $offset_file = "$test_tempdir/test_offset.conllu";
+{
+    open(my $ofh, '>:encoding(UTF-8)', $offset_file)
+        or die "Cannot write test file: $!";
+    print $ofh $offset_data;
+    close($ofh);
+}
+
+my $zipcontent_off = '';
+script_runs(
+    [ 'script/conllu2korapxml', '-f', 'ud', $offset_file ],
+    { stdout => \$zipcontent_off },
+    "conllu2korapxml computes offsets from # text"
+);
+
+my $zipfile_off = "$test_tempdir/test_offset.zip";
+if ($zipcontent_off) {
+    open(my $zfh, '>:raw', $zipfile_off) or die "Cannot write zip: $!";
+    print $zfh $zipcontent_off;
+    close($zfh);
+
+    my $morpho_xml = `$UNZIP -p $zipfile_off 'TEST/TEST/TEST_OFF_001/ud/morpho.xml' 2>/dev/null`;
+
+    # Sentence 1: "Geras rytas." (12 chars, starts at 0)
+    #   Geras: 0-5, rytas: 6-11, .: 11-12
+    like($morpho_xml,
+        qr/id="s1_n1" from="0" to="5"/,
+        "Computed offset: token 'Geras' at 0..5");
+    like($morpho_xml,
+        qr/id="s1_n2" from="6" to="11"/,
+        "Computed offset: token 'rytas' at 6..11");
+    like($morpho_xml,
+        qr/id="s1_n3" from="11" to="12"/,
+        "Computed offset: token '.' at 11..12 (SpaceAfter=No)");
+
+    # Sentence 2: "Kaip sekasi?" (12 chars, starts at 13 = 12 + 1 space)
+    #   Kaip: 13-17, sekasi: 18-24, ?: 24-25
+    like($morpho_xml,
+        qr/id="s2_n1" from="13" to="17"/,
+        "Computed offset: token 'Kaip' at 13..17 (sentence 2)");
+    like($morpho_xml,
+        qr/id="s2_n2" from="18" to="24"/,
+        "Computed offset: token 'sekasi' at 18..24");
+    like($morpho_xml,
+        qr/id="s2_n3" from="24" to="25"/,
+        "Computed offset: token '?' at 24..25 (SpaceAfter=No)");
+
+    # Verify dependency XML has correct offsets (head references)
+    my $dep_xml = `$UNZIP -p $zipfile_off 'TEST/TEST/TEST_OFF_001/ud/dependency.xml' 2>/dev/null`;
+
+    # Token 1 "Geras" (0..5) -> head token 2 "rytas" (6..11)
+    like($dep_xml,
+        qr/id="s1_n1" from="0" to="5".*?<span from="6" to="11"/s,
+        "Dependency: 'Geras' head points to 'rytas' offsets");
+
+    # Token 2 "rytas" (6..11) -> head 0 = sentence span (0..12)
+    like($dep_xml,
+        qr/id="s1_n2" from="6" to="11".*?<span from="0" to="12"/s,
+        "Dependency: 'rytas' head points to sentence span");
+}
+else {
+    fail("Computed offset: token 'Geras' at 0..5");
+    fail("Computed offset: token 'rytas' at 6..11");
+    fail("Computed offset: token '.' at 11..12 (SpaceAfter=No)");
+    fail("Computed offset: token 'Kaip' at 13..17 (sentence 2)");
+    fail("Computed offset: token 'sekasi' at 18..24");
+    fail("Computed offset: token '?' at 24..25 (SpaceAfter=No)");
+    fail("Dependency: 'Geras' head points to 'rytas' offsets");
+    fail("Dependency: 'rytas' head points to sentence span");
+}
+
+# No SpaceAfter at all - every token has normal space separation
+my $no_spaceafter_data = <<'CONLLU';
+# filename = TEST/TEST/TEST_NSA_001/base/tokens.xml
+# text_id = TEST_TEST.TEST_NSA_001
+# text = One two three
+1	One	one	NUM	num.	_	0	root	_	_
+2	two	two	NUM	num.	_	1	flat	_	_
+3	three	three	NUM	num.	_	1	flat	_	_
+
+CONLLU
+
+my $nsa_file = "$test_tempdir/test_no_spaceafter.conllu";
+{
+    open(my $nfh, '>:encoding(UTF-8)', $nsa_file)
+        or die "Cannot write test file: $!";
+    print $nfh $no_spaceafter_data;
+    close($nfh);
+}
+
+my $zipcontent_nsa = '';
+script_runs(
+    [ 'script/conllu2korapxml', '-f', 'ud', $nsa_file ],
+    { stdout => \$zipcontent_nsa },
+    "conllu2korapxml handles tokens with no SpaceAfter"
+);
+
+my $zipfile_nsa = "$test_tempdir/test_no_spaceafter.zip";
+if ($zipcontent_nsa) {
+    open(my $zfh, '>:raw', $zipfile_nsa) or die "Cannot write zip: $!";
+    print $zfh $zipcontent_nsa;
+    close($zfh);
+
+    my $morpho_nsa = `$UNZIP -p $zipfile_nsa 'TEST/TEST/TEST_NSA_001/ud/morpho.xml' 2>/dev/null`;
+
+    # "One two three" = 13 chars
+    #   One: 0-3, two: 4-7, three: 8-13
+    like($morpho_nsa,
+        qr/id="s1_n1" from="0" to="3"/,
+        "No SpaceAfter: token 'One' at 0..3");
+    like($morpho_nsa,
+        qr/id="s1_n2" from="4" to="7"/,
+        "No SpaceAfter: token 'two' at 4..7");
+    like($morpho_nsa,
+        qr/id="s1_n3" from="8" to="13"/,
+        "No SpaceAfter: token 'three' at 8..13");
+}
+else {
+    fail("No SpaceAfter: token 'One' at 0..3");
+    fail("No SpaceAfter: token 'two' at 4..7");
+    fail("No SpaceAfter: token 'three' at 8..13");
+}
+
+# Adjacent tokens - no space between consecutive tokens (SpaceAfter=No)
+my $adjacent_data = <<'CONLLU';
+# filename = TEST/TEST/TEST_ADJ_001/base/tokens.xml
+# text_id = TEST_TEST.TEST_ADJ_001
+# text = foo(bar)baz end
+1	foo	foo	NOUN	n.	_	0	root	_	SpaceAfter=No
+2	(	(	PUNCT	skyr.	_	3	punct	_	SpaceAfter=No
+3	bar	bar	NOUN	n.	_	1	appos	_	SpaceAfter=No
+4	)	)	PUNCT	skyr.	_	3	punct	_	SpaceAfter=No
+5	baz	baz	NOUN	n.	_	1	conj	_	_
+6	end	end	NOUN	n.	_	1	conj	_	_
+
+CONLLU
+
+my $adj_file = "$test_tempdir/test_adjacent.conllu";
+{
+    open(my $afh, '>:encoding(UTF-8)', $adj_file)
+        or die "Cannot write test file: $!";
+    print $afh $adjacent_data;
+    close($afh);
+}
+
+my $zipcontent_adj = '';
+script_runs(
+    [ 'script/conllu2korapxml', '-f', 'ud', $adj_file ],
+    { stdout => \$zipcontent_adj },
+    "conllu2korapxml handles adjacent tokens without spaces"
+);
+
+my $zipfile_adj = "$test_tempdir/test_adjacent.zip";
+if ($zipcontent_adj) {
+    open(my $zfh, '>:raw', $zipfile_adj) or die "Cannot write zip: $!";
+    print $zfh $zipcontent_adj;
+    close($zfh);
+
+    my $morpho_adj = `$UNZIP -p $zipfile_adj 'TEST/TEST/TEST_ADJ_001/ud/morpho.xml' 2>/dev/null`;
+
+    # "foo(bar)baz end" = 15 chars
+    #   foo: 0-3, (: 3-4, bar: 4-7, ): 7-8, baz: 8-11, end: 12-15
+    like($morpho_adj,
+        qr/id="s1_n1" from="0" to="3"/,
+        "Adjacent: token 'foo' at 0..3");
+    like($morpho_adj,
+        qr/id="s1_n2" from="3" to="4"/,
+        "Adjacent: token '(' at 3..4 (no space before)");
+    like($morpho_adj,
+        qr/id="s1_n3" from="4" to="7"/,
+        "Adjacent: token 'bar' at 4..7 (no space before)");
+    like($morpho_adj,
+        qr/id="s1_n4" from="7" to="8"/,
+        "Adjacent: token ')' at 7..8 (no space before)");
+    like($morpho_adj,
+        qr/id="s1_n5" from="8" to="11"/,
+        "Adjacent: token 'baz' at 8..11 (no space before)");
+    like($morpho_adj,
+        qr/id="s1_n6" from="12" to="15"/,
+        "Adjacent: token 'end' at 12..15 (space before)");
+}
+else {
+    fail("Adjacent: token 'foo' at 0..3");
+    fail("Adjacent: token '(' at 3..4 (no space before)");
+    fail("Adjacent: token 'bar' at 4..7 (no space before)");
+    fail("Adjacent: token ')' at 7..8 (no space before)");
+    fail("Adjacent: token 'baz' at 8..11 (no space before)");
+    fail("Adjacent: token 'end' at 12..15 (space before)");
+}
+
+# Repeated token form - same word appears multiple times in sentence
+my $repeat_data = <<'CONLLU';
+# filename = TEST/TEST/TEST_REP_001/base/tokens.xml
+# text_id = TEST_TEST.TEST_REP_001
+# text = the cat and the dog
+1	the	the	DET	det.	_	2	det	_	_
+2	cat	cat	NOUN	n.	_	0	root	_	_
+3	and	and	CCONJ	cc.	_	5	cc	_	_
+4	the	the	DET	det.	_	5	det	_	_
+5	dog	dog	NOUN	n.	_	2	conj	_	_
+
+CONLLU
+
+my $rep_file = "$test_tempdir/test_repeat.conllu";
+{
+    open(my $rfh, '>:encoding(UTF-8)', $rep_file)
+        or die "Cannot write test file: $!";
+    print $rfh $repeat_data;
+    close($rfh);
+}
+
+my $zipcontent_rep = '';
+script_runs(
+    [ 'script/conllu2korapxml', '-f', 'ud', $rep_file ],
+    { stdout => \$zipcontent_rep },
+    "conllu2korapxml handles repeated token forms"
+);
+
+my $zipfile_rep = "$test_tempdir/test_repeat.zip";
+if ($zipcontent_rep) {
+    open(my $zfh, '>:raw', $zipfile_rep) or die "Cannot write zip: $!";
+    print $zfh $zipcontent_rep;
+    close($zfh);
+
+    my $morpho_rep = `$UNZIP -p $zipfile_rep 'TEST/TEST/TEST_REP_001/ud/morpho.xml' 2>/dev/null`;
+
+    # "the cat and the dog" = 19 chars
+    #   the: 0-3, cat: 4-7, and: 8-11, the: 12-15, dog: 16-19
+    like($morpho_rep,
+        qr/id="s1_n1" from="0" to="3"/,
+        "Repeated: first 'the' at 0..3");
+    like($morpho_rep,
+        qr/id="s1_n2" from="4" to="7"/,
+        "Repeated: 'cat' at 4..7");
+    like($morpho_rep,
+        qr/id="s1_n4" from="12" to="15"/,
+        "Repeated: second 'the' at 12..15 (not matching first)");
+    like($morpho_rep,
+        qr/id="s1_n5" from="16" to="19"/,
+        "Repeated: 'dog' at 16..19");
+}
+else {
+    fail("Repeated: first 'the' at 0..3");
+    fail("Repeated: 'cat' at 4..7");
+    fail("Repeated: second 'the' at 12..15 (not matching first)");
+    fail("Repeated: 'dog' at 16..19");
+}
+
+# Single-token sentence - minimal sentence with only one word
+my $single_data = <<'CONLLU';
+# filename = TEST/TEST/TEST_SNG_001/base/tokens.xml
+# text_id = TEST_TEST.TEST_SNG_001
+# text = Hello
+1	Hello	hello	INTJ	intj.	_	0	root	_	_
+
+# text = World
+1	World	world	NOUN	n.	_	0	root	_	_
+
+CONLLU
+
+my $sng_file = "$test_tempdir/test_single.conllu";
+{
+    open(my $sfh, '>:encoding(UTF-8)', $sng_file)
+        or die "Cannot write test file: $!";
+    print $sfh $single_data;
+    close($sfh);
+}
+
+my $zipcontent_sng = '';
+script_runs(
+    [ 'script/conllu2korapxml', '-f', 'ud', $sng_file ],
+    { stdout => \$zipcontent_sng },
+    "conllu2korapxml handles single-token sentences"
+);
+
+my $zipfile_sng = "$test_tempdir/test_single.zip";
+if ($zipcontent_sng) {
+    open(my $zfh, '>:raw', $zipfile_sng) or die "Cannot write zip: $!";
+    print $zfh $zipcontent_sng;
+    close($zfh);
+
+    my $morpho_sng = `$UNZIP -p $zipfile_sng 'TEST/TEST/TEST_SNG_001/ud/morpho.xml' 2>/dev/null`;
+
+    # Sentence 1: "Hello" (5 chars, starts at 0)
+    # Sentence 2: "World" (5 chars, starts at 6 = 5 + 1 space)
+    like($morpho_sng,
+        qr/id="s1_n1" from="0" to="5"/,
+        "Single-token: 'Hello' at 0..5");
+    like($morpho_sng,
+        qr/id="s2_n1" from="6" to="11"/,
+        "Single-token: 'World' at 6..11 (after space separator)");
+}
+else {
+    fail("Single-token: 'Hello' at 0..5");
+    fail("Single-token: 'World' at 6..11 (after space separator)");
+}
+
+# Explicit offsets take precedence over # text when both are present
+my $explicit_wins_data = <<'CONLLU';
+# filename = TEST/TEST/TEST_EXP_001/base/tokens.xml
+# text_id = TEST_TEST.TEST_EXP_001
+# text = Geras rytas.
+# start_offsets = 0 100 200 300
+# end_offsets = 999 199 299 399
+1	Geras	geras	ADJ	bdv.	Case=Nom	2	amod	_	_
+2	rytas	rytas	NOUN	dkt.	Case=Nom	0	root	_	_
+3	.	.	PUNCT	skyr.	_	2	punct	_	_
+
+CONLLU
+
+my $exp_file = "$test_tempdir/test_explicit.conllu";
+{
+    open(my $efh, '>:encoding(UTF-8)', $exp_file)
+        or die "Cannot write test file: $!";
+    print $efh $explicit_wins_data;
+    close($efh);
+}
+
+my $zipcontent_exp = '';
+script_runs(
+    [ 'script/conllu2korapxml', '-f', 'ud', $exp_file ],
+    { stdout => \$zipcontent_exp },
+    "conllu2korapxml uses explicit offsets when both # text and offsets present"
+);
+
+my $zipfile_exp = "$test_tempdir/test_explicit.zip";
+if ($zipcontent_exp) {
+    open(my $zfh, '>:raw', $zipfile_exp) or die "Cannot write zip: $!";
+    print $zfh $zipcontent_exp;
+    close($zfh);
+
+    my $morpho_exp = `$UNZIP -p $zipfile_exp 'TEST/TEST/TEST_EXP_001/ud/morpho.xml' 2>/dev/null`;
+    like($morpho_exp,
+        qr/id="s1_n1" from="100" to="199"/,
+        "Explicit offsets win: token uses from=100 (not computed 0)");
+}
+else {
+    fail("Explicit offsets win: token uses from=100 (not computed 0)");
+}
+
+done_testing;